For many years, Silicon Valley has been perceived as the unchallenged leader of the tech industry, with other cities like London, Beijing, and Tel Aviv vying for secondary roles. However, a pivotal discussion surrounding artificial intelligence is taking place in an unexpected location: Paris.

France has made significant investments in AI research and infrastructure, with startups such as Mistral AI positioning Europe as a formidable contender in the global AI landscape. Meanwhile, the continent's startup ecosystem has seen considerable growth; founders are increasingly inclined to nurture their companies domestically rather than seeking to move to the U.S. at the earliest opportunity.

Paris’ rise in importance isn’t limited to startups alone; the city is becoming a central hub for policymakers, enterprise executives, investors, and researchers who are all engaged in crucial dialogues about the future direction of AI.

One prime example of this shift is the annual VivaTech event, which has transformed from a regional startup fair into a significant platform for AI and innovation on a global scale.

As VivaTech prepares for its 2026 edition, marking its 10th anniversary, the changes within the AI sector you'll witness are striking. Just a year ago, discussions leaned heavily on chatbots and consumer applications. Today, the focus has shifted toward vital areas such as infrastructure, cybersecurity, enterprise deployment, and the intricate task of weaving AI into larger organizations. These themes will be prominent at VivaTech 2026, where founders, investors, policymakers, and business leaders will gather to explore the future of AI in Europe and beyond.

TechCrunch’s involvement with VivaTech is indicative of the event’s growing significance within the worldwide startup landscape. Through this collaboration, TechCrunch and VivaTech will shine a spotlight on emerging entrepreneurs through the VivaTech Innovation of the Year competition, allowing the winner to pitch live in Paris and secure a spot in Startup Battlefield 200 at TechCrunch Disrupt 2026 in San Francisco.

This partnership signals a broader trend within the AI sector: Paris is no longer seen merely as an up-and-coming tech city but is establishing itself as a key player in the global AI narrative.

The ramifications of this evolution extend beyond startups, as investors, enterprise leaders, policymakers, and researchers converge to consider who will influentially shape the next chapter of artificial intelligence.

At VivaTech 2026, Paris will assert that the future of AI may not be exclusively held by Silicon Valley anymore.

Secure your pass today.

Reasons Paris could be the leading AI hub beyond Silicon Valley.

A panoramic sunrise over Paris: the Eiffel Tower and Haussmann rooftops bathed in warm natural light, their iron lattice and slate roofs subtly overlaid with glowing golden AI circuit patterns that stretch like neural pathways across the skyline. A single French flag waves atop a nearby building. In the foreground, a lone professional in a tailored suit stands on a wrought-iron balcony, holding a tablet whose code softly reflects onto the stone balustrade, gazing toward the city’s AI-powered future.

Why Paris may be the most important AI city outside Silicon Valley - TechCrunch

- Paris is emerging as a central hub for global AI discussions and innovation.  
- Europe’s startup ecosystem is maturing, focusing on domestic scaling rather than U.S. relocation.  
- VivaTech 2026 symbolizes the shift toward collaborative AI infrastructure and enterprise solutions.  

For decades, the geography of the tech industry has felt largely fixed, with Silicon Valley dominating the global startup economy. While cities like London, Beijing, and Tel Aviv have competed for secondary influence, one of the most important conversations in artificial intelligence is happening somewhere else entirely: Paris.

France has aggressively invested in artificial intelligence research and infrastructure, with startups like Mistral AI helping Europe become a legitimate force in the global AI race. At the same time, Europe’s startup ecosystem has matured significantly; its founders are increasingly willing to scale companies domestically instead of immediately looking to relocate to the U.S.

But Paris’ growing influence extends beyond startups. The city is becoming a meeting point for policymakers, enterprise leaders, investors, and researchers all trying to answer the same question: What should the next era of AI look like?

Need proof? Look no further than VivaTech, the annual European tech gathering that has evolved from a regional startup expo into one of the world’s most influential AI and innovation events.

VivaTech 2026: 10th anniversary edition

VivaTech’s evolution mirrors the AI industry’s shift over the past few years. Just one year ago, much of the conversation centered on chatbots, copilots, and consumer-facing experimentation. Now the attention is moving toward infrastructure, cybersecurity, enterprise deployment, and the messy realities of integrating AI into large organizations. That focus will be on full display at VivaTech 2026, where founders, investors, policymakers, and enterprise leaders will converge around the future of AI in Europe and beyond.

VivaTech Innovation of the Year 2026

TechCrunch’s partnership with VivaTech reflects the event’s growing influence within the global startup ecosystem. As part of the collaboration, TechCrunch and VivaTech will spotlight emerging founders through the VivaTech Innovation of the Year competition, with the winner earning a chance to pitch live in Paris and secure a place in Startup Battlefield 200 ahead of TechCrunch Disrupt 2026 in San Francisco.

The partnership also underscores something larger happening across the AI industry: Paris is no longer being viewed simply as a rising European tech hub. It is becoming one of the central gathering points for the global AI conversation.

The implications extend far beyond startups. Investors, enterprise leaders, policymakers, and researchers are all converging around the same question: Who gets to shape the next phase of artificial intelligence?

At VivaTech 2026, Paris will make the case that the answer may not belong exclusively to Silicon Valley anymore.

Book your pass now.

Why Paris may be the most important AI city outside Silicon Valley

On Friday, the Chinese startup Moonshot AI unveiled Kimi K3, touted as the world’s largest open AI model with a staggering 2.8 trillion parameters. This launch positions Kimi K3 as a formidable competitor to existing systems from industry leaders like Anthropic and OpenAI, declaring it part of a "new frontier of intelligence."

Moonshot’s Kimi K3 allows developers to download, modify, and run the AI, differentiating it from proprietary models such as OpenAI's ChatGPT and Anthropic's Claude, which keep their systems confidential. While Moonshot acknowledges that Kimi K3 currently performs below Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol, the startup asserts that it still achieves “frontier-level” performance and surpasses several earlier models, including previous versions of GPT, Claude, and SpaceXAI’s Grok. In various benchmarks focused on coding and general tasks, Kimi K3 was found to be competitively effective in comparison to Fable 5 and significantly outperformed Claude Opus 4.8, Sol, and GPT 5.5, according to the company’s assessment. Analysts from Bank of America praised Moonshot’s advancements, noting that despite a restricted access to cutting-edge chips, the startup has made significant strides in model training and design.

In the wake of this announcement, shares of Moonshot's Chinese rivals faced a sharp decline, with Zhipu and MiniMax dropping 28.4% and 15.6%, respectively, in trading on the Hong Kong exchange. Z.ai, which launched a new model in June, experienced a similar 28.4% dip.

Founded in 2023, Moonshot AI has quickly emerged as one of China’s leading AI developers with support from major tech players like Alibaba and Tencent. Recent funding raised approximately $2 billion, boosting its valuation to over $20 billion, after a significant increase from its earlier valuation of $4.3 billion following a $500 million funding round last year. Moonshot reported that its annual recurring revenue has exceeded $200 million, driven largely by subscriptions for Kimi K3 and other AI offerings.

The backdrop to this rapid development is a landscape in which Chinese AI firms face constraints due to U.S. export controls on advanced AI chips, including Nvidia's H200 series. Recent developments suggest that the U.S. has initiated limited exports of these chips, although much discretion remains with the Commerce Department, which continues to regulate approvals stringently. Accusations have arisen alleging that some Chinese companies are attempting to smuggle restricted AI chips into China, while others are navigating the restrictions by securing computing power and purchasing processors through foreign subsidiaries.

Is Kimi K3 Worth a Try? A Comparison of AI Models Against ChatGPT and Claude.

A bright, modern gallery bathed in soft natural light beneath a glass skylight. Three marble pedestals stand in a row: the left one holds a translucent crystal block engraved “Kimi K3,” the center shows bold “ChatGPT” in OpenAI-style text (no logo), and the right bears a sleek plaque reading “Claude.” Behind them, faint data-stream patterns glow on white walls, suggesting a quiet technology exposition.

Should You Try Kimi K3? Here’s How AI Model Compares With ChatGPT And Claude - Forbes

- Moonshot AI launched Kimi K3, the world's largest open AI model with 2.8 trillion parameters.  
- Kimi K3 allows modifications, unlike proprietary models, and shows competitive performance against leading models.  
- Chinese AI firms face chip access constraints, affecting their advancements in AI technology.  

Topline Chinese startup Moonshot AI on Friday launched what it claims is the world’s largest open AI model, Kimi K3, positioning its product as a direct challenger to leading systems offered by Anthropic and OpenAI in what it calls a “new frontier of intelligence.” The Chinese startup called its AI model the “new frontier of intelligence.” NurPhoto via Getty Images

Key Facts

Moonshot introduced Kimi K3 on Friday as the world’s largest open AI model at 2.8 trillion parameters—the information an AI model learns from its training data and relies on to generate responses—though neither OpenAI nor Anthropic has disclosed the total parameter counts of their models. Open-weight models like Kimi K3 or Meta’s Llama allow developers to download, run and modify the AI, unlike proprietary models such as OpenAI’s ChatGPT and Anthropic’s Claude, whose underlying systems are private. Moonshot said its model’s overall performance still trails Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol, but still demonstrates “frontier-level” performance and outperforms other tested models, including earlier GPT and Claude models as well as SpaceXAI’s Grok. In some benchmarks that included coding and general agents, Kimi K3 “performed competitively” with Fable 5 and “substantially outperformed” Claude Opus 4.8, Sol and GPT 5.5, Moonshot said. Bank of America analysts lauded Moonshot’s model in a note, writing that even though Chinese AI firms have limited access to the most advanced chips, Moonshot has demonstrated it can make major advancements by improving how it trains and designs its models.

tangent

Shares of Moonshot’s Chinese competitors Zhipu and MiniMax plunged on Friday, plummeting 28.4% and 15.6% in Hong Kong-based trading, respectively. Shares of Z.ai, which released a new model in June, similarly dropped 28.4%.

what is moonshot ai?

