A version of this report originally appeared in CNBC's Inside Wealth newsletter, curated by Robert Frank, offering weekly insights for high-net-worth investors and consumers. Subscribe to receive future editions directly to your inbox.

Thanks to Jeff Bezos, the summer season is kicking off positively for investment firms catering to the ultra-wealthy.

In June, the Amazon founder’s family office executed five direct investments in startups, which represented 10% of the overall family office deal activity, according to exclusive information from Fintrx, a platform focused on private wealth intelligence. Bezos Expeditions has emerged as the most active family office investor in 2023 so far, making a total of eight direct investments in private enterprises, based on Fintrx's statistics.

The 21-year-old family office took part in five significant funding rounds for artificial intelligence companies last month, including a remarkable $12 billion Series B for Prometheus. This startup, co-founded and co-led by Bezos, has reached a valuation of around $41 billion. During an interview with CNBC's David Faber on June 11, Bezos explained that Prometheus aims to develop an "artificial engineer" designed to expedite the design and production process of physical products, ranging from jet engines to pharmaceuticals.

"What drives national wealth? What underpins the wealth of civilizations? The answer lies in invention," Bezos articulated in his interview on "Squawk Box." He elaborated on Prometheus's mission: "Our goal is to create tools that enhance the invention cycle. How long does it take to refine an idea? From concept to actual production, how quickly can we create a functional object?"

Bezos noted that Prometheus has had to secure significant funding — surpassing $18 billion to date — in order to assemble large datasets, which necessitate substantial computational resources.

Jeff Bezos' family office invested in five AI startups in June.

A sunlit, modern glass boardroom overlooking a city skyline at dawn. On a polished mahogany desk, a gold nameplate reads “Bezos Family Office.” A business-suited hand gently places five tiny circuit-board seedlings—each sprouting robotic arms—onto neat stacks of gold coins, symbolizing five AI startups backed in June. Warm natural light streams through floor-to-ceiling windows, casting soft shadows across the scene.

Jeff Bezos' family office backed five AI startups in June - CNBC

- Jeff Bezos' family office, Bezos Expeditions, leads family office investments with eight deals.  
- In June, Bezos participated in five major investments in AI startups, including Prometheus.  
- Prometheus, co-founded by Bezos, aims to accelerate invention in product design and manufacturing.

A version of this article first appeared in CNBC's Inside Wealth newsletter with Robert Frank, a weekly guide to the high-net-worth investor and consumer. Sign up to receive future editions, straight to your inbox.

Thanks to Jeff Bezos, summer is off to a strong start for investment firms of the ultra-rich.

In June, the Amazon founder's family office made five direct investments in startups, accounting for 10% of family office dealmaking, according to exclusive data provided by Fintrx, the private wealth intelligence platform. Bezos Expeditions is now the most active family office investor thus far this year with eight direct investments in private companies, per Fintrx data.

The 21-year-old family office participated in five megarounds for artificial intelligence startups last month, including a $12 billion Series B for Prometheus. The startup, now valued at about $41 billion, counts Bezos as a cofounder and co-CEO. Prometheus aims to create an "artificial engineer" that will speed up the design and manufacturing of physical products from jet engines to pharmaceuticals, Bezos told CNBC's David Faber on June 11.

"What drives the wealth of nations? What drives civilizational wealth? ... The answer is invention," Bezos said on in an interview on "Squawk Box." "Our goal at Prometheus, what we're working on is building a set of tools that accelerate that invention loop. So, how long does it take to improve something? How long does it take to – from idea to actually manufacturing, seeing it rate and have a useful object?"

He added that Prometheus has had to raise so much capital — more than $18 billion to date — in order to build massive datasets, which requires a lot of compute power.

Jeff Bezos' family office backed five AI startups in June

OpenAI, a leader in artificial intelligence, is exploring the opportunity to set up operations in Israel after years of consideration. Unlike a traditional research and development hub, this initial initiative will focus on business development to foster connections with the local tech community.

Recent reports from Calcalist reveal that OpenAI has hired a senior executive from Amazon, who previously led AWS for Startups in Israel. This new hire will oversee operations in the EMEA region, which includes Europe, the Middle East, and Africa, while primarily concentrating on the Israeli market. He will report to a manager based in London.

The company behind ChatGPT is aiming to integrate into the Israeli tech ecosystem in a manner reminiscent of cloud service providers. These firms typically engage with local startups early in their development, nurturing these relationships as the startups scale and increase their revenue.

OpenAI plans to host a startup event in Israel in collaboration with AWS at the end of the month, where they will showcase their enterprise services. Their careers page also lists a position for an account manager focusing on the Israeli tech landscape. Officially based in Paris, this role is expected to involve working directly from Israel and reporting to European management. If things progress positively, OpenAI may later seek a senior executive to run a sales office specifically in Israel.

