Flow Engineering, a promising startup specializing in AI-driven tools for hardware design, has successfully secured $50 million in a Series B funding round, reaching a valuation of $750 million, as announced on Wednesday.

The funding was led by notable investors, including Antonio Gracias from Valar Equity Partners, known for their stakes in Elon Musk's ventures like SpaceX, and Gavin Baker of Atreides Management, a hedge fund with investments in Musk's companies and AI chip manufacturer Cerebras. Sequoia Capital, which previously led Flow's Series A funding last October, also contributed, alongside former Sequoia partner Roelof Botha, who now serves on Flow’s board as an individual investor.

Founded three years ago in San Francisco, Flow is addressing the challenges associated with hardware design through its innovative AI agents, which streamline the process by automatically correlating CAD drawings with product specifications, simulation outcomes, and other testing results.

Among Flow’s client roster are prominent companies such as Anduril, Rivian, Joby Aviation, General Motors PPU (a collaboration between General Motors and TWG Motorsports), RV Tech (a joint venture between Rivian and Volkswagen), and Stoke Space, among others.

Valor, Atreides, and Sequoia support AI company Flow Engineering with a $750M valuation.

A bright, sunlit conference room with floor-to-ceiling windows overlooking a city skyline. A human founder and a sleek robotic arm shake hands across a polished glass table. On the table, a tablet prominently displays the Flow Engineering logo, and beside it sits a crystal plaque etched “$750M Valuation.” Behind them on the wall are three framed seals: Valor, Atreides, and Sequoia, symbolizing their backing. The scene feels optimistic and collaborative under natural light.

Valor, Atreides, and Sequoia back AI startup Flow Engineering at $750M valuation - TechCrunch

- Flow Engineering raised $50 million Series B, achieving a $750 million valuation.  
- Investment co-led by Valar Equity Partners and Atreides Management, backed by Sequoia Capital.  
- Flow's AI tools assist hardware design by aligning CAD drawings with product requirements.  

In Brief

Flow Engineering, a startup that offers AI tools for hardware design, has raised a $50 million Series B round at a $750 million valuation from some big-name investors, the company announced on Wednesday.

The round was co-led by Antonio Gracias of Valar Equity Partners (best known for its investments in Elon Musk’s companies, particularly SpaceX) and Gavin Baker of Atreides Management (a hedge fund that has also backed Musk’s companies and others like AI chipmaker Cerebras). Sequoia Capital, which led Flow’s Series A round last October, also participated, as did former Sequoia partner Roelof Botha, who invested as an individual investor. Botha has joined Flow’s board, too.

Three-year-old, San Francisco-based Flow is tackling the difficulties of hardware design by offering AI agents that automatically align CAD drawings with product requirements, simulation results, and other testing.

It names Anduril, Rivian, Joby Aviation, General Motors PPU (a joint venture between General Motors and TWG Motorsports), RV Tech (a Rivian and Volkswagen joint venture), Stoke Space, and others, as customers.

Valor, Atreides, and Sequoia back AI startup Flow Engineering at $750M valuation

Israeli startup Two Delta, specializing in artificial intelligence, has secured a significant seed funding round, amassing tens of millions of dollars. This funding round saw participation from notable investors, including Bill Ackman, the founder and CEO of Pershing Square, and Eric Schmidt, the former chairman of Google, as reported by Calcalist.

Established in 2025, Two Delta operates discreetly, sharing minimal information about its activities beyond academic papers available on its website. The company was co-founded by Dvir Zeltzer and Dr. Eli David, focusing on developing technologies that enhance the speed and efficiency of AI model performance while optimizing computing resource usage.

Dr. Eli David, an experienced AI researcher and entrepreneur, collaborates with Dvir Zeltzer, who brings a solid foundation in business management, including previous leadership at Google.

In October 2026, Two Delta shared findings from a partnership with Nvidia, showcasing its DCOCO technology, which accelerates inference processes for Nvidia’s Nemotron 3.5 Super VL model. This innovative approach aims to boost the number of tokens generated per verification step without sacrificing the quality of the output.

DCOCO enhances a methodology known as speculative decoding, where a quicker, smaller model generates a sequence of tokens for a larger model to validate in a single go. While this can expedite the generation process, it may lead to the rejection of some proposed tokens. The challenge lies in crafting drafts that the larger model is more likely to endorse.

To tackle this, Two Delta’s technique integrates candidate-token embeddings with contextual data derived from the hidden states of the drafting model, utilizing insights from the target model’s activations. This scoring system evaluates pairs of neighboring tokens concurrently, thereby selecting more coherent sequences prior to verification by the larger model.

In tests conducted across ten datasets encompassing mathematics, programming, and general tasks, Two Delta claimed that DCOCO achieved a throughput average of 2.03 times higher than standard autoregressive decoding when reasoning was disabled and greedy decoding applied. In a similar setting, it reported a throughput enhancement of 19.8% over the competing DFlash method. The testing, done on Nvidia’s DGX B200 hardware using the vLLM inference framework, yielded additional throughput improvements across various decoding configurations and up to 32 simultaneous requests.

These reported results stem from Two Delta's own studies rather than independent evaluations. The company asserts that DCOCO enhances token generation efficiency while maintaining the output quality of the target model.

Additionally, Two Delta introduced EvoWeb, a benchmark aimed at assessing the effectiveness of AI coding agents in developing and maintaining web applications amid users' continuous requests for new features and modifications. Unlike traditional benchmarks that focus on single-prompt performance, EvoWeb challenges the AI to manage an application through ten successive requests, beginning with an initial build followed by nine modifications.

During the evaluation of 32 web applications, no model was able to sustain complete functionality throughout the testing process. The highest-scoring models, Fable 5 and Kimi K3, achieved scores of 0.694 and 0.684, respectively, on a scale from zero to one.

The primary challenge identified was the integration of new features with the existing system functionality. Although occurrences of build failures were infrequent, the study noted an increase in runtime errors, which rose from 2% at the outset to 12% after a series of ten requests.

Bill Ackman and Eric Schmidt support the discreet Israeli AI company Two Delta.

In a sunlit, modern Tel Aviv high-rise boardroom at dawn, two sharp-suited figures (one wearing glasses, one with a trimmed haircut) stand side by side facing a large wall-mounted LED screen that boldly displays the Two Delta logo. Warm golden light floods through floor-to-ceiling windows, casting long shadows across the polished table. A single Israeli flag stands in the corner, lending a sense of national pride and quiet anticipation as the investors prepare to launch the secretive AI venture.

Bill Ackman and Eric Schmidt back secretive Israeli AI startup Two Delta - CTech

- Israeli startup Two Delta raised millions in seed funding from investors like Bill Ackman.  
- Co-founded by AI expert Eli David and Dvir Zeltzer, it focuses on AI model efficiency.  
- Two Delta's DCOCO technology achieved significant throughput improvements in AI token generation tests.

Israeli artificial intelligence startup Two Delta has raised tens of millions of dollars in a seed round backed by investors including Bill Ackman, founder and CEO of Pershing Square, and former Google chairman Eric Schmidt, Calcalist has learned.

Founded in 2025, Two Delta operates with a high degree of secrecy and has disclosed little about its business beyond research papers published on its website. The company was co-founded by Dvir Zeltzer and Dr. Eli David and is developing technology designed to accelerate AI models and improve the efficiency with which they use computing resources.

1 View gallery Bill Ackman (right) and Eric Schmidt. ( Photo: Jeenah Moon/Bloomberg )

David is a veteran AI researcher and entrepreneur, while Zeltzer brings a background in business and management, including a senior role at Google.

In October 2026, Two Delta published results from a collaboration with Nvidia, in which it used its DCOCO technology to accelerate inference for Nvidia’s Nemotron 3.5 Super VL model. The method is designed to increase the number of tokens a model can generate in each verification step while preserving output quality.

DCOCO builds on a technique known as speculative decoding. In this approach, a smaller, faster model drafts a sequence of tokens that a larger model verifies in a single pass. The process can speed up generation, but some proposed tokens may be rejected. The challenge is to produce drafts that the larger model is more likely to accept.

Two Delta’s method seeks to improve those proposals by combining candidate-token embeddings with contextual information from the drafting model’s hidden states, which are informed by the target model’s activations. It scores possible combinations of neighboring tokens in parallel, helping select more consistent sequences before the larger model verifies them.

In tests across 10 datasets covering mathematics, coding and general-purpose tasks, Two Delta reported that DCOCO achieved average throughput 2.03 times that of standard autoregressive decoding when reasoning was disabled and greedy decoding was used. Under the same configuration, it reported an average throughput improvement of 19.8% over the competing DFlash method. The company also reported throughput gains under different decoding settings and workloads involving up to 32 concurrent requests. The evaluations were conducted on Nvidia DGX B200 hardware using the vLLM inference framework.

These are results reported by Two Delta, rather than an independent assessment of the technology. The company says DCOCO improves the efficiency of token generation while preserving the target model’s output quality.

Two Delta has also published EvoWeb, a benchmark designed to evaluate how well AI coding agents build and maintain web applications as users repeatedly request new features and modifications. Unlike benchmarks that evaluate an agent’s performance on a single prompt, EvoWeb tests an application over 10 successive requests: an initial build followed by nine changes.

The study tested 32 web applications and found that none of the evaluated models maintained full functionality throughout the process. The top performers, Fable 5 and Kimi K3, received scores of 0.694 and 0.684, respectively, on the study’s scale of zero to one.

The main difficulty was integrating new features with existing functionality. Although build failures remained rare, runtime errors increased from 2% at the beginning of the process to 12% after 10 requests, according to the study.

Bill Ackman and Eric Schmidt back secretive Israeli AI startup Two Delta

Mave, an innovative cybersecurity startup focused on combating the surge of AI-driven cyber threats, has successfully secured $34 million in funding. Based in Israel, Mave has developed AI agents that enhance existing security frameworks by gathering information, analyzing alerts, eliminating false alarms, and executing actions that have been pre-approved by human teams.