Moonshot AI was founded in 2023 and is among China’s leading model developers with backing from Chinese tech giants Alibaba and Tencent. The firm raised about $2 billion in its latest funding round, valuing it at more than $20 billion, according to Bloomberg. That follows Moonshot’s prior valuation at $4.3 billion from a $500 million raise at the end of last year and a $700 million raise at a $10 billion valuation earlier this year. Moonshot reported in April that its annual recurring revenue topped $200 million, thanks to growing subscriptions for Kimi and its other AI services.

key backgrond

Chinese AI firms have been constrained by U.S. export controls that limit access to the most advanced AI chips, including Nvidia’s H200 AI chips. The U.S. began allowing limited exports of Nvidia’s chips earlier this week, though the Commerce Department said approvals remain limited and are subject to licensing. The U.S. has alleged that some Chinese companies are smuggling banned AI chips into China from the U.S., and some firms have sought to work around the export restrictions by renting computing power overseas and buying processors through foreign subsidiaries.

further reading

Should You Try Kimi K3? Here’s How AI Model Compares With ChatGPT And Claude

Sable, a groundbreaking AI startup founded by four Harvard alumni less than a year ago, has successfully secured $45 million in funding from Sequoia Capital and 8VC. The company aims to develop an innovative “AI employee” designed to handle customer interactions comprehensively—from initial inquiry to final sale. Sable claims that its technology surpasses traditional chatbots by enabling capability to navigate software interfaces, showcase products, answer intricate questions, and provide real-time guidance to buyers.

The funding round was spearheaded by Sequoia Capital and 8VC, with additional contributions from BoxGroup, SV Angel, and the Valor Atreides AI Fund, along with the involvement of notable figures including founders from Suno, HubSpot, and Cognition. Shaun Maguire from Sequoia and Joe Lonsdale, co-founder of both 8VC and Palantir, will join Sable's board of directors.

Founded in October 2025, Sable was established by Nim Ravid (CEO), Leon Chen, Linda He, and Itamar Rocha, who met while studying at Harvard University and conducted research in various fields such as reinforcement learning and multimodal AI. The team comprises individuals with experience at prestigious organizations including SpaceX, Google, Meta, and Together AI.

Currently, Sable employs around 15 team members in the United States and is expanding its recruitment efforts in Israel, where Ravid sees extensive AI talent. He emphasized the unique and experienced technological skills in Israel, stating that the company has a chance to engage such talent with significant AI challenges.

Prior to launching Sable, Ravid garnered recognition both in Israel and internationally through the Survived To Tell initiative, developed in response to the Hamas attack on October 7. This initiative aimed to share the personal narratives of survivors and hostage families, reaching millions online. Additionally, Ravid served as a student representative on Harvard University’s Committee to Combat Anti-Semitism.

Sable is venturing into a market where businesses are increasingly exploring AI assistants, yet many customer interactions still heavily rely on human employees. Sable points out that the current online purchasing experience—especially for complex software and AI products—is often disjointed. Customers typically navigate various stages of the sales process, encountering different representatives for qualification calls, product demonstrations, sales discussions, and onboarding. This fragmentation leads to a lack of context, redundant conversations, and prolonged response times.

While conventional AI chatbots have automated simple functions like drafting text or summarizing content, they struggle to explain product functionalities effectively. Traditional chatbots mostly engage in text-based dialogue and are incapable of demonstrating products, navigating software interfaces, or collaborating live with users.

In response, Sable has introduced Aidan, an advanced AI system intended to function as a digital employee managing customer interactions from start to finish. Aidan stands out from conventional chatbots by being able to “see” what users are viewing on their screens, maneuver through software interfaces, articulate features, and guide customers through live demonstrations.

The technology utilizes Sable’s concept of “Interactive Intelligence,” which merges real-time browser navigation, computer vision, along with voice and video capabilities to facilitate interactions in a shared environment. Ravid described Aidan as capable of communicating directly with buyers throughout the purchasing process, operating independently of human involvement. He noted that achieving natural, low-latency, and high-accuracy interactions posed significant AI challenges that the company has addressed.

Central to Sable’s offering is LiveBox, a virtual workspace allowing Aidan to showcase products directly to prospects. The platform can tackle complex inquiries, gauge customer engagement, collect feedback, and extract insights from interactions.

Companies supply Aidan with their internal assets such as sales call recordings, documentation, marketing content, and interviews with leading employees. Sable transforms this input into a continuously evolving “Brain”—a knowledge base that enhances the AI employee’s responses and comprehension of the organization’s product positioning.

Currently, Aidan is operating in production for clients like Notion and Decagon, alongside various large public enterprises, though additional client names remain undisclosed.

Sequoia invests $45 million in Israeli entrepreneur developing the next generation of

A sunlit R&D lab bathed in warm, natural light, where a focused Israeli entrepreneur leans over a sleek, futuristic circuit board prototype. A bold Sequoia Capital red tree logo dominates the back wall, while an Israeli flag hangs beside it. On the workbench lies a stylized “$45M” grant document, and scattered next-generation hardware components catch the daylight streaming through tall windows.

Sequoia leads $45 million bet on Israeli entrepreneur building the next generation of AI employees - calcalistech.com

- Sable, an AI startup, raised $45 million to create a comprehensive AI employee for customer interactions.  
- The AI system, named Aidan, can navigate software and provide live demonstrations to users.  
- Sable’s technology enhances online sales processes, improving customer engagement and reducing context loss.

Sable, an AI startup founded by four Harvard graduates less than a year ago, has raised $45 million from Sequoia Capital and 8VC to develop what it calls an “AI employee” capable of managing customer interactions from start to finish. The company says its technology can navigate software interfaces, demonstrate products, answer complex questions and guide buyers in real time, moving beyond the limitations of traditional chatbots.

The round was led by Sequoia Capital and 8VC, with participation from BoxGroup, SV Angel, Valor Atreides AI Fund, Sabrina and Evan Hahn, as well as investors including founders from Suno, HubSpot and Cognition. Sequoia partner Shaun Maguire and 8VC co-founder and Palantir co-founder Joe Lonsdale are joining Sable’s board of directors.

1 View gallery Nimrod Ravid ( Photo: Boris Zharkov )

Founded in October 2025 by Nim Ravid (CEO), Leon Chen, Linda He and Itamar Rocha, Sable was created by a team that met while studying at Harvard University and previously conducted research in areas including reinforcement learning, post-training and multimodal AI. Members of the team have backgrounds at companies including SpaceX, Google, Meta and Together AI.

The company currently employs around 15 people, all based in the United States, but is beginning to recruit engineers in Israel, where Ravid says the company sees a deep pool of AI talent.

“Israel has leading, experienced and unique technological talent, and we have an opportunity to expose it to the great challenges of the AI field,” Ravid said. “We are working to recruit the best engineers in Israel.”

Before founding Sable, Ravid became known in Israel and abroad for founding the Survived To Tell initiative following the October 7 Hamas attack, with the goal of bringing the personal stories of survivors and families of hostages to an international audience. The initiative reached tens of millions of views, and Ravid later served as the student representative on Harvard University’s Committee to Combat Anti-Semitism.

Sable is entering a market where companies are increasingly experimenting with AI assistants, but where many customer-facing interactions remain dependent on human employees.

The company argues that existing online purchasing experiences, particularly for complex software and AI products, remain fragmented. Customers often move between different stages of the sales process, from initial qualification calls to product demonstrations, sales discussions and onboarding, with different employees handling each step.

The result, according to Sable, is a loss of context, repeated conversations and delays in getting answers.

Traditional AI chatbots have helped automate simple tasks such as drafting text or summarizing information, but they are limited when customers need to understand how a product actually works. They typically operate through text-based conversations and cannot demonstrate products, navigate interfaces or collaborate with users in real time.

Sable’s solution is Aidan, an AI system designed to act as a digital employee that can manage customer interactions end-to-end.

Unlike a chatbot that waits for prompts, Aidan can “see” what appears on a user’s screen, navigate software interfaces, explain features and guide customers through live demonstrations.

The technology is based on what Sable calls “Interactive Intelligence”, a combination of real-time browser navigation, computer vision, voice and video capabilities that allow the AI system to interact with users in a shared environment.

“Aidan interfaces directly with buyers on your behalf, end-to-end, without a human in the loop,” Ravid said. “Achieving the low-latency and high-accuracy interaction required for a natural conversation with an AI employee meant overcoming a set of frontier AI challenges.”

A key component of Sable’s platform is LiveBox, a virtual workspace that allows Aidan to demonstrate products directly to potential customers.