OpenAI’s strategy for international expansion has generally prioritized marketing and business development before establishing R&D facilities, a pattern similar to their entry into India last year, where they swiftly expanded and expect a fully operational office by 2025.

While the establishment of a business development office in Israel is not equivalent to a dedicated R&D center like those set up by Amazon, Google, Apple, Microsoft, or Nvidia, which employ thousands in the country, it signifies an important step. Google, for instance, began with a focus on business before expanding its R&D efforts.

Currently, OpenAI operates only two R&D centers outside the U.S.—a primary location in London and another in Zurich, recognized as a significant AI hub due to its academic institutions. The move into Israel highlights the value OpenAI sees in this thriving tech environment. Despite not having developed its own large language model, Israel is recognized as a global leader in AI utilization, especially in enterprise applications.

Recent data from Anthropic, OpenAI's competitive rival, shows Israel leading globally in per-capita use of Claude, Anthropic’s AI tool, surpassing other advanced nations. The report emphasized Israel's rapid adoption of AI technologies within its workforce, reflecting the innovative spirit of the local ecosystem.

A recent Bank of America analysis outlined Israel's position in the global AI landscape, noting that while it lags behind in infrastructure and public sector investment, it stands out for its robust private sector involvement. LinkedIn data places Israel first in AI talent concentration and third in investment in AI firms from 2013 to 2025, amounting to $19 billion.

With numerous startups utilizing language models, OpenAI is keen to tap into this burgeoning market. The popularity of Claude and Google's Gemini in Israel has led to potential market disadvantages for OpenAI.

Establishing a foothold in Israel could not only enhance OpenAI's language model through increased user interaction but also lead to lucrative business opportunities as these startup companies grow, mirroring the substantial cloud contracts awarded to Israeli tech firms over the last decade.

Amid plans for a Wall Street IPO next year, OpenAI faces stiff competition from Anthropic, which has outpaced it in enterprise-market penetration and international reach. Despite being the first to release a widely available AI application, OpenAI has seen Anthropic gaining ground in revenue per client and European and East Asian expansion, with offices in Germany, Japan, and South Korea.

As a result, Anthropic is anticipated to surpass OpenAI in annualized revenue this year, nearing $50 billion, compared to OpenAI's current $25 billion, despite the latter's larger user base. Both companies are not yet profitable, although Anthropic appears to be managing its finances more efficiently.

In a notable shift, Anthropic's valuation has recently exceeded OpenAI's for the first time, hitting $965 billion against OpenAI's $852 billion. Following this trajectory, Anthropic is expected to go public ahead of OpenAI, which has postponed its initial public offering until 2027, moving away from its initial goal of late 2026 due to uncertainties regarding achieving a valuation exceeding $1 trillion.

OpenAI considers expanding in Israel by hiring employees and exploring a business center.

A sunlit high-rise conference room in Tel Aviv: warm natural light floods through floor-to-ceiling windows overlooking the city skyline. Around a sleek wooden table, a diverse team of professionals leans over blueprints and laptops, one map of Israel pinned with a single Israeli flag. Behind them, the word “OpenAI” appears in bold text on a frosted glass wall. Papers, coffee mugs, and digital tablets hint at a business hub in the making, as staff exchange ideas in a bright, optimistic atmosphere.

OpenAI eyes Israel expansion, hires staff and weighs business hub - Ynetnews

- OpenAI plans to establish a business development operation in Israel, not R&D.  
- A senior Amazon executive has joined OpenAI to lead operations in the EMEA region.  
- Israel ranks high in AI adoption, making it a critical market for OpenAI.  

It took several years, but artificial intelligence pioneer OpenAI is considering establishing operations in Israel. The initial move would not involve a research and development center but rather a business development operation aimed primarily at building ties with the local tech sector.

According to information obtained by Calcalist, a senior Israeli Amazon executive who until recently led AWS for Startups — Amazon’s cloud operations for the local tech industry — has been recruited by OpenAI and is beginning work at the company. He will be responsible for the EMEA region, covering Europe, the Middle East and Africa, and will report to a manager in London, though he is expected to focus primarily on Israel.

The AI giant, which launched ChatGPT , its large language model-based application for the general public, wants to build a presence in the Israeli ecosystem using a model similar to that of cloud companies. Such companies establish relationships with local startups soon after they are founded, then increase their revenue as the startups grow and develop.

OpenAI is expected to hold an event for startups in Israel with AWS toward the end of the month, where it will present its services for the enterprise market. At the same time, OpenAI’s careers website currently lists an opening for an account manager serving Israel’s tech sector.