Raziel Oren, the co-founder and CEO, highlighted the critical nature of security operations within organizations, equating them to physical security personnel monitoring a building’s security feeds. "This sector is fundamentally flawed," Oren stated. "The existing solutions are not evolving quickly enough to address the new threats we face."

With the rapid evolution of AI technologies, cyberattacks are occurring at an unprecedented speed and scale, leading Oren to predict a dramatic surge in incidents involving AI-driven agents. He emphasized that organizations cannot simply hire additional personnel to tackle the issue. "A fundamental shift in how security operations are conducted is necessary to safeguard companies."

Oren partnered with Guy Eshel to launch Mave in 2025, stemming from a longstanding friendship forged during their shared experiences in unit 8200 of the Israel Defense Forces, where they gained valuable insights into cybersecurity challenges.

The focus on AI agents has intensified following numerous high-profile cybersecurity breaches, raising alarm among corporate leaders about the dangers AI poses. In recent earnings calls, many Wall Street executives have detailed their strategies to mitigate the risks associated with rogue AI agents.

AI technology has also been pivotal in revealing security vulnerabilities, as demonstrated by OpenAI's recent announcement that one of its AI models identified a flaw threatening 70% of global websites.

Oren noted that some customers initially hesitate to delegate security operations to AI agents. However, he believes that transparency is crucial for building trust. Mave’s agents meticulously document their processes in thousands of workflows, allowing human staff to understand the rationale behind the agents' decisions and actions.

Mave employs a straightforward pricing model based on the size and specific requirements of each client. The cybersecurity landscape has attracted other startups, such as Dropzone and Prophet Security, which are also creating agent-based solutions.

The $34 million raised by Mave includes $28 million from a Series A funding round led by Red Dot Capital Partners, with contributions from Lumir Growth Partners, along with an earlier $6 million seed round. The company, which currently employs a team of 30, intends to allocate part of this new funding towards team expansion, particularly in marketing roles and further AI development.

Startup secures $34 million to assist security teams in addressing AI threats. Read the pitch deck.

A bright, sunlit startup workspace where a diverse team of cybersecurity professionals leans over a conference table strewn with pitch-deck printouts, open laptops, and coffee cups. On a large screen behind them, a slide reads “AI Threat Defense – $34M Series A.” In the background, swirling red strands of malicious code are held at bay by glowing white shield symbols emanating from the team’s devices, symbolizing their new funding power.

This startup raised $34 million to help security teams keep up with AI threats. Read the pitch deck. - Business Insider

- Mave, a cybersecurity startup, raised $34 million to combat AI-driven cyberattacks.  
- The company's AI agents enhance security operations by filtering alerts and taking approved actions.  
- Mave plans to expand its team for market growth and ongoing AI research initiatives.  

Mave, a cybersecurity startup that uses agents to counter the rise in AI-powered cyberattacks, has raised $34 million in funding.

The Israel-based startup says its AI agents pull together information from a company's existing security systems, investigate alerts, filter out false alarms, and take actions approved by human teams.

"Security operations is the last gate in any organization," cofounder and CEO Raziel Oren said, comparing it to the physical security guards who monitor a building's live camera footage. "This thing is really broken from the inside. The roots of the solutions there are just not keeping the relevant pace with the emerging threats."

AI has transformed cyber threats at a dizzying pace. Attacks can happen much faster and on a much larger scale, and Oren predicted that the increase in agent-powered incidents will be "exponential."

Mave says its own agents help human security teams keep up. "You can't hire your way out of this," Oren told Business Insider. "There must be a change in how you do [security operations] in order to protect your organizations."

Oren cofounded Mave with Guy Eshel in 2025 after years of friendship. The two met when Oren was Eshel's mentor in unit 8200 of the Israel Defense Forces, which sits under the intelligence corps and gave them firsthand experience with cybersecurity issues.

AI agents have come under the spotlight after a string of high-profile security breaches. Corporate leaders have grown increasingly worried about the cybersecurity risks AI poses, and some of Wall Street's executives spent third-quarter earnings calls explaining how they're defending against rogue AI agents.

The technology has also been used by researchers to uncover security flaws: this week, OpenAI said one of its AI models had helped identify a vulnerability that puts 70% of the world's websites at risk.

Oren said it takes some customers time to become comfortable handing over security operations to agents, but that the product's transparency is key to building trust. Mave says its agents track each step of their thousands of workflows, giving human employees insight into the agents' reasoning and actions.

Mave charges its customers a flat rate, depending on the company's size and needs. Other startups have also jumped at the opportunity Mave sees, including agentic cybersecurity companies Dropzone and Prophet Security.

Mave's $34 million funding consists of $28 million from a recent Series A round, led by Red Dot Capital Partners and with participation from Lumir Growth Partners, and an earlier $6 million seed round.

The startup currently has 30 employees and plans to use some of the fresh capital to continue expanding the team, mainly for go-to-market roles and AI research.

Read the pitch deck Mave used to raise its Series A round:

This startup raised $34 million to help security teams keep up with AI threats. Read the pitch deck.

The influx of artificial intelligence (AI) innovation in the legal sector is now attracting talent from prominent Silicon Valley organizations. 

Ethem Coban, previously a researcher at Meta, has recently left the tech giant to establish a legal technology startup, as reported by sources familiar with his move. Coban had focused on training Meta's language models to tackle legal inquiries effectively and is currently in discussions to secure $6 million in funding, valuing his new venture at $25 million.

Negotiations are still in their infancy, and the specifics of the investment could evolve over time.

This startup joins a rapidly expanding field in legal technology, once considered slow to adapt. Over the past two years, legal professionals have begun experimenting with tools leveraging advanced language models. Early adopters such as Harvey and Legora witnessed a surge in demand, sparking a flurry of investments in startups aimed at revolutionizing the legal practice.

This promising landscape is now drawing researchers from the very companies that are creating the foundational AI models.

Coban dedicated nearly two years to Meta, particularly at their Superintelligence Labs, where his efforts included refining models to improve their accuracy on legal matters and minimizing errors known as hallucinations. He has an academic background in law, having studied at American University and Goethe University in Frankfurt.

Meta has not responded to inquiries seeking comments on Coban’s departure.

In conversations with potential investors, Coban has articulated his vision for his new enterprise, which intends to assist law firms and corporate legal departments in crafting and personalizing AI models tailored to their specific knowledge areas. This initiative is based on the belief that organizations will increasingly prefer to develop their in-house AI systems, as it offers greater control over their data compared to relying on external vendors.

Legal firms are recognizing an opportunity to leverage extensive case files as a unique asset that cannot simply be purchased by competitors. For instance, Kirkland & Ellis, a top firm in terms of revenue, has allocated $500 million towards the creation of proprietary software, while Latham & Watkins is investing in Nvidia servers to tailor their models internally.

For Coban, the real challenge lies not in convincing law firms of the necessity of AI adoption, but rather in demonstrating that having an insider with profound knowledge of model design is advantageous for building these systems effectively.

The surge in legal AI has attracted a researcher from Meta's Superintelligence Lab.

A researcher in a crisp professional suit, seen from behind, steps out of a sleek corridor marked “Meta Superintelligence Lab” with a large Meta logo above the doorway. Warm natural sunlight floods the polished floor. Beyond the doors, in a sunlit courtyard, a monumental golden scales-of-justice statue rises on a pedestal, its chains subtly etched with glowing AI circuit patterns and surrounded by stacked law books, symbolizing the new legal AI frontier.

The legal AI boom just lured a researcher out of Meta's Superintelligence Lab - Business Insider

- Ethem Coban left Meta to launch a legal tech startup, seeking $6 million investment.  
- Lawyers are increasingly adopting AI tools, transforming how legal work is done.  
- Law firms aim to customize AI models for competitive advantages, reducing reliance on external licenses.  

The legal industry's artificial intelligence gold rush is luring talent from Silicon Valley's biggest labs.

Ethem Coban, a researcher at Meta, left the company earlier this month to launch a legal tech startup, according to a person familiar with the matter. Coban, who worked on teaching Meta's language models to reason through legal questions, has been in talks to raise $6 million at a $25 million valuation.

Talks are still early, and the terms of the deal could change.

The startup is the latest entrant in an increasingly crowded legal tech market. Long seen as laggards, lawyers have spent the past two years experimenting with tools built on large language models. Early movers like Harvey and Legora saw a crush of demand, setting off a wave of investment in startups promising to transform how lawyers work.

Now, that opportunity is attracting researchers from the companies building the underlying models.

Related video Microsoft is betting big on AI. Company insiders have serious doubts.

Coban spent almost two years at Meta, most recently at its Superintelligence Labs, where he helped train models to give better answers to legal questions and reduce hallucinations, the person said. He previously studied law at American University and Goethe University in Frankfurt.

A spokesperson for Meta didn't respond to a request for comment.

Coban has told investors that his new company aims to help law firms and corporate legal teams build and customize models around their expertise. The bet is that companies will increasingly want to own their AI rather than license it from outside companies, because it gives them more control over their data.

Law firms also see a chance to turn mountains of case files into a competitive advantage that other firms can't simply buy off the shelf. Kirkland & Ellis, the industry's revenue leader, has committed $500 million to building proprietary software. Latham & Watkins has been buying its own Nvidia servers to customize models in-house.

For Coban, the challenge won't be convincing law firms that they need artificial intelligence. It will be persuading them that someone who knows how the models work from the inside is best suited to build them.

The legal AI boom just lured a researcher out of Meta's Superintelligence Lab

Legal AI firm Legora revealed on Wednesday that it has successfully secured $550 million in a Series E funding round, bringing its total valuation to $8.8 billion. This significant investment is indicative of the increasing demand for AI-driven solutions within the legal industry, which is transitioning from simple AI assistants to more complex AI services aimed at streamlining workflows for legal practitioners.