The company says the system can answer complex questions, measure engagement, collect feedback and surface insights from customer interactions.

Companies provide Aidan with their internal knowledge, including sales call recordings, documentation, marketing materials and interviews with top performers. Sable converts this information into what it calls a “Brain”, a continuously updated knowledge base that allows the AI employee to improve its responses and understand how an organization positions its products.

The company says Aidan is already being used in production by companies including Notion and Decagon, as well as large public companies, although it did not disclose the names of those additional customers.

Sequoia leads $45 million bet on Israeli entrepreneur building the next generation of

Parag Agrawal, the former chief executive of Twitter, has been intensifying his collaboration with the leading search engine giant through his AI search venture, Parallel Web Systems. Established over two and a half years ago, Parallel is designed to enhance the way AI agents navigate the internet. In a significant development, Google is set to integrate Parallel into its offerings for cloud clients who are building AI models using the Gemini framework, thereby granting Parallel access to Google's extensive customer base.

The startup has attracted investment from prominent firms such as Sequoia Capital, Khosla Ventures, and Kleiner Perkins, amassing a total of $230 million in funding, with its latest valuation reaching $2 billion as of April.

Agrawal's motivation for founding Parallel stemmed from his conviction that AI agents will eventually surpass human capabilities in web searching, necessitating a new approach to information retrieval. Unlike conventional search engines that prioritize user-friendly web pages, Parallel specializes in extracting data for AI systems, including information that is often hidden within extensive documents.

As AI agents increasingly take on various tasks, ensuring they have access to the most current information has become vital. This concept, known as grounding, is a challenge that Parallel, along with other emerging startups like Exa, aims to tackle.

According to Matt Renner, president and chief revenue officer of Google Cloud, the partnership with Parallel aligns with Google's objective of providing customers with diverse options. While Google possesses its own grounding solutions, organizations may seek out different tools based on their specific requirements.

In addition to its collaboration with Google, Parallel can also be utilized across other cloud platforms, including Amazon Web Services. Agrawal emphasizes that the relationship with Google extends well beyond mere distribution, representing their most extensive technical integration to date with a leading cloud provider.

Since its inception, Parallel has largely built its technology on Google Cloud. Recently, engineers from both Parallel and Google have been working closely to unify their respective toolsets, enabling users to leverage them without needing to establish those interconnections themselves. Major clients, such as the legal AI firm Harvey, have already begun utilizing Parallel's search capabilities to enhance their AI models, merging real-time web data with proprietary internal information.

Google's collaboration with former Twitter CEO Parag Agrawal's $2 billion AI search company explained.

A bright, sunlit corporate boardroom with floor-to-ceiling windows overlooking a city skyline. Two business executives—one wearing a lapel pin shaped like the Google “G”—firmly shake hands across a sleek glass table. Behind them, a large presentation screen displays the Google logo in bold, and next to it a modern graphic reading “$2 B AI Search Startup” with a stylized search-bar icon. A laptop perched on the table shows “Agrawal AI Search” on its screen. Warm natural light floods the room, creating an optimistic, collaborative atmosphere.

Why Google partnered with former Twitter CEO Parag Agrawal's $2 billion AI search startup - Business Insider

- Parag Agrawal's startup Parallel Web Systems partners with Google for AI search solutions.  
- Parallel has raised $230 million and is valued at $2 billion as of April.  
- The startup helps AI agents access deep web information, improving their grounding capabilities.  

Former Twitter CEO Parag Agrawal's AI search startup, Parallel Web Systems, is deepening its partnership with the world's largest search company.

Agrawal founded Parallel over two and a half years ago to help AI agents search the web. Now, Google will offer Parallel to cloud customers who are building AI agents with the Gemini model, giving Parallel access to Google's vast clientele.

Investors like Sequoia Capital, Khosla Ventures, and Kleiner Perkins back Parallel. It has raised $230 million to date, and last reported a valuation of $2 billion in April.

Agrawal said he started Parallel after becoming obsessed with the idea that AI agents would eventually search the web exponentially more than humans — and would need a different way to do so.

While traditional search engines prioritize webpages that are easy for people to read, Parallel retrieves information for AI models, including material buried deep within documents.

Related video Why Google partnered with former Twitter CEO Parag Agrawal's $2 billion AI search startup

As AI agents take on more work, equipping them with up-to-date information has become crucial. That process, known as grounding, is what Parallel and other startups like Exa are trying to solve.

Google Cloud president and chief revenue officer Matt Renner. Google

Google Cloud president and chief revenue officer Matt Renner told Business Insider the Parallel partnership reflects its strategy of giving customers choice. While Google has its own grounding tools, businesses may prefer different products depending on the use case.

Parallel is available on other clouds beyond Google, including Amazon Web Services. Agrawal said the Google relationship goes beyond distribution, marking "our deepest technical integration with a hyperscaler model lab to date."

Parallel has built much of its product on Google Cloud since day one. And in recent months, engineers from both companies have worked to bring together tools from both companies so customers don't have to build those connections themselves, Agrawal said.

Parallel has landed major customers like the legal AI startup Harvey, which uses Parallel's search tools to give its AI models access to current web information alongside customers' internal data.

Why Google partnered with former Twitter CEO Parag Agrawal's $2 billion AI search startup

Applied Computing, a startup based in London, is making strides in the oil, gas, and petrochemical sector by developing a foundational AI model aimed at improving data management in these industries. The company has successfully secured $20 million in Series A funding, led by renowned engineering firm KBR, with participation from Databricks Ventures.

Established in 2023, Applied Computing focuses on the complex data landscape of oil, gas, refining, and petrochemical operations, where a single facility may incorporate thousands of sensors that monitor variables such as temperature, pressure, velocity, and viscosity. The vast potential for enhancing data tracking in the energy sector faces challenges due to significant fragmentation.

As a result, operators often rely on less than 8% of the data at their disposal when making operational decisions, according to Callum Adamson, co-founder and CEO of Applied Computing. He explains that while the necessary data is often collected, facilities struggle to effectively integrate sensor data, engineering documentation, and the principles of physics and chemistry in a timely manner for predictive analytics.

Adamson emphasizes the importance of real-time communication among these data sources, stating that this integration is crucial for effective decision-making.

Rather than relying on conventional large language models that predict the next word, Applied Computing's model, named Orbital, employs a combination of a time series model, a physics-based model, and a language model to forecast a facility's operational state. Orbital synthesizes sensor data while considering physical and chemical constraints of the equipment and the activities of operators. This model also enables technicians to simulate the effects of modifications in one part of a facility on its entire operation.

In essence, Applied Computing is delivering rapid solutions; Orbital is touted to identify anomalies, analyze the causes, and simulate the impact of adjustments within minutes. Adamson notes that investigations, which would typically take days or weeks, can now be condensed to mere seconds, thereby helping operators minimize energy consumption and sustain production levels.

This value proposition of speed has resonated well with the market. According to the company, it has transitioned from a stealth mode to achieving double-digit millions in annual recurring revenue in less than 18 months. While Adamson mentions that Orbital is already in use by several well-established upstream oil and gas, along with downstream refining and petrochemical firms, he refrains from disclosing specific customer numbers.

Among the company's collaborators is Wipro, an Indian energy firm, and KBR, which has incorporated Orbital into its INSITE 3.0 digital platform for energy initiatives, notably for ammonia production. Moreover, Applied Computing is planning to announce a partnership with a significant European oil company and is currently engaging with a major U.S. upstream operator.

Nonetheless, Applied Computing is entering a competitive environment filled with established industrial software providers and niche AI startups. Competitors like AspenTech offer simulation and AI modeling tools while AVEVA provides solutions for physics-based process simulation and “what-if” analysis for industrial facilities. Other companies such as Cognite and Seeq focus on the data aspect, helping facilities analyze industrial data and implement AI-driven workflows.

Adamson contends that the company's distinct advantage lies not in accessing industrial data or process knowledge but rather in its ability to assemble top-tier AI researchers to develop a model that competes effectively with Orbital.

“It’s an AI challenge, not merely a data or energy issue,” he states. “If you're a leading AI researcher, companies like Shell might not be at the top of your list for where to work.”

He also highlights that Orbital benefits from the operational data it accesses through its deployments, noting that data from refinery operations is generally not publicly available and that simulated data does not accurately depict the complexities of functioning plants.

Additionally, the partnership with KBR is expected to enhance Applied Computing's capabilities by providing operational data, industry insights, and openings to potential new clients.

The startup intends to utilize the $20 million to further its international expansion, recruit for research and development roles, and explore additional deployments with energy industry clients. Recently, Applied Computing announced the establishment of a new office in Houston, adding to its headquarters in London and its operational hub in Bengaluru. Adamson indicates that this U.S. location will position the company close to two existing clients in North America, with plans for an expansion into the Middle East also underway.

Applied Computing aims to provide oil and gas operators with a comprehensive AI model for their facilities.

A sprawling oil refinery at dawn, bathed in warm natural sunlight that glints off towering distillation columns and winding pipelines. In the foreground, an adult engineer’s silhouette stands before a large, wall-sized touchscreen displaying a vibrant digital twin of the entire plant—flowing process streams rendered in soft greens and oranges. The Applied Computing logo is prominently displayed atop the screen, symbolizing their AI model orchestrating every unit of the refinery.