Although the position is officially listed as being based in Paris, the managers are expected to work from Israel and report to the company’s European offices. Depending on how its Israeli operations develop, people familiar with OpenAI’s plans believe the company will later seek a senior executive to lead an Israeli sales office in a country manager role.

Depth of AI adoption in the ecosystem

OpenAI’s expansion outside the United States has generally begun with marketing and business development operations rather than R&D centers, followed later by the recruitment of engineers. In comparison, OpenAI began operating in India a year ago by hiring for positions similar to those it is now seeking to fill in Israel. By the end of 2025, it already had an active office in the country.

This is not yet the holy grail for Israel, a local R&D center like those operated by Amazon, Google, Apple, Microsoft and Nvidia, each of which employs thousands of people in the country, but it represents an important initial foothold.

Google’s operations in Israel also began with a greater focus on marketing and business activity before expanding into R&D. Historically, development centers in Israel have often been established through acquisitions of local startups or larger companies, as in the case of Mellanox, which was already publicly traded when it was acquired by Nvidia.

OpenAI currently has only two R&D centers outside the United States: one in London, its largest, and another in Zurich, which has become a significant AI hub because of the city’s academic institutions.

OpenAI’s interest in establishing business development operations in Israel illustrates the importance Sam Altman’s company places on the Israeli tech industry. Although Israel has not developed its own large language model, it is considered one of the world’s most advanced countries in the use of AI tools, particularly in the enterprise market among tech companies.

According to recent data from Anthropic, OpenAI’s fierce rival, Israel ranks first worldwide in per-capita use of Claude, Anthropic’s AI tool. The study, conducted in March, examined use among working-age populations around the world and found that Israel surpassed countries including the United States and Singapore.

The usage figures highlighted both the depth of AI adoption in the Israeli ecosystem and the speed at which new technologies are embraced.

Calcalist also reported last week on a Bank of America study examining the countries leading in various aspects of AI behind the two dominant powers, the United States and China.

Israel ranked just outside the global top 10. Its position was hurt primarily by weak national infrastructure and limited government investment, while it ranked much higher in measures related to the private sector.

Israel ranked first in the concentration of AI talent, according to LinkedIn data, and third in investment in AI companies between 2013 and 2025, with a total of $19 billion.

Hundreds of Israeli startups now build their products on language models, and OpenAI does not want to miss out on this significant market. The widespread adoption of Claude, which is highly popular among developers, and Gemini, aided by Google’s strong presence in Israel, has put OpenAI at a disadvantage.

A growing startup that eventually generates hundreds of millions of dollars in revenue could also become a major customer spending large sums on AI models, much as Israeli tech companies became major cloud customers over the past decade.

Establishing a presence in Israel could also help OpenAI improve its language model, which develops as more people use it.

Annual revenue run rate of $50 billion

OpenAI, which is seeking a Wall Street flotation in the coming year, significantly trails Anthropic in both enterprise-market penetration and expansion outside the United States.

OpenAI was the first company to launch a widely accessible artificial intelligence application and remains the better-known brand. That has attracted many customers who use either the free version, which accounts for the overwhelming majority of users, or the basic paid version, which costs $20 a month.

Anthropic , by contrast, focused on the enterprise market from the outset, generating significantly higher revenue per customer. It also invested more heavily in expanding across Europe and East Asia, establishing centers in Germany, Japan and South Korea — countries that, like Israel, rank among the leaders in AI adoption.

As a result, Anthropic is expected to overtake OpenAI this year in annualized revenue, approaching $50 billion compared with OpenAI’s $25 billion, despite OpenAI’s commanding lead in user numbers.

Neither company is profitable, though Anthropic is believed to be burning less cash.

Anthropic also recently surpassed OpenAI’s valuation for the first time in a private funding round, reaching a valuation of $965 billion compared with OpenAI’s latest valuation of $852 billion.

Anthropic is now expected to go public before its older rival. OpenAI recently decided to postpone its initial public offering until 2027, rather than proceed with its original plan for late 2026, amid concerns that it would not secure the valuation of more than $1 trillion it is seeking.

OpenAI eyes Israel expansion, hires staff and weighs business hub

AI infrastructure firm Infinity announced on Monday that it has successfully raised $15 million at a valuation of $100 million, attracting investment from firms such as Touring Capital, Principal VC, and experts from renowned companies like OpenAI and Anthropic.

Infinity is focused on developing software that simplifies the implementation of AI models on AI chips. The dominance of Nvidia in the market is attributable not only to its powerful chips but also to its CUDA software (Compute Unified Device Architecture), enabling its GPUs—initially designed for graphics processing—to operate as general-purpose CPUs. Popular AI development frameworks like PyTorch and TensorFlow are built on CUDA, which allows developers to create applications using widely-used programming languages such as Python, resulting in seamless operation on Nvidia hardware.