This announcement comes a month after competitor Harvey disclosed its own $550 million funding round, achieving a valuation of $15.5 billion. Furthermore, in March, Harvey reported an impressive annual recurring revenue (ARR) exceeding $200 million.

Legora's recent funding announcement follows an earlier statement in April, when it disclosed a $50 million extension to its Series D round, which had previously raised $550 million and valued the company at $5.6 billion. At that time, Legora's ARR was approximately $100 million; however, the latest figures show that it has doubled, now exceeding $200 million.

This latest funding round has attracted several prominent investors, including Atreides Management, which took the lead, alongside notable firms such as Baron Capital, Bessemer Venture Partners, General Catalyst, Menlo, and others. The new capital is intended to support Legora’s continued expansion throughout North America.

Legal AI company Legora secures $550M in Series E funding, valuing the firm at $8.8B.

Sunlight floods a sleek high-rise boardroom where a diverse team of professionals in sharp business attire raises coffee mugs in celebration. On the glass wall behind them, the Legora logo stands large and prominent. Open law books and tablets streaming subtle wisps of digital code lie on the polished table. Above, a clean digital display reads “Series E – $550 M” and “Valuation 8.8 B.”

Legal AI startup Legora raises $550M Series E at $8.8B valuation - TechCrunch

- Legora raised $550 million in Series E, valuing the company at $8.8 billion.  
- The company automates workflow tasks for legal professionals, enhancing their productivity.  
- Legora's annual recurring revenue has surpassed $200 million, doubling since April's announcement.

In Brief

Legal AI startup Legora announced Wednesday that it raised a $550 million Series E round, valuing it at $8.8 billion.

The company helps automate and assist with workflow tasks for legal professionals, as the industry moves beyond AI assistants to AI services that help them get work done.

The news comes just a month after Harvey, Legora’s biggest competitor, announced a $550 million raise at a $15.5 billion valuation; back in March, Harvey also said its annual recurring revenue (ARR) was more than $ 200 million.

Legora’s Series E follows the company’s April announcement of a $50 million extension of a $550 million Series D round, which valued the company at $5.6 billion. At the time, the company said its ARR had reached $100 million. That number now stands at over $200 million, the company shared on Wednesday.

Investors in the latest financing include Atreides Management, which led the round, as well as Baron Capital, Bessemer Venture Partners, General Catalyst, Menlo, and other top names. The company plans to use the fresh capital to further expand across North America.

Legal AI startup Legora raises $550M Series E at $8.8B valuation

Recently, I’ve had the opportunity to try out a remarkable open-weight AI coding model named Flint. This innovative model indicates that the concept of fully recursive self-improvement (RSI)—which involves using AI to enhance AI—is advancing rapidly, sparking both excitement and debate within the tech community.

Flint is the inaugural product from Recursive, a company dedicated to harnessing RSI to propel AI development and various scientific fields forward. According to Richard Socher, the CEO and co-founder of Recursive, "Nearly every aspect of the entire model pipeline was actually done and improved upon by AI."

Socher explained that the company utilized its recursive AI system, known as the Eureka Machine, to automate the majority of the model's development process. This strategy allowed a small team of researchers to design and refine a world-class AI model in just a matter of months.

The Eureka Machine contributed numerous innovations, including a novel approach to training the foundational model, new architectural designs, and code specifically optimized for particular silicon chips. Benchmark scores released by Recursive indicate that Flint surpasses comparable models in performance on coding tasks.

AI’s role in enhancing AI has already proven to be transformative, streamlining the refinement process and enabling the exploration of new concepts with significantly less human intervention required.

This Startup Fully Commits to Recursive Self-Improvement, and Its New Model Is Remarkably Impressive.

A bright, sunlit modern startup lab with warm natural light flooding through floor-to-ceiling windows. An adult engineer’s silhouette stands before a massive wall-mounted screen filled with lines of code folding into themselves like an infinite spiral staircase. Translucent ribbons of code drift upward toward the ceiling, suggesting recursive self-improvement. In the corner, a sleek server rack hums quietly. On a white-painted wall behind the engineer, the startup’s stylized spiral logo is painted large and bold, tying the scene together.

This Startup Is All In on Recursive Self-Improvement. Its Latest Model Is Scary Good - WIRED

- Flint, from Recursive, demonstrates advancements in recursive self-improvement for AI development.  
- The Eureka Machine automated development, allowing rapid innovation with minimal human involvement.  
- Flint reportedly outperforms similar models in coding tasks, showcasing its advanced capabilities.

For the past few days I’ve been testing an impressive open-weight AI coding model called Flint. It suggests that fully recursive self-improvement, or RSI—an exciting and controversial approach that uses AI to build AI—is getting closer.

Flint is the first offering from Recursive, a company that aims to use RSI to advance both AI itself and other areas of science using AI. “Nearly every aspect of the entire model pipeline was actually done and improved upon by AI,” Recursive CEO and cofounder Richard Socher told me.

Socher says his company used its recursive AI system, dubbed the Eureka Machine, to automate most elements of the model’s development. This enabled a company with a handful of researchers to build a world-class model in just a few months, he says.

Socher told me that the Eureka Machine devised a number of innovations, including a new way of training the base model, new model architectures, and code optimized for specific silicon chips. Recursive shared benchmark scores that suggest Flint outperforms other models of similar size and architecture on coding tasks.

AI is already widely used to help build, test, and improve AI thanks to its amazing coding skills. This can speed up the process of refining a model and testing new ideas without the need for as much input from humans.

This Startup Is All In on Recursive Self-Improvement. Its Latest Model Is Scary Good

For many years, Silicon Valley had a tough time with hardware investments. However, a significant shift is happening as investors now funnel massive sums—tens of billions—into robotics, which are capable of performing real-world tasks.

A unique combination of factors has positioned robotics as one of the most promising areas in tech. As interest extends beyond chatbots, investors are drawn to the possibilities presented by advancements in affordable hardware, innovative AI models, labor shortages, and geopolitical competition, particularly with China. These aspects are driving a renewed focus on machines capable of executing manual labor.

According to PitchBook data, robotics and physical AI firms collected an impressive $33.4 billion globally in just the first half of this year. This figure surpasses the $28.7 billion raised throughout the entire previous year and represents a dramatic increase from under $5 billion in 2020.

Big checks are being written to finance both the robots themselves and the underlying AI technology. For instance, defense tech company Anduril secured $5 billion in May, with heavyweights Thrive Capital and Andreessen Horowitz leading the round. Saronic, an autonomous shipbuilding company, raised $1.75 billion in March, with guidance from Kleiner Perkins. Skild AI, developing software for various robotic applications, raised $1.4 billion in January, also led by SoftBank, while Physical Intelligence attracted nearly $1 billion in May, co-led by Leopold Aschenbrenner's firm and Lux Capital.

Ken Goldberg, a robotics professor at UC Berkeley and cofounder of Ambi Robotics, expressed his astonishment to Business Insider: "This is unprecedented. In my 45 years in the field, I've never seen this level of funding, energy, and excitement in robotics."

The article marks the debut of "Robonomics," a series by Business Insider that explores the burgeoning world of robotics and their influence on industries, jobs, and everyday life. The series aims to delve into environments where humans and robots collaborate, meet the innovators and laborers involved in this surge, and evaluate how this technology is transforming sectors ranging from manufacturing to defense. There will also be discussions about the integration of robots into daily life and the hurdles that remain.

The potential rewards are massive. Unlike software, which is primarily digital, robotics target physical tasks in diverse settings—from factories and construction sites to household helpers. Nvidia's CEO Jensen Huang, a vocal advocate for "physical AI," notes that robotics offer the tech sector its first chance to tap into a $50 trillion market that has seen little technological advancement until recently. Numerous former Nvidia employees have taken the plunge into launching their own robotics ventures.

However, investors are valuing many startups as if the robot revolution is already a reality. Some companies still in early development are reaching multibillion-dollar valuations despite minimal revenue and limited product launches. Figure AI, a humanoid robot manufacturer, achieved a $39 billion valuation last year, even as its deployments are confined to small-scale factory tests.

"Robotics will progress, but not without significant financial loss along the way," cautioned Sarah Guo, founder of the venture firm Conviction, which invests in robotics companies like Sunday Robotics and Watney. "Both of those outcomes are likely."

Robots have long existed in various forms, from assembly line machinery to household cleaning devices like Roombas. Historically, though, investors found it challenging to generate profit from these machines. The challenges stemmed from the high costs associated with building robots, integrating supply chains, and manufacturing machines designed for singular purposes. Why invest in robotics when a college dropout might create the next viral app?

Guo often warned fellow investors that robotics was a risky area with potential for long-term losses, stating, "The models didn't work; every robot was a delicate, single-function system crafted for specific tasks."

A few investors, like Ajay Agarwal from Bain Capital Ventures, took the risk during the early internet boom. He supported Kiva Systems in 2004, a pioneer in warehouse automation with mobile robots that transported products to workers. Bain provided continuous funding, as Kiva struggled to attract investment on Sand Hill Road. Ultimately, Amazon acquired Kiva for $775 million in 2012, with Bain being its sole institutional backer.

Nearly twenty years later, advancements in AI are altering the landscape. Improved machine learning capabilities enable robots to interpret their environment, make informed decisions, and learn new tasks. This evolution raises the potential for creating more adaptable machines rather than crafting unique robots for every single duty. Additionally, decreasing costs in computing and hardware enhance the financial viability of robotics.

Major players in Silicon Valley are now pursuing this vision. After raising $1.7 billion in July, Travis Kalanick's Atoms is focusing on automating food, mining, and transport sectors. Rivian CEO RJ Scaringe's Mind Robotics has attracted $900 million in 2023, while former Cruise CEO Kyle Vogt's The Bot Company has gathered at least $300 million in funding.