Applied Computing wants to give oil and gas operators an AI model for the entire plant - TechCrunch

- Applied Computing, a London startup, raised $20 million for AI in the energy sector.  
- Their AI model, Orbital, integrates data from sensors, physics, and chemistry for analysis.  
- The funding will aid international expansion and hiring while enhancing energy sector deployments.  

Applied Computing, a London-based startup that’s building a foundation AI model for the oil, gas and petrochemical industry, has raised a $20 million Series A led by engineering giant KBR, with Databricks Ventures participating.

Founded in 2023, the startup targets oil, gas, refining and petrochemical systems, where a single facility can have thousands of sensors measuring everything from temperature and pressure to velocity and viscosity. While there’s a huge market for helping energy companies solve the data tracking problem, the fragmentation that presents a significant hurdle.



Facilities consequently make operating decisions using less than 8% of the data available to them, says Applied Computing’s co-founder and CEO Callum Adamson (pictured above, right). Operators already collect much of this information, he said, but they struggle to combine the sensor readings, engineering documentation, and physics and chemistry quickly enough to analyze and make predictions.



“It’s getting those three data sources to talk to each other in real time. That’s the real key,” he told TechCrunch.



Unlike large language models, which predict the next word, Applied Computing says its foundation model, Orbital, combines a time series model, a physics-based model, and a language model to predict the state of a facility. It does this by analyzing sensor readings, keeping physics and chemistry in mind, and recognizing a facility’s equipment constraints and operator activity. It also allows technicians to run simulations of how a change in one part of a facility could affect the rest of its operations.

Image Credits:Applied Computing / Applied Computing

Essentially, Applied Computing is pitching speed: It claims Orbital can flag anomalies, investigate what caused them, and model whether a proposed fix could create problems elsewhere in the facility, all within minutes. Adamson claims the product can compress investigations that previously took days or weeks into seconds, helping operators reduce energy use and maintain output.

That promise of speed seems to have found believers. The startup says it has gone from stealth to double-digit millions in annual recurring revenue in under 18 months. Adamson said Orbital is in use at some “large, publicly listed” upstream oil and gas, downstream refining and petrochemicals companies, although he declined to mention how many customers it has.

Its partners include Indian energy company Wipro, and KBR, which has integrated Orbital into its INSITE 3.0 digital platform for energy projects, and is using the product for ammonia production. Adamson said the startup is also working with a “major U.S. upstream operator,” and plans to announce a partnership with a European oil major in the coming weeks.

Still, Applied Computing is entering a market that has entrenched industrial software suppliers as well as more focused AI startups. AspenTech sells simulation and AI-powered modeling software for upstream, refining and chemical operations, while AVEVA offers physics-based process simulation, optimization, and “what-if” modeling for industrial plants. Cognite and Seeq target the data layer, helping facilities analyze industrial data, and apply AI to design workflows.

Adamson argues that the company’s moat is not access to industrial data or process knowledge, but assembling AI researchers to build a model that can compete with Orbital.



“It’s an AI problem. It’s not a data problem, and it’s not an energy problem,” he said. “If you’re a tier-one AI researcher, where are you going to work? … I don’t think Shell’s on that list.”

Adamson also pointed to the data Orbital receives through its deployments. Operational data from refineries and other energy facilities is generally not available publicly, he said, while simulated data cannot fully reproduce what happens inside a working plant.

The KBR partnership may help the company, too. Adamson said the partnership gives Applied Computing access to operational data, industry expertise, and also introductions to more potential customers.

Applied Computing plans to use the $20 million to expand internationally, hire for research and engineering roles, and explore deployments with energy clients.

The company on Thursday said it’s also opened an office in Houston, adding to its headquarters in London and operational hub in Bengaluru. Adamson said the U.S. base puts the startup closer to two existing customers in North America, and an expansion into the Middle East is also in the works.

Applied Computing wants to give oil and gas operators an AI model for the entire plant

Mira Murati, the former technology leader at OpenAI, is pioneering the development of more adaptable artificial intelligence models through her company, Thinking Machines Lab. This initiative aims to narrow the gap that established frontier labs, including OpenAI and Anthropic, currently maintain in the AI landscape.

On Wednesday, Thinking Machines Lab unveiled its inaugural AI model, named Inkling, which features "open weights." This means that users can personalize the model using their own data. Inkling has 975 billion total parameters, placing it significantly smaller than the highly advanced, proprietary models offered by competitors like OpenAI and Anthropic.

The company stated, “We trained it to be a broad, balanced foundation model: robust across various domains and flexible enough to adapt. While Inkling may not be the strongest overall model available, whether open or closed, it is designed for versatility.”

Thinking Machines Lab's foray into the decentralized realm of open-weights AI models coincides with a growing criticism within the industry regarding the “walled garden” strategies that major labs like OpenAI and Anthropic employ. Executives such as Palantir's Alex Karp and Microsoft's Satya Nadella have raised concerns that businesses could jeopardize their own models by entrusting their essential data to centralized, generalized AI systems that they don’t govern.

This development is also part of a larger trend in Silicon Valley to foster the creation of domestic open-weight models as a countermeasure to offerings from China's Alibaba and emerging startups like Z.ai. A number of U.S. firms have recently turned to Chinese open-weight models to tackle less complex AI tasks, striving to manage costs and enhance their strategic options.

Mira Murati's AI Company Launches Initial Model to Challenge Dominance of Major AI Firms

A bright, sunlit startup lab: Mira Murati, mid-30s, stands beside a sleek pedestal holding a softly glowing crystalline AI core labeled “Model v1.” Through large windows behind her, warm morning light spills in, illuminating her and a small team of engineers applauding. In the distance, subtle silhouettes of towering corporate skyscrapers stamped “AI Giants” loom on the horizon, their shadows receding before the startup’s newfound light.

Mira Murati’s AI Startup Releases First Model in Bid to Loosen AI Giants’ Grip - WSJ

- Mira Murati introduces Inkling, an AI model with customizable "open weights" for users.  
- The model has 975 billion parameters, smaller than leading closed-source competitors' models.  
- Industry leaders advocate for open-weights models to prevent reliance on centralized AI solutions.  

Mira Murati, CEO of Thinking Machines Lab, in March. David Paul Morris / Bloomberg News

Former OpenAI technology chief Mira Murati is betting on more customizable artificial-intelligence models to chip away at the lead that frontier labs such as her former employer hold over the technology.

Thinking Machines Lab, the company led by Murati, released its first AI model on Wednesday—and did it with “open weights,” meaning others can modify it with their data. Called Inkling, the model has 975 billion total parameters, making it far smaller than estimates of the most advanced closed-source models from rivals such as OpenAI and Anthropic.

“We trained it to be a broad, balanced foundation model: strong across many domains, flexible enough to adapt. Inkling is not the strongest overall model available today, open or closed,” the company said.

Thinking Machines’s push into the decentralized ecosystem of open-weights AI models comes amid a broader industry backlash against the “walled garden” approach of frontier labs such as OpenAI and Anthropic.

Industry leaders such as Palantir Chief Executive Alex Karp and Microsoft’s Satya Nadella have warned that companies risk undermining their own business models by feeding their core institutional data into centralized, generalist models they don’t control.

The release is also part of a push within Silicon Valley to build homegrown open-weights models as an alternative to those developed by China’s Alibaba and a younger crop of startups such as Z.ai. Many U.S. companies have been turning to Chinese open-weight models to help complete less-sophisticated AI tasks, in an effort to control costs and diversify their approach.

Mira Murati’s AI Startup Releases First Model in Bid to Loosen AI Giants’ Grip

Emergent, an Indian startup specializing in AI coding solutions, has successfully secured $130 million in its Series C funding round, achieving an impressive post-money valuation of $1.5 billion, marking a drastic five-fold increase in just six months.

The funding was spearheaded by private equity firm Creaegis, with participation from new investors like MNI Ventures-Claypond and Sentinel Global, as well as existing backers including Khosla Ventures, SoftBank’s Vision Fund 2, Lightspeed, and Y Combinator. With this round, Emergent's total funding has reached $230 million, following a previous Series B financing of $70 million at a $300 million valuation in January.

The AI coding sector has garnered significant attention from investors, with companies like Lovable, Replit, and Cursor collectively attracting billions to enhance tools that streamline developers' workflows. AI powerhouses such as OpenAI and Anthropic have also intensified their focus on coding capabilities.

Emergent aims to carve out its space in this competitive landscape by catering to entrepreneurs launching startups and small to medium enterprises that have historically relied on conventional tools like email, spreadsheets, and messaging platforms for operations.

In a conversation with TechCrunch, Emergent's co-founder and CEO Mukund Jha shared, “Our vision has consistently been to create a production-grade application tailored for serious builders. Essentially, we’re offering an engineering team in a box.”

Jha noted that Emergent has achieved an annual run-rate revenue of $120 million, reflecting a robust 70% increase over the past four months, along with a customer base exceeding 200,000. He co-founded the company with his brother, Madhav Jha, who serves as CTO, in June of the previous year.

The startup's clientele includes diverse sectors like trucking, where companies are developing software to monitor shipments; factories; constructions firms implementing enterprise resource planning systems; and property managers working on internal customer management solutions.

Emergent's revenue distribution shows that approximately one-third comes from North American customers, another third from Europe, with the remainder sourced from various other regions, including about 8% to 9% from India, as Jha explained to TechCrunch.

Focusing on small businesses and entrepreneurs puts Emergent in direct competition with Replit, which Jha identifies as its primary rival. He seeks to differentiate Emergent from other developer-centric coding platforms such as Anthropic’s Claude Code, OpenAI’s Codex, and Cursor by emphasizing the need for a comprehensive platform that simplifies deployment, hosting, testing, and debugging for non-technical users.

However, Jha admits that design is an area that requires improvement, as many AI-generated websites often share a similar aesthetic.

Plans for the newly acquired funding include accelerating product development and research initiatives, enhancing the success rates of applications constructed through Emergent’s platform, and refining the core workflows of its AI agents. The company is also exploring the incorporation of more advanced AI applications, including local and open-source models, and aims to bolster its market presence.

Additionally, Emergent is contemplating the establishment of an office in Europe due to notable customer engagement in that region.

Currently, the company employs around 200 individuals, predominantly based in Bengaluru, with a smaller team in San Francisco. Jha further indicated plans to increase the San Francisco office staff by 30 to 40 employees before the year's end.

Indian AI coding startup Emergent reaches unicorn status after securing $130M in Series C funding.

A sunlit, modern Indian startup office with floor-to-ceiling windows. In the center, a gleaming golden unicorn statue stands on a sleek podium engraved “$130M Series C.” The Emergent logo is boldly displayed on a feature wall, and an Indian flag hangs nearby. A group of adult developers in business-casual attire share quiet smiles, laptops open with lines of code glowing softly.

Indian AI coding startup Emergent becomes a unicorn with $130M Series C - TechCrunch

- Emergent raised $130 million in Series C funding, achieving a $1.5 billion valuation.  
- The startup targets small businesses and entrepreneurs, facing competition from Replit.  
- Emergent plans to enhance product development and expand its presence in Europe.  

Indian AI coding startup Emergent has raised $130 million in a Series C funding round at a $1.5 billion post-money valuation, a five-fold jump in six months.

The funding round was led by private equity firm Creaegis. New investors MNI Ventures-Claypond, Sentinel Global, and existing backers Khosla Ventures, SoftBank’s Vision Fund 2, Lightspeed, and Y Combinator also participated. The deal takes Emergent’s total funding to $230 million. The startup had previously raised a $70 million Series B at a $300 million valuation in January.

AI coding has attracted hordes of investors, with startups such as Lovable, Replit, and Cursor raising billions in funding to develop tools that allow developers to speed up their work. AI labs such as OpenAI and Anthropic have also pushed deeper into coding.

Emergent is looking to gain a share of this crowded market by targeting entrepreneurs looking to start new businesses and small and medium-sized companies that have traditionally relied on email, spreadsheets, and messaging apps to run their operations.

“Our thesis has always been to build a production-grade application for serious builders,” Emergent co-founder and chief executive Mukund Jha (pictured above, right) told TechCrunch in an interview. “So you’re basically getting an engineering team in a box.”

Jha said the startup has reached an annual run-rate revenue of $120 million, up 70% in the last four months, and has more than 200,000 paying customers. Jha started Emergent with his brother Madhav Jha (CTO) in June last year.

Customers include trucking companies building software to track shipments; factories; construction businesses creating enterprise resource planning systems; and property managers developing internal customer management tools.

North American customers account for about a third of Emergent’s revenue, Europe makes up another third, and the rest comes from other markets, Jha told TechCrunch. India accounts for about 8% to 9%.

Emergent’s focus on small businesses and entrepreneurs pits it directly against Replit, which Jha described as the startup’s closest rival. He sought to distinguish Emergent from developer-focused coding tools such as Anthropic’s Claude Code, OpenAI’s Codex, and Cursor, arguing that non-technical users need a platform that handles deployment, hosting, testing, and debugging alongside the work of programming.

However, Jha acknowledged that design remains a weakness, pointing out that many websites built using AI tools tend to look similar.

Emergent plans to use the fresh capital to accelerate product development and research, including improving the success rate of applications built on its platform and its core AI agent workflows. The company is working to support more complex AI applications, including those that use local and open source models, Jha said, adding that it will also invest in expanding its go-to-market operations.

The company is also considering opening an office in Europe, where Jha said Emergent is seeing significant customer traction.

Emergent has about 200 employees, most of whom work in Bengaluru, with a handful in San Francisco. The startup plans to expand its San Francisco office by 30 to 40 people by the end of the year, Jha said.

Indian AI coding startup Emergent becomes a unicorn with $130M Series C

An innovative AI advertising initiative is making waves and may stir some controversy within the film industry.

OpenArt AI, a text-to-video platform, is launching a promotional campaign for its "Director" product in major AMC theaters across Los Angeles, San Francisco, Chicago, and New York this week, as confirmed exclusively to CMO Insider. This effort includes billboards in these urban centers as well as digital and social media placements.

The Director tool allows users to articulate ideas, aesthetic preferences, and storylines in everyday language, generating videos that can be up to five minutes long.

The objective of this campaign, entirely crafted in-house with the Director tool, is to ignite a trend of "vibe directing" similar to the emergence of "vibe coding," while encouraging creative newcomers to produce short dramas, music videos, or advertisements.

According to Stella Guan, OpenArt AI's head of growth and operations, the decision to target moviegoers wasn’t meant to provoke, but rather to engage an audience seen as aligned with the product's capabilities.

AI technologies have faced scrutiny in Hollywood, especially with concerns surrounding job loss and the danger of low-quality content termed "AI slop." Nevertheless, some industry figures—including notable names like Ben Affleck and Martin Scorsese—are beginning to recognize the potential of AI for enhancing productivity and technical aspects of filmmaking.

Since its establishment in 2022 by two ex-Googlers, OpenArt AI has seen its user base swell to 8 million monthly active users. The company also attracted attention earlier this year with a $30 million Series A funding round led by Canaan Partners.

The company’s in-house creative team, comprising six directors, spent four days brainstorming concepts for the advertisement using the Director product, culminating in a "movie night" screening session to finalize the chosen spot.

The advertisement features a man enjoying a hot dog on a park bench adjacent to a basketball court, where he is approached by a character described as a "coach," brandishing a whistle and a laptop. The man expresses his long-standing desire to make a film about a penguin in the desert, prompting the coach to input the idea into OpenArt, promoting the notion that any video can be created, including "micro dramas."

Despite the excitement around AI in advertising, there are notable consumer hesitations. A survey from January by IAB and Sonata Insights revealed that 30% of Gen Z individuals found brands using AI in advertisements to be "inauthentic," while 26% said they felt such ads were "disconnected," and 24% described them as "unethical."

James Poulter, CEO of the consulting firm ThreePoint Labs, advises caution, suggesting that AI companies should avoid overstating what their platforms can accomplish for novice users.

Poulter remarked, "All of the models are overselling how easy it is to create the top 1% of excellence. Even if the models can generate cinematic content, without the foundational language of cinema or directing, users may struggle to guide the model effectively."

Guan acknowledges that much of the skepticism regarding AI in creative endeavors stems from a lack of awareness about the technological advances made since its earlier versions.

She envisions that this advertising campaign will demonstrate the elevated quality of AI-generated content, asserting, "This alone serves as a great illustration of our product's capabilities and the technology behind it."

An AI startup aims to make 'vibe directing' the new trend in vibe coding.

A bright modern creative studio bathed in warm natural light through floor-to-ceiling windows. A poised adult “vibe director” stands at a sleek wooden console, her hands gliding over touch-sensitive sliders. Above her, soft pastel streams of light and color swirl in the air like living brushstrokes. Behind her, a widescreen display shows a clean interface of interconnected mood nodes labeled “Energy,” “Tempo,” and “Color.” A potted plant and a steaming coffee cup on the console add a human touch, suggesting this startup is pioneering “vibe directing” as the next evolution of coding.

An AI startup wants 'vibe directing' to become the new vibe coding - Business Insider

- OpenArt AI launches a targeted ad campaign in AMC theaters across major U.S. cities.  
- The Director tool generates up to five-minute videos from user-described ideas in natural language.  
- AI-generated ads face skepticism, with concerns about authenticity and potential job loss in Hollywood.  

Here comes an AI ad campaign that could ruffle some feathers in Hollywood.

AI text-to-video platform OpenArt AI is launching ads in AMC movie theaters across Los Angeles, San Francisco, Chicago, and New York this week to promote its Director product, the company exclusively told CMO Insider. The push also includes billboards in those cities, as well as digital and social placements.

Director lets users describe an idea, visual style, and narrative arc in conversational language, then generates videos up to five minutes long.

The company said it hopes the campaign — created entirely in-house using the Director tool — will spark a wave of "vibe directing," echoing the rise of "vibe coding," and inspire newcomers to create micro dramas, music videos, or ads.

"I don't think we're trying to be provocative, but it's just that we think the audience is a really good fit," Stella Guan, head of growth and operations at OpenArt AI, told me when I asked whether targeting the campaign to cinemagoers was designed to poke the hornet's nest.

AI has provoked a backlash within Hollywood's creative community over the technology's potential to destroy jobs and the proliferation of low-quality "AI slop." A growing segment of the industry — including Ben Affleck and Martin Scorsese — has become more open to embracing it to streamline and enhance their technical work, though.