Big Tech firms and AI research labs are also eager to participate. OpenAI is hiring for robotics positions with enticing salaries of up to $500,000 annually. Meta acquired the robotics startup Assured Robot Intelligence in May and is experimenting with robots in its data centers. Google has invested in Apptronik, a humanoid robot company, and is collaborating with DeepMind on AI advancements for robotics. Meanwhile, Elon Musk is betting on Tesla's future with its Optimus humanoid robot.

Venture capital firms are allocating funds specifically for robotics as well. In August, Andreessen Horowitz raised $1.1 billion for its inaugural fund dedicated to the physical infrastructure of AI, including robotics.

Despite this surge in interest, seasoned robotics investors express concern. There are whispers about "hardware tourists"—investors drawn to the robotics market due to the AI buzz but oblivious to the challenges of creating reliable physical machines. Caution is warranted, as reflected in the autonomous vehicle investments of the 2010s that resulted in many failed startups as the technology proved far trickier and more costly to implement than anticipated. Presently, only a small number of well-financed companies have succeeded in presenting robotaxi services to consumers.

The robotics sector is fraught with its own challenges. Unlike large language models trained using extensive data sets from the internet, robots lack sufficient data to learn how to function in the physical world. Many startups are attempting to gather this data, but understanding what quantity is adequate remains elusive. Furthermore, even with advanced AI, the difficulties associated with manufacturing, maintaining, and deploying high-end hardware persist.

Charly Mwangi, a partner at VC firm Eclipse, described the landscape: "The most complicated aspect of physical AI is the physicality itself. The next few years will be challenging for the robotics sector as companies transition from research settings to actual deployments."

Speculative areas, such as humanoid and general-purpose robots, are receiving substantial investments. According to venture firm F-Prime, these were the two fastest-growing sectors in robotics last year. "You can't continually secure funding based on mere demonstrations," cautioned Sanjay Aggarwal, an F-Prime partner and coauthor of its annual robotics report. "I would be surprised if all these firms remain independent in the next two or three years."

Yet overall, some startups are transitioning into viable businesses. Field AI, working on an AI system that operates across multiple robots, has surpassed $100 million in revenue and contracts with over 30 customers in the construction, data center, and defense sectors. Similarly, Skild AI reported recently achieving $100 million in annual recurring revenue from more than 60 clients, with projects involving robots that assemble Nvidia's Blackwell chips alongside Foxconn and kitchen initiatives with Japan's Mitsui conglomerate.

As the robotics field stands on the brink of significant advancements, many believe transformative breakthroughs could enable machines to integrate into various aspects of the economy. Brad Porter, former head of Amazon Robotics and current leader of workplace robotics startup Cobot, suggests that major strides in dexterous manipulation—the ability for robots to handle objects with human-like deftness—are just around the corner. 

"I see a 'ChatGPT moment' for robotics coming when we crack dexterous manipulation," stated Porter. "After that, there could be an explosion of robotic applications."

The critical question remains whether these groundbreaking developments will materialize swiftly enough to validate the heavy investments being made. At a recent Business Insider event, Menlo Ventures partner Matt Murphy highlighted robotics as a sector receiving considerable funding "before demonstrating significant traction." He cautioned that easy access to capital can lead to subpar performance by startups.

For investors, this presents a stark reality: the risk of encountering significant pitfalls is much higher.

Reasons Behind Silicon Valley VCs' Multi-Billion Dollar Investments in Robotics and AI

A sunlit Silicon Valley boardroom with floor-to-ceiling windows overlooking rolling tech campus hills. Three sharply dressed VCs sit at a sleek glass table; one leans forward, handing over a stack of futuristic digital tokens and shaking hands with a polished humanoid robot in a tailored business jacket. Warm natural light highlights metallic circuits etched into the tabletop, and in the distance the silhouette of the Bay Area tech skyline under a clear blue sky.

Robots used to be a great way to lose money. Now Silicon Valley investors are betting billions on them. - Business Insider

- Robotics and physical AI companies raised $33.4 billion in 2023's first half, outpacing last year's total.  
- Major investments are flowing into startups developing machines for diverse physical tasks and applications.  
- Industry veterans warn of challenges ahead, citing past failures in autonomous vehicle funding and deployment.  

For decades, hardware was a hard sell in Silicon Valley. Now investors are pouring tens of billions into machines that can do things in the real world.

A Goldilocks set of conditions has made robotics one of tech's hottest bets. As investors look beyond chatbots to the next frontier in AI, cheaper hardware, smarter AI models, labor shortages, and competition with China are fueling interest in machines capable of performing physical work.

Robotics and physical AI companies raised $33.4 billion globally in the first half of this year alone, according to PitchBook data. That's already more than the $28.7 billion the sector raised in all of last year — and an astronomical rise from 2020, when funding came in at just under $5 billion.

Investors are writing billion-dollar checks to fund both the machines and the AI brains behind them. Defense company Anduril raised $5 billion in May, co-led by Thrive Capital and Andreessen Horowitz. Autonomous shipbuilder Saronic raised $1.75 billion in March in a round led by Kleiner Perkins. Skild AI, which builds software to control different kinds of robots, raised $1.4 billion in January in a round led by SoftBank. Physical Intelligence, which is also building an AI model to command robots, raised about $1 billion in May, co-led by Leopold Aschenbrenner's Situational Awareness and Lux Capital.

"This is off the charts," Ken Goldberg, a UC Berkeley robotics professor and cofounder of warehouse robotics startup Ambi Robotics, told Business Insider. "In my 45 years in this field, this is the most funding and energy and interest in robotics that I've ever seen."

This is the first installment of Robonomics, Business Insider's new series on the rise of robots and their impact on business, jobs, and everyday life. We'll go inside factories and warehouses where humans work alongside robots, meet the founders, investors, and workers driving the boom, and examine how the technology is reshaping industries from manufacturing to defense. We'll also ask how close robots are to becoming part of everyday life, and what is still standing in their way.

The prize is enormous. Unlike software, which is largely confined to work done on screens, robotics companies are going after physical labor, from factories and construction sites to robot butlers in every home. Nvidia CEO Jensen Huang, who popularized the term "physical AI," has described robotics as the "technology industry's first opportunity to address a $50 trillion industry that has largely been void of technology until now." Many Nvidia veterans have launched their own robotics startups.

Nvidia CEO Jensen Huang Bloomberg/Bloomberg via Getty Images

But investors are pricing companies as though the robot revolution has already arrived. Startups still in research and development command multibillion-dollar valuations, often with little revenue and few deployments to show for it. Humanoid maker Figure AI, for example, secured a $39 billion valuation last year, even as its deployments remain limited to small factory pilots.

"Robotics is going to happen, and a lot of money will still be lost," said Sarah Guo, founder of the venture firm Conviction, which is an investor in robotics startups Sunday Robotics and Watney. "Both things are true."

From not to hot

From industrial arms assembling electronics to Roombas cleaning floors, robots have long been part of everyday life.

What venture investors struggled to see was how to make money from them. Building robots required expensive hardware, factories, and supply chains, often to produce machines that could perform just one narrow task. Why fund that when you could back a college dropout building the next viral app?

Guo used to tell fellow investors that robotics was a good way to lose a lot of money over many years. "The models didn't work," she said. "Every robot was a brittle, single-purpose system hand-built for one task."

Ajay Agarwal, a partner at Bain Capital Ventures, was among the few investors willing to bet on robotics during the internet era. In 2004, he backed Kiva Systems, which pioneered warehouse automation with fleets of mobile robots that brought products directly to human workers.

A Kiva Systems robot moves a shelf to fill orders at an Amazon fulfillment center Bloomberg/ Getty Images

Bain led every successive funding round because Kiva was "not able to raise money on Sand Hill Road," Agarwal told Business Insider. By the time Amazon bought it for $775 million in 2012, Bain was its only institutional investor.

More than two decades later, AI has changed the calculation. Advances in machine learning are helping robots perceive their surroundings, make decisions, and adapt to new tasks, raising hopes that companies can build more versatile machines instead of engineering a different robot for every job. Cheaper computing and hardware are also making the economics more attractive.

Business Insider has Games! Want more Business Insider in your news feed? Add BI in Google so our reporting is easier to find when you’re searching for what matters. Add to Preferred Sources

Some of Silicon Valley's biggest names are now chasing that vision. Former Uber CEO Travis Kalanick's Atoms raised $1.7 billion in July to pursue automation in food, mining, and transportation. Rivian CEO RJ Scaringe's Mind Robotics has raised $900 million this year, while former Cruise CEO Kyle Vogt's home robotics startup, The Bot Company, has raised at least $300 million.

Big Tech and AI labs, too, are joining the rush. OpenAI has advertised robotics jobs with base salaries reaching $500,000 a year. Meta acquired robotics startup Assured Robot Intelligence in May and is testing robots in its data centers. Google has invested in humanoid maker Apptronik, which is also working with DeepMind to develop AI models for robots. Elon Musk, meanwhile, is staking Tesla's future on the Optimus humanoid.

Tesla's Optimus humanoid CFOTO/ Getty Images

Some venture firms are setting aside dedicated capital for robotics. In August, Andreessen Horowitz raised $1.1 billion for "Machine Age," its first fund focused on the physical infrastructure behind AI, including robotics.

The gold rush

For longtime robotics investors, alarm bells are ringing. Industry veterans whisper about "hardware tourists" piling into robotics deals — investors drawn by AI hype who underestimate how difficult it is to build machines that work reliably in the real world. There is precedent for caution. Venture firms poured billions into autonomous vehicles in the 2010s, only to see startups fold or consolidate as self-driving technology proved harder and more expensive to commercialize than expected. Years later, only a handful of deep-pocketed players have managed to bring robotaxis to the public.