Founded in 2022 by two former Googlers, OpenArt AI says it has grown to 8 million monthly active users. In January, the company raised a $30 million Series A round led by Canaan Partners.

The company's in-house studio of six creative directors spent four days coming up with ideas for the ad using the Director product. The company held a "movie night" screening marathon before selecting the final spot.

The ad depicts a man eating a hot dog on a bench beside a basketball court. He's then approached by a "coach" character who blows a whistle and wields a laptop. The man tells the coach he has always wanted to make a movie about a penguin in the desert.

The coach types it into OpenArt, explaining he can create any video he wants, "even a micro drama."

The man protests, "I don't watch micro dramas."

OpenArt AI said its budget for the Director marketing campaign is in the "low hundreds of thousands" of dollars.

Countering AI criticism

Some consumers have an aversion to AI-generated ads. A survey published in January by the IAB and Sonata Insights found 30% of Gen Z respondents said brands using AI in ads were "inauthentic," while 26% said they were "disconnected," and 24% "unethical."

Moral panic about AI aside, James Poulter, CEO of the AI consultancy ThreePoint Labs, said AI companies need to be cautious about overselling their platforms' capabilities to everyday users.

"All of the models are overselling how easy it is to create the top 1% of excellent," Poulter said. "No matter how capable the models are of creating cinematic things out of the box, if you don't have the language of cinema or being a director, you cannot steer the model to deliver something akin to someone who does have that language."

Guan said a lot of the criticism about the use of AI in advertising and other creative works could be attributed to people not being aware of how far the technology has progressed since its early iterations.

She hopes the ad will show that AI is producing higher-quality output fit for the big cinema screen.

"I think that by itself is a great showcase of our product's capability as well as of the technology," Guan said.

An AI startup wants 'vibe directing' to become the new vibe coding

In the competitive landscape of artificial intelligence, cost considerations are prompting American startups to explore alternatives beyond the high-priced AI models offered by leading U.S. companies like Anthropic. One notable case is Lindy.ai, a San Francisco-based enterprise founded by Flo Crivello that develops AI-powered "assistants" for email and calendar management. Initially, Lindy relied on Anthropic’s premium models, but Crivello soon realized that the expense was unsustainable, eclipsing even payroll for his team of over two dozen employees.

In response, Lindy.ai made a strategic shift last month by transitioning fully to the Chinese AI model DeepSeek-V4, which offered a staggering cost reduction, saving the company millions. Crivello described this switch as an obvious financial decision in light of the rising costs associated with AI development.

The escalating expenses tied to AI technologies are becoming a significant concern for many U.S. businesses. As firms grapple with the balance between leveraging AI for operational efficiency and managing escalating costs, a growing number are turning to cost-effective Chinese models. While U.S. companies like Anthropic, OpenAI, and Google are currently at the forefront of AI innovation, experts note that Chinese models lag behind by approximately six to twelve months in performance capabilities. However, China has made significant strides in the open-source AI space, enabling companies to access and adapt these models freely.

Crivello pointed out that many entrepreneurs in the AI field are either contemplating a switch to Chinese models or have already made the transition. The need for cost efficiency is echoed by Uber's CEO, Dara Khosrowshahi, who shared concerns on a podcast about exceeding their AI budget in just one quarter, compelling the company to reevaluate its AI spending.

While some key industry players, such as Airbnb, have found success using Chinese models like Alibaba's Qwen, many brands are hesitant to openly discuss their use due to political implications. Nonetheless, structures facilitating easier access to these models, including platforms like Hugging Face and GitHub, have made them readily available for companies looking for alternatives.

For firms like Featherless, which provides access to approximately 30,000 AI models, the growing popularity of Chinese offerings demonstrates a shifting preference among developers. CEO Eugene Cheah remarked on the practical advantages of these models, drawing an analogy between high-end luxury cars and more practical vehicle choices. He noted that many open-source AI groups are content to utilize models that may not be top-tier but can still deliver effective performance at scale.

As the adoption of Chinese AI models grows—evident in platforms like OpenRouter where the use of DeepSeek has doubled—companies are finding a balance between performance and cost. Some businesses opt to self-host these models, while others leverage paid hosting solutions that maintain compliance with U.S. regulations regarding data privacy.

Yet not every company is convinced. Jon Gordner, co-founder of Comment.io, emphasized that for startups focused on developing high-quality software rapidly, the savings offered by switching to cheaper models may be outweighed by the cost incurred from potential errors in performance. He acknowledged that while current discounts from companies like Anthropic and OpenAI are advantageous, the long-term landscape may compel them to reconsider Chinese and other open-source models as competition intensifies.

Economist Ara Kharazian from Ramp supports the notion that U.S. firms will continue to adapt their AI strategies. He believes that the pressure from emerging Chinese models could prompt American companies to innovate and offer high-quality alternatives. However, Gordner remains skeptical, suggesting that leading U.S. AI companies might feel compelled to increase their prices as they approach potential initial public offerings, which could alter the competitive dynamics. 

As the market evolves, the balance of power in the AI landscape remains fluid, with startups weighing cost against performance, and a critical eye on the viability of both American and Chinese models.

U.S. AI is costly, prompting some startups to adopt affordable Chinese alternatives.

A bright, sunlit modern startup office: a young founder stands between two minimalist desks. On the left, a sleek monitor displays the American flag with a dangling gold price tag showing high dollar signs. On the right, another screen shows the Chinese flag with a simple red price tag marked much lower. Natural light streams through floor-to-ceiling windows, casting soft shadows over sketch-filled whiteboards and a steaming coffee mug.

American AI is expensive. Some startups are turning to cheap Chinese models - NPR

- Startups like Lindy.ai are switching to cheaper Chinese AI models, saving significant costs.  
- Chinese AI models are currently six to 12 months behind U.S. counterparts in capabilities.  
- Companies are increasingly adopting open-source Chinese models despite potential performance concerns.  

American AI is expensive. Some startups are turning to cheap Chinese models

toggle caption Imen Ben Youssef/Hans Lucas/AFP via Getty Images

Sign up for NPR's Up First morning newsletter for the stories you need to start your day.

SAN FRANCISCO — Flo Crivello's San Francisco-based startup, Lindy.ai, creates artificial intelligence "assistants" to manage your email and calendar. At first, the company leaned heavily on Anthropic's top-of-the-line AI models.

But in meeting after meeting with his finance guy, Crivello said, one thing became clear: "By far, our No. 1 expense was Anthropic," he said. "Like, more than payroll."

Sponsor Message

More than payroll — for over two dozen employees. More than rent. More than for anything else. So last month, Crivello announced that Lindy had migrated 100% of its traffic to the Chinese AI model DeepSeek-V4.

"It was just 10x cheaper," he said, adding that it had saved the company millions of dollars. "So it was a very, very simple business decision."

Artificial intelligence has become one of the — it not the — fastest-growing costs for U.S. businesses. But for many companies, it's a double-edged sword: necessary but expensive. To survive, a growing number of firms are switching from American models to cheaper Chinese AI.

In the race to create the best AI models, U.S. companies like Anthropic, OpenAI and Google lead the world. Experts say Chinese models are six to 12 months behind in terms of capabilities.

But China has carved out a niche in open-source models, which are free to download and adapt. "The open-source scene right now is absolutely dominated by the Chinese. It's not even close," Crivello said.

He said every founder he knows who is working in the AI space either is thinking about switching to Chinese models or has done so already.

And ballooning AI costs are not just a startup issue either. Uber CEO Dara Khosrowshahi spoke about it last month on the Invest Like the Best podcast. "We blew through our AI budget in a quarter, you know, for the whole year, essentially. And it is forcing us to adjust," he said.

Sponsor Message

(Uber did not respond to NPR's request for information about whether it uses Chinese models.)

Bloomberg reported Airbnb CEO Brian Chesky as saying that last year the company relied on Alibaba's Qwen model, which was "good," "fast and cheap." Perplexity and Nvidia have also made use of Qwen.

Like a Ferrari or a Honda

Many companies are wary of trumpeting their use of Chinese models due to political sensitivities, but the models are widely available on AI-model hubs like Hugging Face, on the code hosting platform GitHub and via model aggregators and inference providers based outside China.

That includes the San Francisco-based company Featherless, which offers access to some 30,000 AI models. Founder and CEO Eugene Cheah said Chinese models are popular, even if they aren't "frontier" models, or best in class.

"It's like the difference between driving a Ferrari and a Honda. You can have the best luxury car, or you can just have a Honda at scale that works," he said.

"Actually, a lot of open-source AI groups are perfectly fine being N-1, N being where the frontier is," he continued. "Because as the gap keeps shrinking, at some point the question is: Does it actually matter?"

For many, like Lindy, it doesn't matter. The Honda of AI is perfectly good.

OpenRouter, another platform where startups can access a range of AI models, reported that use of China's DeepSeek has gone from around 9% to nearly 20% since January. Use of models from the Chinese companies MiniMax, Xiaomi and Tencent have also risen.

Some users download and self-host open-source Chinese AI models, but many use them via paid AI-hosting companies, like Featherless and OpenRouter, so that user data is kept in the United States.

Victor Su-Ortiz, who does global product marketing at the Shanghai-based MiniMax, attended a recent AI engineers conference in San Francisco. Companies pay to use AI models by paying for tokens, or units of AI work. Su-Ortiz said it all comes down to the cost per token.

"A lot of repetitive tasks can be done with a model that's just as performant but has much lower cost per token" when compared with leading AI models, he said. "And this is essentially what has brought these open-weight models into the United States."

Sponsor Message

He said companies are shifting from "tokenmaxxing" — using as much AI as possible — to saving costs by limiting usage, switching to cheaper models or routing different types of AI work to different kinds of models.

For research or "deep reasoning," for instance, the cutting-edge models may perform better, said Su-Ortiz. "But if you're routing for a coding task that is repetitive, high volume … that's where one of our models, especially MiniMax M3, will perform exceptionally well at like only one-tenth the cost."

Saving a few dollars isn't worth it for everyone

For some companies, Chinese models still aren't good enough. Jon Gordner is CEO and co-founder of Comment.io, which was founded just weeks ago and is developing a product that he said is like Google Docs for coders and AI agents.

"We need to make as good software as we can as fast as possible. And for us, saving a few dollars on a cheaper model isn't worth it if we have to spend two or three more weeks fixing its mistakes," he said.

Gordner said his company is getting value out of Anthropic and OpenAI models in part because both companies are subsidizing users to hook customers. He said monthly subscriptions offer tokens at a huge discount now — but that probably won't last forever.

"Then for us, it's going to make a lot more sense to start evaluating Chinese models and open-source models," he said.

Ara Kharazian is lead economist at Ramp, a company that helps businesses track, control and automate spending. It has insight into AI spending, and Kharazian said he thinks U.S. companies will keep adapting — in other words, they may keep prices in check or introduce high-quality open-source models in a bid to outcompete Chinese rivals.

"The rise of these Chinese models is indicative of the fact that businesses want something that is today not being offered by the American model companies," he said. "The only reason why I'm bearish about the Chinese models is because I assume that the American model companies will respond competitively."

Sponsor Message

Gordner, of Comment.io, is less certain. He thinks the major U.S. AI companies may have to start charging more for AI as pressure to demonstrate profitability rises, possibly as they get closer to going public. Both Anthropic and OpenAI have filed confidential paperwork with the U.S. government to get the ball rolling on eventual initial public offerings.

"At some point," Gordner said, "the music's going to stop."

Anthropic is a financial supporter of NPR.

American AI is expensive. Some startups are turning to cheap Chinese models

The proportion of equity given to employees under the age of 30 in startups is on the decline, coinciding with a surge in venture capital investment focused on artificial intelligence. Historically, joining a startup early and accepting equity in lieu of a larger salary has been a reliable strategy for many employees to accumulate wealth. However, this opportunity appears to be shrinking for younger workers.

Recent analysis from altshare, an equity management firm, reveals that stock grants to under-30 employees have plummeted from around 8% to just 3% in recent years. This data comes from a substantial sample—over 3,000 private companies and 120,000 stakeholders—encompassing age information for around 60% of those involved. While altshare's business focuses on startup equity management, independent research corroborates their findings.

The decline in equity distribution is symptomatic of broader shifts in the job market. Young adults are increasingly missing out on equity-linked positions as investment flows more heavily into a select group of AI and cybersecurity firms. Additionally, the initial ownership stake created at startup inception is being divided among fewer individuals, primarily due to the rise in single-founder startups. These trends are shaping ownership dynamics and the potential wealth of future companies.

Ronen Solomon, the founder and CEO of altshare, clarifies that it's not so much that young employees are being stripped of existing equity but rather that they are being less frequently hired for positions that offer equity. Increased focus on specialized hires means that entry-level roles are dwindling. According to Solomon, “Young workers aren’t being cut out of the upside; they’re just not being placed in positions that offer it.” This aligns with broader employment trends: smaller teams with fewer junior roles are less likely to offer equity, reflecting changes identified in recent economic research.

Venture capital trends show an unmistakable shift towards AI. Carta reports that over 60% of venture capital raised in the first quarter of 2026 was directed towards AI companies—the highest proportion recorded to date. The median valuation for Series A funding in AI firms is approximately $300 million, compared to around $55 million for companies in other sectors. While overall early-stage valuations are rising, Solomon warns against assuming that concentrated investment guarantees better returns. The average pre-seed round has reached $1.9 million, indicating a trend of founders raising larger sums before establishing valuations.

Another significant shift is occurring in the composition of startup founding teams. From 2021 to 2025, the percentage of solo founders among startups has surged from around 12% to nearly 25%. Despite this growth in solo entrepreneurship, a Carta report found that while solo founders comprised 35% of new startups in 2024, only 17% of those startups secured venture funding. This disparity suggests that although founding alone is becoming more common, it doesn’t necessarily translate into successful fundraising.

Interestingly, emerging founders are retaining a larger share of their companies over time. The median dilution—the percentage of a company’s equity sold to new investors at each funding stage—has reduced from about 18% to 16% recently, indicating terms that favor founders are becoming more prevalent. Solomon points out that the trend towards smaller founding teams predates the current boom in AI, although it has certainly been accelerated by the evolving landscape where advanced software tools enable individual entrepreneurs to manage businesses more effectively than they could in the past.

AI Is Transforming Wealth Distribution Among Startup Founders

A sunlit modern co-working loft at sunrise: at a long wooden table, seasoned investors push back their chairs in surprise as a sculptural AI “brain” of warm golden circuits floats above the tabletop, pouring streams of amber data lines and coins into the outstretched hands of diverse new founders—visualizing how AI is redistributing startup wealth.

AI Is Changing Who Gets Rich From Startups - Forbes

- The share of startup equity for employees under 30 has dropped from 8% to 3%.  
- Venture capital is increasingly concentrated in AI, with 60% going to AI companies.  
- Solo founders now account for 25% of startups, up from 12% in 2021.  

The share of startup equity going to employees under 30 is falling while venture money crowds into artificial intelligence. getty

For years, one of the surer ways for an ordinary employee to build wealth was to join a startup early and take equity instead of a bigger salary, on the chance the shares were worth something later. Fewer young workers are being handed that chance now.

According to altshare , an equity-management firm, the share of stock grants going to employees under 30 has fallen from about 8% to 3% over the past few years. The firm drew the figure from more than 3,000 private companies and 120,000 stakeholders on its platform, with ages on file for about 60% of them. altshare sells software for managing startup equity, so its data comes with an interest attached. The number is not an outlier, though. It matches what economists studying the job market have found using far larger datasets.

The under-30 figure is one strand of a wider change. Young people are landing fewer of the jobs that come with equity. Venture money is concentrating in a narrow set of AI and cybersecurity companies. And the stock created when a company is founded is being split among fewer people, as more startups are run by a single founder. Each of these decides who owns a piece of the next wave of companies, and how big that piece is.

Equity Follows Hiring Ronen Solomon, altshare’s founder and chief executive, is careful about what the drop does and does not mean. Young workers are not having equity taken away, he says. They are being hired into the roles that carry equity far less often. “Young workers aren’t so much being cut out of the upside as they’re not getting into the roles that carry it in the first place,” Solomon says. Teams are smaller and take on fewer junior people, and grants track headcount. When entry-level hiring falls, the under-30 share of equity falls with it. That fits what economists have been seeing in the wider labor market. In a study titled “Canaries in the Coal Mine,” first released in August 2025, researchers at Stanford’s Digital Economy Lab, led by Erik Brynjolfsson with Bharat Chandar and Ruyu Chen, used payroll records from ADP covering millions of American workers. They found that after generative AI came into wide use, employment for workers aged 22 to 25 in the most exposed occupations fell 13% relative to older colleagues, even once company-specific shocks were accounted for. Software development and customer support, both common ways into a technical career, were among the hardest hit. In the same high-exposure roles, employment for workers over 30 rose by 6% to 12%. The equity data lines up with that. If companies are not hiring juniors into engineering and support, those juniors are not receiving the grants such jobs used to include.

The Money Is Going To AI Venture funding is running in one direction. Carta reported that more than 60% of all the venture capital raised on its platform in the first quarter of 2026 went to AI companies, the largest share it has recorded. Prices show the crowding. The same report put the median Series A valuation for an AI foundational-model company at about $300 million, against roughly $55 million for a company at the same stage doing something else. Early-stage valuations have risen generally, not only for AI: the median Series A valuation reached a record $78.7 million in the last quarter of 2025, up 37% on the year, with the median seed at $24 million. Companies that can credibly attach AI to their pitch tend to price above even those levels. Solomon is wary of reading concentration as a promise of returns. “Concentration on its own doesn’t tell you returns will be stronger,” he says. His platform sees how companies are priced and owned at the point of investment, not how they do at exit, where returns are settled. Cybersecurity carries the highest valuation premiums in altshare’s data, he says, though he points to restraint within those sectors: the typical cybersecurity Series A check has come down from its peak, by roughly a quarter on his numbers, and revenue multiples have narrowed. The risk he flags for investors is the entry price. Paying a premium to get in raises the return the investment then has to earn to be worthwhile. One altshare figure looks, at first, to contradict the rest: a record median pre-seed round of $1.9 million, when most reports describe early rounds getting smaller. Solomon says the two can sit together. His $1.9 million measures the median pre-seed SAFE, the instrument founders use to raise money before agreeing on a valuation, and founders are choosing to raise more at that first, unpriced stage. Pre-seed data from Carta shows the typical round shrinking, with a small number of very large rounds pulling the average up. The two are counting different things.