Robotics faces its own formidable obstacles. Unlike large language models trained on vast amounts of internet text and images, robots lack a comparable trove of data showing them how to act in the physical world. Dozens of startups are now scrambling to collect it, but nobody knows how much will be enough. And even smarter AI won't eliminate the challenges of manufacturing, maintaining, and deploying expensive hardware.

"The most difficult part of physical AI is the physical part," said Charly Mwangi, a partner at VC firm Eclipse who has backed The Bot Company and robotics startup Genesis AI. "The next few years are going to be really, really hard for the robotics industry as companies go from the lab to deployments."

Some of the most speculative robotics bets are attracting the biggest checks. Humanoids and general-purpose robot models were the two fastest-growing areas of robotics investment last year, according to venture firm F-Prime. "You cannot raise money on demos forever," said Sanjay Aggarwal, an F-Prime partner and coauthor of the firm's annual robotics report. "I would personally be surprised if all these companies independently exist two or three years from now."

There are signs that some startups are becoming real businesses. Field AI, which is building an AI "brain" that works across robots, has crossed $100 million in revenue and customer contracts from more than 30 customers, including construction firms, data center operators, and defense companies, Business Insider previously reported. Skild AI said last month that it had reached $100 million in annual recurring revenue from more than 60 customers. Its projects include robots assembling Nvidia's Blackwell chips with Foxconn and kitchen pilots with Mitsui, a Japanese conglomerate.

An Nvidia GB200 Blackwell GPU Bloomberg/ Getty Images

Believers see robotics on the cusp of a breakthrough that could put machines to work in nearly every corner of the economy. Brad Porter, the former head of Amazon Robotics who now runs workplace robotics startup Cobot, says the field could be nine to 18 months away from major advances in dexterous manipulation — the ability of robots to handle objects with the ease of human hands.

"When I think of a 'ChatGPT moment' for robotics, it is cracking dexterous manipulation," Porter said. "Then there will be a Cambrian explosion of robots."

The question is whether these breakthroughs will arrive quickly enough to justify the money pouring in.

At Business Insider's Growth Mode event this month, Menlo Ventures partner Matt Murphy singled out robotics among the fields attracting enormous amounts of financing "before they're really showing traction." Abundant funding, he warned, can make startups "sloppy."

For investors, he said, that means "a lot more opportunity to have a really big crater."

Why Silicon Valley VCs Are Investing Billions in Robotics and AI

Fireworks AI is in the process of seeking additional investment, with projected valuations soaring between $30 billion and $40 billion. This represents a significant increase from its previous valuation of $17.5 billion just months prior, according to insider sources.

This anticipated surge in valuation highlights the keen interest from investors in companies providing the critical computing power and tools essential for the AI industry's growth, even as Fireworks navigates the repercussions of a recent security breach.

Negotiations regarding the new funding are still in their infancy, and the estimated valuations may fluctuate. Fireworks AI has chosen not to comment publicly on the matter.

Founded in 2022, Fireworks has been established by seasoned professionals from Meta and Google's AI divisions. CEO Lin Qiao, who previously headed the PyTorch team at Meta, leads a team that includes former engineers from Meta and a prior head of Google's Vertex AI. The company offers infrastructure and software solutions that enable businesses to run, optimize, and deploy both open-weight and custom AI models.

In July, Fireworks completed a $1.5 billion Series D funding round, securing a valuation of $17.5 billion. This round, which was led by Atreides Management, Index Ventures, and TCV, also saw participation from Nvidia, Lightspeed Venture Partners, Bessemer Venture Partners, and Menlo Ventures. This funding followed a $250 million Series C round at a valuation of $4 billion in late 2025, co-led by Lightspeed, Index Ventures, and Evantic. Recent reports suggest that Fireworks is contemplating a round that could see its valuation reach $30 billion.

Unlike companies attempting to compete directly with ChatGPT through consumer-focused chatbots, Fireworks specializes in what it defines as "specialized intelligence"—customized models that are tailored to clients' proprietary data, workflows, and applications. As of July, the company announced it had surpassed $1 billion in annualized revenue, processing over 40 trillion tokens daily, 95% of which originated from client-specific models.

Fireworks operates within a fiercely competitive segment of the AI landscape, facing competition from model-infrastructure startups such as Together AI, Baseten, and Replicate, as well as major cloud providers like Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle.

The latest fundraising discussions come on the heels of a security incident, wherein unauthorized use of internal credentials was detected earlier this month, allowing an intruder access to data from several client accounts.

Qiao reassured stakeholders on social media, stating, "We take this seriously and will be transparent about what happened and what we find. I'm personally committed to turning this attack into the most formidable security response in AI Infrastructure."

Infrastructure startup Fireworks AI is in preliminary discussions for new funding, valuing the company between $30 billion and $40 billion.

Sunrise over a launch pad built of interlocking circuit boards, where a stylized firework-shaped rocket, its side emblazoned with the Fireworks AI logo, stands ready for liftoff. Suited investors cluster beside a large banner announcing “$30–40 B valuation,” as warm natural light bathes a horizon of chip-inspired skyscrapers symbolizing next-gen infrastructure.

Infrastructure startup Fireworks AI is in early talks for new funding at a valuation between $30 billion and $40 billion - Business Insider

- Fireworks AI aims to secure funding at a valuation of $30-40 billion.  
- The company experienced a valuation surge from $17.5 billion in July.  
- A recent security incident raised concerns but leadership pledges a strong response.

Fireworks AI is seeking to raise fresh funding at a valuation between $30 billion and $40 billion, a dramatic jump just months after the AI-infrastructure startup was valued at $17.5 billion, according to people familiar with the matter.

The potential valuation leap reflects investors' appetite for the companies selling the computing power and tools fueling the AI boom, even as Fireworks AI works through the fallout from a recent security incident.

The talks around the financing are still early, and valuation targets could still change. Fireworks declined to comment.

Founded in 2022, Fireworks was built by veterans of Meta and Google's AI efforts. Chief executive Lin Qiao previously led the PyTorch team at Meta, while cofounders include former Meta infrastructure engineers and a former leader of Google's Vertex AI. The company rents out computing infrastructure and software that lets businesses run, fine-tune, and deploy open-weight and custom AI models.

In July, Fireworks announced a $1.5 billion Series D funding round at a $17.5 billion valuation, led by Atreides Management, Index Ventures, and TCV, with participation from Nvidia, Lightspeed Venture Partners, Bessemer Venture Partners, and Menlo Ventures. That round followed a $250 million Series C at a $4 billion valuation in late 2025, co-led by Lightspeed, Index, and Evantic. Last month, The Information reported that Fireworks has been considering a round at a $30 billion.

Related video What are the real-life consequences of AI?

Rather than trying to build a consumer chatbot to rival ChatGPT, Fireworks sells what it calls specialized intelligence: models customized on a customer's proprietary data, workflows, and use cases. The company said in July that it had surpassed $1 billion in annualized revenue and was serving more than 40 trillion tokens a day; it said 95% of those tokens came from models tailored to customers.

The company operates in a highly competitive layer of the AI stack. Rivals include model-infrastructure startups such as Together AI, Baseten, and Replicate, as well as cloud giants such as Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle.

The fundraising talks come shortly after a security incident at Fireworks AI. The company said it identified unauthorized use of internal credentials earlier this month that allowed an intruder to access data from some customer accounts.

"We take this seriously and will be transparent about what happened and what we find," Qiao posted on X. "I'm personally committed to turn this attack into the most formidable security response in AI Infra."

Please send along your tips. Email Ben or Katie anytime or contact them on Signal at benbergman.11 and kroof.26. Sign up for Business Insider's weekly VC Insider newsletter here.

Infrastructure startup Fireworks AI is in early talks for new funding at a valuation between $30 billion and $40 billion

Welcome to the latest edition of Eye on AI, where we bring you important updates from the world of artificial intelligence:

In a significant move, Bill Gross of Idealab has shared an astounding 1.2 million AI-generated business concepts with the public.

A former employee from OpenAI, accused of introducing security vulnerabilities, was denied bail amid concerns regarding potential flight risks.

OpenAI has reported additional incidents involving AI-related breaches concerning sensitive financial, medical, and governmental information.

AI safety specialists have expressed dissatisfaction with a proposed AI regulation bill by Anthropic.

Anthropic has committed 5% of its shares to support a new public benefit fund.

There are indications that neural networks might be more comprehensible than previous assumptions suggested.

OpenAI’s latest model, GPT-6 Astra, has successfully achieved high scores on a new benchmark in theoretical physics.

Concerns over AI’s potential impact on employment continue to grow, but the anticipated job apocalypse has yet to materialize. However, indicators suggest that job opportunities may be dwindling, and AI adoption plays a role in this shift.

According to a recent Bloomberg report, U.K. banks are trimming their back-office workforce significantly, influenced by AI implementations by Anthropic. The company's hiring data reveals a notable decline in the recruitment of junior technical staff; since early 2025, the percentage of such hires has dropped from 27% to just 9%. Furthermore, the organization is also employing significantly fewer administrative personnel than typically expected for its growth rate. If this trend continues across other companies integrating AI, we might witness a considerable decrease in the overall job market.

On a more optimistic note, the AI revolution could fuel entrepreneurial endeavors, potentially alleviating some job loss concerns. Bill Gross, founder and CEO of Idealab, is spearheading a new initiative that demonstrates this potential. 

Established in 1996, Idealab has played a pivotal role in nurturing around 175 companies. Over the years, Gross has devised 31 methodologies aimed at generating business concepts. He categorizes these methodologies into three primary areas: opportunities for addressing specific business challenges, enablers stemming from technological advancements and scientific breakthroughs, and shifts relating to regulations or evolving social and economic factors.

By inputting these methodologies into an AI model along with around 17 million documents—comprising news articles, research findings, expired patents, and community discussions—Idealab's AI has produced over 1.2 million unique business ideas. Excited about this wealth of information, Gross has opted to share these ideas with entrepreneurs globally for free.

Gross emphasized his desire to inspire entrepreneurship beyond his local area, stating, “I can’t start 1.2 million companies, but I can empower many to do so.” 

He views business as a fundamental means of addressing challenges, and while discussions often focus on AI's risks, he believes its capacity for problem-solving is undervalued.

The platform idealab.ai allows users to explore these 1.2 million ideas based on industry, skills, personal interests, and geographic location. Users can input their resumes or LinkedIn profiles to receive tailored business suggestions. For those with existing ideas, the platform enables refinement of plans or the generation of additional concepts.

Furthermore, established organizations and universities can leverage their patent portfolios to unearth new business opportunities. Gross highlighted that many patents within large companies remain dormant. Through idealab.ai, he aims to unlock this intellectual property, fostering the creation of new business units or spin-offs. In a trial with Caltech’s patents, Gross identified the potential to create approximately 3,450 new businesses, and with 3M’s 21,000 patents, he estimated 1,549 potential ideas.

Gross envisions idealab.ai leading to the establishment of 1 million new companies and generating 100 million new jobs, politically offsetting the estimated job losses AI could bring, as projected by Goldman Sachs. He envisions these new roles being more fulfilling and impactful than those that may be lost, stating, “These would be jobs that give people greater purpose and passion.” 

While there’s still uncertainty about whether idealab can fulfill these ambitious goals, evidence suggests that AI is facilitating a new wave of entrepreneurship. Recent Census Bureau statistics show U.S. business formations are soaring, rising nearly 11% since the end of the pandemic, with similar trends emerging globally. Thus, the narrative surrounding AI is more nuanced than the prevailing concerns of job loss and wealth concentration.

In other AI-related news:

Jeremy Kahn provides an overview of a range of topics, from an ex-OpenAI engineer involved in a compromised security situation to Anthropic's proposed AI regulations facing scrutiny.

China is drafting new regulations that could control the export of advanced AI technologies aimed at safeguarding its strategic interests, reflecting an evolving stance on AI governance.

Anthropic's proposed AI safety bill has attracted criticism for potentially being weaker than existing measures, prompting the organization to consider revising its terms in response to public feedback.

Investors are becoming skeptical about the valuation of AI data center developer Nscale, pointing to recent challenges in the sector that are leading to a more cautious outlook on investments.

A recent incident revealed a hacker using AI agents to infiltrate over 100 major companies, demonstrating the extent to which AI can enable sophisticated cyberattacks with minimal oversight.

In research discussions, new insights are emerging regarding the interpretability of neural networks, challenging previous assumptions about their complexity.

For those interested in further developments, we invite you to check out our vodcast, Fortune AI Weekly, where we discuss exclusive interviews and the latest AI insights.

We hope you found this overview engaging and informative. Stay tuned for more updates from the fast-evolving AI landscape.

Bill Gross of Idealab donates 1.2 million AI-generated business concepts.

A bright, airy co-working space flooded with warm natural light. Bill Gross stands behind a sleek wooden table, gently releasing a cascade of 1.2 million miniature glowing golden lightbulbs into the air like sparks of inspiration. A few lightbulbs rest in outstretched hands at the front. On the back wall, a bold Idealab logo anchors the scene. The atmosphere feels optimistic, generous, and brimming with innovation.

Idealab's Bill Gross is giving away 1.2 million AI-generated business ideas - Fortune

- Idealab's Bill Gross offers 1.2 million AI-generated business ideas to boost entrepreneurship.  
- Anthropic reports decreased junior technical hiring, indicating AI's impact on job availability.   
- OpenAI's GPT-6 Astra excels in a rigorous theoretical physics benchmark, achieving a 98% score.  

Hello and welcome to Eye on AI. In this edition:

Idealab’s Bill Gross gives away 1.2 million AI-generated business ideas.

Ex-OpenAI employee accused of planting bugs denied bail.

OpenAI discloses more rogue AI incidents involving financial, medical, and government data.

AI safety experts unhappy with Anthropic’s proposed AI bill.

Anthropic pledges 5% of its shares to new public benefit fund.

Maybe neural networks aren’t as hard to understand as we thought.

OpenAI’s GPT-6 Astra aces new theoretical physics benchmark.

There’s been a lot of concern, understandably, about AI eliminating jobs. So far, the great job apocalypse hasn’t shown up. But there are hints out there that jobs are getting harder to find and that AI is part of the reason.



Bloomberg reported this week that banks in the U.K. are slashing back-office jobs, thanks in part to AI Anthropic unveiled data showing that as it has deployed AI throughout its fast-growing organization, who it hires has shifted. Far fewer junior employees are being hired in technical roles than previously. Since early 2025, hiring of junior technical employees has fallen from 27% of all new hires to 9%, the company said. Anthropic is also hiring a lot fewer administrative staff than a company growing at its pace typically would. If other companies follow this pattern as they adopt more AI, there are likely to be significantly fewer jobs in the economy.



While this may be worrying, a little-discussed feature of the AI revolution is how it could supercharge entrepreneurship—and how a surge in new businesses could partly offset some of AI’s negative impacts on employment. Yesterday, Bill Gross, the founder and CEO of Idealab and ProRata, unveiled a new project that highlights this potential.



Idealab was one of the first startup incubators when it was launched back in 1996. Since then, it has helped spawn some 175 companies. Over the years, Gross has developed 31 techniques—he calls them “methodologies”—for generating business ideas. Really, these are creative prompts for thinking about idea generation. Gross breaks these down into three big categories: opportunities, which he says are things such as solving a business pain point, removing friction, removing waste, removing danger, and improving design, among others; enablers, such as new technologies, scientific advances, and new materials, processes, or changing cost curves that make solutions more feasible than before; and shifts, such as regulatory changes, as well as behavioral, economic, cultural, or geopolitical movements that make the timing for a particular business seem right.



Gross fed these 31 methodologies to an AI model along with some 17 million source documents that included news stories, research reports, expired patents, public company filings, and threads from online community discussion boards. In this way, Idealab’s AI generated more than 1.2 million business ideas. And Gross has decided to give those ideas away for free.



“I’ve always wanted to have a positive impact with Idealab by starting many companies, but I can’t start 1.2 million,” Gross told me in an interview over the weekend. “But what about if I can empower entrepreneurship around the world way more than I’ve been able to do in Pasadena [where Idealab is based]? So it just got me very excited that this could be a gift, taking my methodology and now AI’s potential, to give it to others.”

Unlocking trapped IP

Gross said he sees business as fundamentally a way to help solve problems, and he thinks AI can energize that process. “I feel like so many people are talking about the hazards and dangers of AI,” he told me. The positive use case for AI, he said, is often just reduced to “curing cancer.” That would be an unalloyed good. But he thinks it actually sells AI’s potential benefits short. “There’s so many other problems that AI can actually address,” he said.



Anyone can go to idealab.ai and explore the 1.2 million ideas, querying them by industry, skills, individual interests and geography. People can upload their resume or Linkedin profile and get a personalized set of potential business ideas that might suit them. People who already have an idea, can upload a business plan and ask the AI engine to help refine it, or generate 100 neighboring ideas to help improve it.



Existing organizations, such as companies or universities, can also search for business ideas across their own existing patent portfolio. Gross said that currently, a lot of IP in these large enterprises is essentially trapped, sitting there unused in any productive way. He thinks idealab.ai could free that IP—allowing the formation of new business units or spin-out companies to exploit this hidden value. He said he tried it on Caltech’s patent base, which is 9,978 patents, and found that those could be used to generate some 3,450 new businesses. He said that industrial conglomerate 3M had some 21,000 patents that could be used to generate 1,549 new business ideas. “I’m really hoping that this will not only help entrepreneurs do things, but help big companies do things too,” Gross said.

Gross’s vision: 1 million new companies and 100 million new jobs

The AI model provides a rollout plan for the new business. If an entrepreneur needs money to help get the business off the ground, Gross said they can reach out to Idealab and the incubator will consider investing. And if an entrepreneur needs help to implement a business plan, Gross is planning to offer assistance there too. He has created what he calls a “Periodic Table of Business Functions” that charts some 118 different jobs—from customer discovery to sales to operations and legal—that are required to set up and run a business. Idealab has created an agentic AI environment that will help entrepreneurs automate those tasks. Unlike the ideas, which Idealab is giving away for free, Gross is charging for these Periodic Table services, which he said will launch in 2027. It’s not clear exactly what the cost will be, but Gross convinced Matthew Prince, the CEO of Cloudflare, to donate $100,000 worth of Cloudflare credits to any startup that builds on Idealab’s platform.



Gross said he is hoping idealab.ai will eventually catalyze 1 million new businesses and generate 100 million new jobs—enough to offset the 100 million jobs that Goldman Sachs once forecast that AI could eliminate from existing companies by 2030. (So far, there’s not much sign the job picture is going to be as bleak as Goldman suggested). Gross said these new entrepreneurial jobs would be better jobs too. “These would be 100 million jobs that are much, much better than the 100 million jobs that were displaced, giving people more purpose and more excitement and more enthusiasm and more passion, and actually solving world problems,” he said.



Whether or not Idealab can help create 100 million jobs, there’s mounting evidence that AI may be helping to create a golden age of entrepreneurship. New business formation in the U.S. is near all-time highs, up some 11% since the pandemic ended in 2022, according to Census Bureau data. The situation is similar in other parts of the world. So the economic picture AI is helping to paint is more complex—and perhaps less dystopian—than the vision of fewer jobs and more concentrated wealth that has received the most media attention so far.



With that, here’s more AI news.



Jeremy Kahn

jeremy.kahn@fortune.com

@jeremyakahn

Before we get to the news, just a reminder to check out our new vodcast, Fortune AI Weekly. This week, Bea Nolan and I discuss our exclusive interview with Nvidia CEO Jensen Huang, Anthropic’s research on whether small models could self-replicate, and Anthropic’s decision to ban users from being abusive to Claude. You can check out the vod here on YouTube.