Smaller Founding Teams The people starting companies are changing too. Among the businesses altshare works with, the share founded by one person has climbed to about a quarter of those started in 2025, from around 12% in 2021. Solomon is specific about the measure: companies that have taken outside money and set up a cap table, grouped by the year they were founded. A separate Founder Ownership Report from Carta points the same way, with one caveat. It found that solo founders accounted for about 35% of new startups formed on that platform in 2024, up from 17% in 2017, but only about 17% of the companies that went on to raise a venture round. Founding alone is far more common at incorporation than among companies that get funded. altshare’s figure counts a first SAFE as funding, a lower bar. Both sets show founding teams shrinking. Two cofounders remains the most common arrangement. There is a counterweight to any account of workers losing out. Founders are keeping more of their companies than they used to. Median dilution, the share of a company sold to new investors at each round, has fallen across the early stages, from about 18% to 16% over the past year, with down rounds now rare and terms generally favoring founders. Owners are doing better at the same time that fewer employees are getting onto the cap table at all. Solomon agrees with the direction and adds a caveat. The move toward solo founders, he says, “predates the current AI wave, which has accelerated it rather than created it.” Cheaper software has made it easier for one person to run a company further on their own than they could a few years ago.

AI Is Changing Who Gets Rich From Startups

Jack Manning has navigated the complex emotional landscape of creating a fitting tribute to a loved one. After the loss of his sister over a decade ago, he and his family dedicated significant time and effort to organizing a volunteer project that honored her memory.

“It required years of bake sales and 5K runs to fund a memorial that held real meaning for us. We truly experienced how much work goes into establishing a significant tribute,” he shared with CBS News. “It got me thinking, surely technology could ease this process."

This is where artificial intelligence steps in. As a doctoral candidate in information science at the University of Colorado Boulder, Manning is delving into the potential of large language models (LLMs) to develop what some refer to as "generative ghosts." These AI-driven chatbots are designed using a deceased person's digital footprint—social media posts, emails, audio and video materials, and photographs—to create a virtual simulation that friends and family members can engage with posthumously.

“It’s definitely a burgeoning field. Interest is rising in exploring the intersection of AI technology and grief,” Manning noted.

Creating a "lifelike" persona

A number of startups are now pioneering these solutions, such as Séance AI, You, Only Virtual, and re;memory. For instance, re;memory allows individuals to generate realistic avatars of their departed loved ones using their images and recorded voices.

“By enabling interaction with a realistic digital avatar, users can find comfort in expressing love and forgiveness, creating a connective experience to the treasured memories they share,” the company explains. Their platform allows users to visualize and hear their parents as they appeared in beloved photographs, making the experience more vivid than mere imagination.

In a research paper published in June by the Association for Computing Machinery, lead researcher Jed Brubaker—an associate professor at the University of Colorado Boulder specializing in socio-technical systems—along with Manning and his team, observed how participants reacted to these AI simulations of their deceased relatives.

“Participants completed a basic survey sharing fond memories of their loved ones, discussing their backgrounds and defining traits,” Manning explained. “They were captivated by the capabilities of an LLM to generate responses from minimal input.”

Manning outlined two fundamental types of generative ghosts. The first, a basic "death bot," might only replay recorded statements from a deceased individual without mimicking their unique communication style.

“You can achieve a basic, albeit simplistic, representation of someone by using just one prompt in a free ChatGPT account,” Brubaker indicated. “Just as you could prompt it to articulate like Shakespeare, you might say, 'Be like my grandfather—here are some characteristics about him.'"

Facilitating ongoing dialogue

Generative ghosts go beyond this basic functionality, employing the deceased’s voice and speaking style, and allowing real-time interactions with the living, including answering questions in a manner synonymous with the deceased.

“They can articulate responses that a loved one might not have conveyed. These AI models can respond to inquiries that remained unanswered in life,” Manning said.

Pricing for these services varies. For instance, re;memory offers individual users the chance to create three customized avatars for $24 monthly. Meanwhile, Sceance AI provides animated representations of loved ones—capable of smiling, moving their heads slightly, and speaking in a voice akin to the deceased—for $19.99 per month.

Researchers discovered that participants typically favored interacting with an AI version of their loved one in the first person, appreciating a direct reincarnation rather than an avatar narrating in the third person. However, if a chatbot used affectionate terms that the deceased wouldn’t typically say, such as "champ" or "pal," it tended to diminish the user's desire to continue the conversation.

While generative ghosts may share traits with deepfakes, which can distort public figures’ speech and actions, Brubaker emphasized the crucial difference between the two.

“The intent behind a deepfake is to mislead,” Brubaker explained. “In contrast, generative ghosts aim to provide comfort and connection, not to deceive.”

AI firms are producing "generative ghosts" of people who have passed away.

A dim, moody living room at night: a lone adult sits at a wooden table bathed in warm lamplight, facing a translucent, ethereal figure of a loved one—its edges flickering with digital noise and faint lines of code. On the table are scattered old photographs and a sleek, unbranded device emitting a soft glow. Deep shadows hug the corners, the scene heavy with longing and the uncanny.

AI companies are creating "generative ghosts" of deceased loved ones - CBS News

- Jack Manning is exploring AI-driven "generative ghosts" to honor deceased loved ones meaningfully.  
- Startups like Re;memory create lifelike avatars allowing interaction with deceased individuals' personas.  
- Users prefer generative ghosts responding in the first person, enhancing the emotional connection.  

Jack Manning knows firsthand the emotional and practical challenge of memorializing a loved one in a way that feels appropriate. After losing his sister more than 10 years ago, he and his family spent hundreds of hours raising funds to organize a volunteer project honoring her.

"It took years of bake sales and 5Ks to raise money to honor her memory in a way that was meaningful to us. We got to see how much effort is required to create a memorial that's impactful," he told CBS News. "I thought, surely there is a way that technology can help us."

Enter artificial intelligence. Manning, a doctoral candidate in information science at the University of Colorado Boulder, is now studying the use of large language models (LLM) to create what some are calling "generative ghosts" — AI chatbots trained on a dead person's social media posts, emails, audio and video recordings, photos and other data to create a digital simulation that family and friends can interact with after the individual's death.

"It is definitely a growing sector. We're seeing growth in interest in how AI and grief intertwine," Manning said.

A "lifelike" avatar

A number of startups now offer these services, including Séance AI, You, Only Virtual and re;memory. For example, Re;memory, lets users create lifelike avatars of their loved ones based on photos and voice recordings.

"By allowing you to interact with a lifelike digital avatar of your loved one, you can find solace in expressing your love and forgiveness, creating a bridge to the cherished moments you hold dear," the company says on its website. The tool "allows you to see and hear your parents as they were in their cherished photos, rather than just imagining them."

In a June paper published in the Association for Computing Machinery by lead researcher Jed Brubaker, an associate professor at the University of Colorado Boulder who studies socio-technical systems, Manning and his colleagues observed study participants' reactions to the AI ghosts of people they have lost.

"Participants filled out a simple survey sharing positive memories of a loved one, talking about their upbringing and core traits," he explained. "And they were fascinated by how much an LLM can do with so little information."

There are two basic kinds of generative ghosts, according to Manning. A simple "death bot" could be limited to playing back verbatim any statements recorded by a loved one, but would be incapable of simulating the deceased's manner of communication.

"You can get a reasonable, although thin representation of a loved one with a single prompt on a free ChatGPT account," Brubaker said. "The same way you might prompt it to say, 'Talk to me like Shakespeare,' you can say, 'Talk to me like my grandfather — here are some details about how he acted.'"

Continuing the conversation

Generative ghosts go significantly further by recreating the deceased's voice and manner of communication, and offering the ability to interact with the living, even responding to questions in a recognizably familiar way.

"They are capable of saying things a loved one might not have said. Ghosts are AI models that can talk for somebody, maybe answering a question that they didn't answer when they were alive," Manning said.

The cost to consumers is variable. Re;memory charges $24 a month for individuals to create three custom avatars. For $19.99 per month, Sceance AI users can create animated images of their loved ones that can smile, make gentle head movements and speak in the deceased's voice.

The researchers found that study participants generally preferred communicating with an AI "ghost" of their loved one in the first person, with the chatbot serving as a direct reincarnation, rather than an avatar speaking in the third person. One turnoff: if a bot used a term of endearment, like "champ" or "pal," that their loved one typically didn't say, a user wanted to end the interaction.

Generative ghosts share some commonalities with so-called deepfakes, which can be designed to mislead the public by fabricating a public figure's speech and actions. But Brubaker said there is a distinct difference.

"The fundamental premise of a deep fake is the intent to deceive," Brubaker said. "The intent of generative ghosts is not to deceive another person."

Reasons Paris could be the leading AI hub beyond Silicon Valley.

More from AI Startups & Investments

Is Kimi K3 Worth a Try? A Comparison of AI Models Against ChatGPT and Claude.

Sequoia invests $45 million in Israeli entrepreneur developing the next generation of

Google's collaboration with former Twitter CEO Parag Agrawal's $2 billion AI search company explained.

Applied Computing aims to provide oil and gas operators with a comprehensive AI model for their facilities.