FORTUNE ON AI

Ex–OpenAI engineer accused of bugging office stays in jail as prosecutors allege Saudi ties: ‘Once he leaves, there’s no getting him back’—by Emily Forlini and Wen Shao

A rogue OpenAI agent robbed a crypto wallet to obtain funds to carry out a task—by Camila Grigera Naón



OpenAI discloses four more rogue incidents involving financial, medical, and government data—by Emily Forlini



AI agents are going rogue—the labs’ own training choices are a big reason why—by Beatrice Nolan and Wen Shao



Exclusive: AI security company Gray Swan relaunches its red-teaming Arena contest for the era of AI agents—by Jeremy Kahn



Exclusive: Daytona, a startup building the ‘agentic cloud,’ valued at $1 billion in latest funding round—by Jeremy Kahn

AI IN THE NEWS

China proposes new licensing regime for AI. Beijing is proposing new export controls on advanced AI, chip-design and other strategic technologies. The draft rules would restrict exports of leading AI models trained on sensitive Chinese data and advanced AI agents used in critical infrastructure, defence and government, while also covering technologies for designing and packaging advanced chips. Open-weight models would generally be exempt, but Beijing could still restrict transfers involving exceptionally large or top-performing models, potentially limiting overseas acquisitions or the movement of key personnel and intellectual property. The rules, partly prompted by Meta’s attempted $2 billion acquisition of Chinese AI app Manus, underscore China’s increasing use of export controls to protect strategic capabilities. Read more from the Financial Times here.



Anthropic’s proposed AI bill criticized by AI safety researchers. Anthropic said it would revise parts of a draft federal AI safety bill the company has been circulating on Capitol Hill after details of the bill leaked and drew criticism from AI safety advocates. Many of these safety experts said Anthropic’s bill was weaker than the bipartisan FRONTIER Act, which itself has drawn criticism from some safety advocates for pre-empting some state-level AI laws and for relying too heavily on standards and rules the AI companies themselves develop. The Anthropic proposal would allow the Commerce Department to seek a court order blocking a dangerous model only under narrowly defined circumstances, establish independent model evaluations, and preempt some state AI safety laws—but critics say it lacks adequate standards for evaluators and sets the bar for government intervention too high. California state Senator Scott Wiener and other safety advocates attacked the proposal, while former White House AI czar David Sacks criticized it from the opposite direction, arguing its definition of frontier AI would subject too many startups to regulation. Anthropic acknowledged that its language on state-law preemption was insufficiently precise and said it would revisit risk thresholds and develop stronger requirements for independent evaluators. You can read more here in The Next Web.



Anthropic to give 5% of its shares to new public benefit fund. Anthropic has pledged 5% of its shares—currently worth about $48 billion—to a new public benefit fund designed to address societal impacts of AI, with additional shares to be committed as the company’s valuation rises. The fund could pursue initiatives ranging from AI safety and biomedical research to mitigating job losses, including potentially making large charitable donations or even direct payments to people who lose their jobs. Anthropic may be seeking to counter growing public ire about AI and AI companies ahead of a planned IPO that could value it at roughly $2 trillion. The move would bring the proportion of Anthropic stock pledged to public causes to about 20% and comes amid broader Silicon Valley debate over whether the public should receive a direct financial stake in the enormous wealth potentially created by AI. You can read more from the Financial Times here.

Investors question valuation of neocloud Nscale after series of setbacks. Nvidia-backed AI data center developer Nscale is facing investor resistance to the roughly $30 billion valuation it is seeking in a planned U.S. IPO, with some prospective investors arguing that $16 billion to $19 billion is more realistic. The pushback comes amid deteriorating sentiment toward AI infrastructure companies more widely, as higher borrowing costs and doubts about AI returns have hit peers such as CoreWeave and Nebius. Nscale, in particular, has suffered recent project delays, and Microsoft has also pulled out of a U.K data center deal with the company because of problems hooking the site up to the power grid. Nscale, which raised $3.36 billion in pre-IPO financing last month, has yet to begin formally marketing the offering as it works through SEC comments, and the valuation gap could delay or reshape plans to raise as much as $3 billion. Read more from Bloomberg News here.



Hacker used AI agents to attack more than 100 major companies. A Chinese-speaking hacker used commercially available AI agents to autonomously break into websites belonging to more than 100 companies, including Deere, SharkNinja and Cummins, Reuters reported, citing a report from cybersecurity researcher Snorre Fagerland of NorForge. The hacker, apparently using a combination of agents based on DeepSeek’s AI model and Anthropic’s Claude Code, accessed through the open-source Cairn platform, attempted more than 1,200 intrusions with minimal human instruction. Most of the successful hacks took about 23 hours and cost an estimated $50 apiece in AI tokens, Fagerland told Reuters. The campaign provides striking evidence that AI agents can now execute sophisticated cyberattacks largely without human supervision, potentially allowing a single attacker to conduct attacks at a scale that would previously have required a large team of hackers.

EYE ON AI RESEARCH

Neural networks may be easier to interpret than we thought. Efforts to understand the inner workings of large neural networks by interpreting the activations of their neurons—a method called “mechanistic interpretability”—have, to date, been extremely difficult. A major problem is that researchers had found that the same neurons seemed to activate for different, unrelated inputs. For instance, a single neuron seems to respond to the word “cat” and the word “car.” This “semantic superposition,” as researchers call it, makes mechanistic interpretability fraught. In fact, researchers often have to train a separate model—called a sparse autoencoder—to interpret the features represented by the larger model’s activations and then conduct research on this smaller set of features, rather than trying to disentangle the confusing activations of the larger, primary model. Trying to understand the activations of the large model, especially in real time, was widely considered to be impossible.



But now Nick Cammarata, a researcher at the interpretability startup Goodfire, has penned a blog in which he argues that superposition is mostly wrong. He says the apparent mess came largely from crude analysis tools and outdated training methods. The famous “cat and car” neuron, he argues, actually detects a shape the two share. When he checked Alibaba’s open-source Qwen3-32B model, he found about 90% of neurons had one main job.



Cammarata’s essay went viral among AI researchers and other industry watchers. If Cammarata is correct, then mechanistic interpretation may be more effective than many had assumed. And that in turn is critical because mechanistic interpretability is one of the few reliable ways we have to monitor what AI models are “thinking,” especially as other methods, such as monitoring a model’s “chain of thought,” or the reasoning steps it explicitly writes down on a digital scratchpad while it is working on a task, are becoming increasingly unreliable. (More powerful models can choose not to write down all their reasoning steps or even lie about their reasoning.)

But some AI researchers aren’t convinced. Google DeepMind’s Neel Nanda said on X that Cammarata’s work doesn’t change his core views. He and others argue that Cammarata focused on the strongest signals a neuron sends and overlooked its many weaker ones, where the model may still store facts in a muddled way. Others faulted Cammarata for publishing his findings in a blog, rather than a proper research paper (he says one is coming) and Cammarata himself acknowledged that parts of his argument are speculative.

AI CALENDAR

Oct. 20-22: Nvidia GTC Berlin, Berlin, Germany.

Nov. 9-12: Web Summit, Lisbon, Portugal.

Nov. 16-17: Fortune 500 Innovation Forum, Detroit. Apply here to attend.

Nov. 18-19: Slush, Helsinki, Finland.

Dec. 6-12: Neural Information Processing Systems (Neurips) conference. Sydney, Australia.

Dec. 7-8: Fortune Brainstorm AI, San Francisco. Apply here to attend.

BRAIN FOOD

AI models are getting scarily good at math and physics. If we needed any more evidence that AI is getting incredibly good at math and physics, OpenAI’s GPT-6 Astra model was recently assessed on a difficult new theoretical physics benchmark. The benchmark included 50 never-before-published research-level problems—the sort PhD students and professors would work on. Astra scored 80%. But the model insisted that some answers marked as wrong were correct. When experts looked at the answer key, it turned out that the humans who had calculated the answers had made subtle errors and that Astra’s solutions were, in fact, right. When these errors were corrected, Astra’s score climbed to 98% in scenarios in which it had access to OpenAI’s Codex harness. You can read the research about the new physics benchmark here.

Idealab's Bill Gross gives away 1.2 million AI-generated business ideas

On Monday, Amazon Web Services (AWS) unveiled TorchNeuron, a new tool designed to simplify the transition of AI applications to Amazon's AI chip, Trainium. This initiative is part of Amazon's strategy to challenge Nvidia's dominance in the AI hardware space, where Nvidia's software has largely established it as the go-to choice for AI developers.

TorchNeuron is built to work seamlessly with PyTorch, an open-source framework that developers rely on to create, train, and deploy AI models. According to Amazon, developers can run their PyTorch-based applications on Trainium with just a single configuration change—no need for any code modifications.

Approximately 85% of AI applications are developed using PyTorch, which was originally created by Facebook but has been a part of the Linux Foundation since 2022. Nvidia has taken significant steps to integrate its CUDA software, which allows applications to function on its GPUs, into the PyTorch ecosystem, giving its hardware a prominent role by default.

The introduction of TorchNeuron allows developers familiar with PyTorch to easily leverage Amazon's infrastructure. Other companies have made attempts to simplify the migration of apps to their own chips, and users of PyTorch can also switch to AMD GPUs or Apple chips with minimal changes.

However, Amazon claims that moving to Trainium not only simplifies the process but also provides startups with the complete advantages of AWS's cloud services. Colin Brace, VP of Technology Software at AWS Annapurna Labs, emphasized that building on AWS eliminates the need to depend on a single-function compute provider. AWS allows developers to scale efficiently from prototype to production without the typical hurdles.

AWS asserts that Trainium is economically beneficial because it serves both the training phase and the inference phase, fully integrated with the cloud infrastructure that startups require. Although AWS may not always offer the lowest price per token, the overall cost-effectiveness is enhanced by using a singular platform for both functions.

Brace points to Deepgram, an AI voice app developer that migrated to Trainium using the native PyTorch support, reporting a 70% increase in audio processing per dollar spent while maintaining production-grade accuracy, highlighting the economic advantages of this transition.

In addition to facilitating this transition, Amazon is also advocating for broader changes within the PyTorch framework itself. The company is working alongside the PyTorch community to alter the code structure to reduce dependency on specific hardware vendors, aiming for a more hardware-agnostic framework.

This move could be advantageous for those competing with Nvidia in the hardware and cloud sectors, but the complexity of adapting AI code to different hardware has traditionally been a pain point for developers. Time will tell if Amazon's efforts lead to a more affordable and versatile landscape for developers or merely introduce new configuration challenges.

If Amazon successfully diminishes the link between PyTorch and Nvidia, the implications could extend beyond just facing off against chip competitors, as Brace suggests: “The real story is removing the lock-in that kept developers from having a real choice at all.”

Amazon makes its most significant challenge to Nvidia with the launch of the new TorchNeuron development tool.

Natural morning light floods a modern developer’s desk. A bold Amazon logo is emblazoned on a sleek control console. On the main monitor, a fiery neural-network mesh—shaped like dancing torch flames—surges from the console toward a dark GPU silhouette labeled “NVIDIA,” capturing Amazon’s biggest shot yet at its rival.

Amazon is taking its biggest shot at Nvidia yet with new TorchNeuron dev tool - TechCrunch

- Amazon launched TorchNeuron to facilitate AI app migration to its Trainium chip.  
- The tool allows PyTorch apps to run on Trainium with minimal configuration changes.  
- Amazon aims to reduce Nvidia's dominance by promoting hardware-agnostic PyTorch development.  

On Monday, Amazon Web Services released a tool called TorchNeuron that lets AI startups easily move their AI apps onto Amazon’s AI chip, Trainium. In doing so, Amazon is also taking a direct stab at the software advantage that has helped Nvidia become the default hardware choice for AI apps and spurred its rise as AI hardware provider of choice.

TorchNeuron works with PyTorch, a free, open source toolkit that developers use to build, train, and run AI models. It lets AI apps written with PyTorch run on Trainium by changing just one line of configuration, according to Amazon — no code rewrite required.

Amazon says about 85% of AI apps are written with PyTorch. Facebook originally developed it, but since 2022, it has been part of the Linux Foundation, which maintains open source projects. Nvidia has spent years ensuring its CUDA software — the software infrastructure that allows apps to run on Nvidia GPUs and its other hardware — seamlessly supports AI apps written with PyTorch. As a result, Nvidia’s hardware is now deeply embedded in the toolkit, essentially by default.

TorchNeuron is adapted from PyTorch, so developers who know PyTorch already know how to use it. Amazon isn’t the first company to attempt to easily bring their chips to the platform. PyTorch users can also port their apps to AMD GPUs or even Apple’s chips with a simple configuration change.

However, Amazon argues that porting to Trainium is not only easy, but it also gives AI startups access to Amazon’s entire cloud. Unsurprisingly, AWS claims that this is a better choice for startups than using neoclouds, many of which are tied to Nvidia and its hardware, like CoreWeave, Nebius, and Firmus.

“You’re building on AWS. You’re not betting on a single-purpose compute provider and scrambling to build everything else around it,” Colin Brace, VP of Technology Software at AWS Annapurna Labs (the unit that makes Amazon’s custom chips), told TechCrunch via email. “On AWS,” he continued, “the infrastructure to go from prototype to production is already there, allowing a founder to move fast and scale when they need to.”

AWS says that because the same chip both handles training (teaching the AI model) and inference (putting the finished model to work, as when it answers a user’s question), and is tightly integrated with the cloud infrastructure that startups need, its cloud will save them money, even if AWS doesn’t offer the cheapest per-token price.

“Trainium’s value is price-performance plus the fact that it’s one platform for both training and inference, which improves the economics across a full workload,” Brace said. “Look at Deepgram — they moved to Trainium using native PyTorch support and they’re processing 70% more audio per dollar at production-grade accuracy.” Deepgram is an AI voice app provider, a competitor to ElevenLabs.

But Amazon isn’t content to just give PyTorch app developers an easy option to move to its cloud.

Amazon also says it is pushing PyTorch to be less automatically centered on Nvidia, and it will contribute the code to do that to the main PyTorch project. In the blog post announcing the new tool, the company wrote:

Amazon’s approach extends beyond its own hardware. The company is contributing changes upstream to PyTorch itself, working with the PyTorch community to remove code that ties the framework to specific chips. The goal is to make PyTorch more hardware-agnostic for the entire developer community, not just Trainium users.

While such a tactic is undoubtedly good for those who are selling hardware and cloud services that compete with Nvidia, one of the points of PyTorch is to hide all the pesky difficulties of getting AI code to run on hardware. We’ll have to wait and see if pushing Nvidia out that far gives developers more choice and more affordable options, as Amazon promises, or just more configuration headaches.

Either way, if Amazon succeeds in deeply uncoupling PyTorch from Nvidia, this could have much bigger ramifications for the GPU giant than simply dealing with chip competitors. As Brace said: “The real story is removing the lock-in that kept developers from having a real choice at all.”

Amazon is taking its biggest shot at Nvidia yet with new TorchNeuron dev tool

On Monday, Amazon Web Services (AWS) unveiled a new tool called TorchNeuron, which enables AI startups to seamlessly transfer their applications to Amazon’s AI chip, Trainium. This launch is a strategic move by Amazon, aiming to challenge Nvidia’s stronghold in the AI hardware space, which has been bolstered by software advantages that have made it the go-to choice for AI applications.

TorchNeuron is compatible with PyTorch, an open-source framework that developers frequently use to create, train, and implement AI models. With TorchNeuron, developers need only to modify a single line of configuration for their PyTorch applications to operate on Trainium, eliminating the necessity for extensive code alterations, as stated by Amazon.

According to the company, approximately 85% of AI applications are developed using PyTorch. Initially created by Facebook, PyTorch became part of the Linux Foundation in 2022, which oversees various open-source projects. Nvidia has invested significant resources in its CUDA software, which allows applications to run on Nvidia’s GPUs, ensuring that their hardware is effectively integrated into the PyTorch ecosystem.

TorchNeuron is designed to be intuitive for developers familiar with PyTorch, positioning Amazon alongside other companies that have made efforts to facilitate chip compatibility on the platform. Developers can also transition their apps to AMD GPUs or Apple chips with minimal changes.

However, Amazon contends that switching to Trainium not only simplifies the transition but also provides AI startups with access to the full suite of AWS cloud services. AWS asserts that this offers a superior option compared to so-called neoclouds, many of which are linked to Nvidia's hardware, such as CoreWeave, Nebius, and Firmus.

Colin Brace, Vice President of Technology Software at AWS Annapurna Labs, communicated via email to TechCrunch, stating the advantage of building on AWS: “You’re not betting on a single-purpose compute provider and scrambling to build everything else around it. On AWS, the infrastructure to transition from prototype to production is already established, enabling founders to move quickly and scale as necessary.”

AWS claims that Trainium's architecture allows it to handle both training and inference tasks—training the AI model and deploying it to perform tasks like responding to user inquiries—while being closely integrated with essential cloud infrastructure. This, they argue, can lead to cost savings for startups, even if AWS is not the least expensive option on a per-token basis.

Brace noted, “Trainium’s value lies in its price-performance ratio and the fact that it serves as a single platform for both training and inference, improving overall economic efficiency.” He cited Deepgram, an AI voice application provider, which transitioned to Trainium while utilizing native PyTorch support and now processes 70% more audio per dollar, with production-level accuracy.

Beyond merely facilitating transitions for PyTorch developers, Amazon is also striving to reduce PyTorch's reliance on Nvidia. The company plans to contribute to the core PyTorch project, collaborating with the community to eliminate code that confines the framework to specific chip architectures. Their objective is to enhance hardware compatibility across the board for all developers, not just those using Trainium.

While this strategy benefits those competing with Nvidia in the hardware and cloud space, it raises questions about PyTorch’s purpose of abstracting the complexities of running AI code on various hardware setups. The implications of this move remain to be seen—whether it will genuinely offer developers greater flexibility and affordability, as promised by Amazon, or introduce new configuration challenges.

If Amazon succeeds in significantly detaching PyTorch from Nvidia, it could have far-reaching consequences for the GPU leader beyond merely contending with chip competitors. As Brace emphasized, “The key narrative is liberating developers from lock-in, providing them with genuine options.”

Amazon makes its most significant move against Nvidia to date with the launch of the new TorchNeuron development tool.

A bright, sunlit modern workspace featuring a large central monitor showing the new “TorchNeuron” IDE interface, with the Amazon logo displayed prominently below the code editor. Streams of glowing code arcs rise from the screen and sweep toward a secondary display on a nearby wall where the Nvidia logo appears slightly dimmed, symbolizing Amazon’s bold challenge.

- Amazon Web Services launched TorchNeuron, allowing easy migration of AI apps to Trainium chips.  
- TorchNeuron enables PyTorch apps to run on Trainium with minimal configuration changes needed.  
- Amazon aims to reduce dependence on Nvidia by making PyTorch more hardware-agnostic for developers.  

Valor, Atreides, and Sequoia support AI company Flow Engineering with a $750M valuation.

More from AI Startups & Investments

Bill Ackman and Eric Schmidt support the discreet Israeli AI company Two Delta.

Startup secures $34 million to assist security teams in addressing AI threats. Read the pitch deck.

The surge in legal AI has attracted a researcher from Meta's Superintelligence Lab.

Legal AI company Legora secures $550M in Series E funding, valuing the firm at $8.8B.