Many startups in the application space lack the expertise or resources to create their own kernels, the fundamental software that drives hardware, or to adapt their applications for various AI chips. Infinity aims to create an alternative kernel software compatible with a diverse range of chips, including SRAM, GPUs, mobile processors, and Systolic Arrays. This initiative is part of a broader trend where startups are systematically challenging Nvidia's market supremacy.

Infinity is working towards a universal inference library capable of functioning across different chip brands, enabling these chips to automate the replication of cutting-edge research results.

Founded last year by Jeremy Nixon, a former researcher at Google Brain and the creator of the AGI House hacker network community, Infinity emerged from Nixon's fascination with the concept of “automated invention.” He envisions AI systems as a transformative meta-technology. Nixon developed a machine learning algorithm named Omega, which autonomously generated and evaluated new algorithms within a feedback loop.

Inspired by this success, he explored applications in hardware, believing that automated systems could similarly produce the fundamental code necessary for enhanced chip performance.

Infinity’s AI research agent, Ignition, is designed to create the low-level code essential for AI inference on alternative chips to Nvidia. It autonomously tests, debugs, and assesses the hardware's performance with the generated code, adjusting it as needed to optimize speed. This self-optimizing system continuously learns and adapts to different chip architectures, regardless of proprietary constraints, according to Nixon. Infinity claims it has developed a software stack comparable to CUDA.

Infinity's client roster includes D-Matrix, an AI chip manufacturer positioned as a competitor to Nvidia, and Nixon mentioned ongoing discussions with other major chip and cloud providers.

While the agent handles the more labor-intensive tasks, human oversight remains essential for providing strategic direction. In one instance, the startup noted that the agent significantly outperformed human efforts, condensing a potentially years-long project into just hours or days. Infinity doesn't impose an upfront licensing fee; instead, it earns a portion of the performance improvements and cost savings, based on changes measured in tokens per second.

Currently, Infinity employs 26 staff members across design, operations, and engineering roles.

Inference startup Infinity secures $15M from Touring Capital along with researchers from OpenAI and Anthropic.

A sunlit, modern boardroom scene centered on a polished infinity-symbol sculpture perched atop a glowing pile of golden coins stamped “$15M.” Warm daylight streams through floor-to-ceiling windows, illuminating three banners: one with Touring Capital’s logo, one with the Anthropic logo, and a frosted glass partition etched simply with the word “OpenAI.” Subtle confetti sparkles drift in the air, capturing the celebratory energy of a successful funding round.

Inference startup Infinity raises $15M from Touring Capital, OpenAI and Anthropic researchers - TechCrunch

- Infinity raised $15 million at a $100 million valuation for AI chip software development.  
- The startup aims to create software alternatives for Nvidia's CUDA, enhancing chip performance.  
- Infinity’s AI agent automates low-level code writing, significantly speeding up development processes.  

AI infrastructure company Infinity announced a $15 million raise at a $100 million valuation on Monday from investors including Touring Capital, Principal VC, and researchers from companies such as OpenAI and Anthropic.



The startup is building software to make it easier for AI chips to run AI models. One big reason Nvidia became the top player is not just its high-performance chips, but also its CUDA software (Compute Unified Device Architecture), which allows its GPUs (originally designed to run graphics) to act as general-purpose processing CPUs. The largest AI development frameworks PyTorch and TensorFlow have been built on top of CUDA. This allows developers to write their apps in popular languages like Python, use those major AI frameworks and their apps will, by default, run on Nvidia chips.

Most of these app-level startups wouldn’t have the resources or know-how to write their own kernels — the low-level software that operates chips — and port their apps to other AI chips. So Infinity is trying to build CUDA-alternative kernel software that works with any type of chip, like SRAM, GPUs, phone chips, and Systolic Arrays. Infinity is part of a new wave of startups that are attempting, product by product, to chip away at Nvidia’s market dominance.

Infinity is attempting to build a universal inference library to run on all chips, allowing these chips to automate replicating state-of-the-art research results.

Infinity was launched last year by Jeremy Nixon, once a researcher at Google Brain and creator of the hacker network community AGI House. Nixon told TechCrunch he decided to launch this company because he was obsessed with the idea of “automated invention” — the belief that “AI systems can actually be a meta technology.” He himself had invented a machine learning algorithm called Omega, he said, which essentially created new machine learning algorithms and automatically evaluated them in a feedback loop.

That success got him thinking about other cases where this approach could work, and he turned to hardware, believing that automated systems could also generate the low-level code, like the kernels and so forth, needed to help run chips more effectively.

Infinity’s AI research agent Ignition is intended to write the low-level code needed for AI inference on Nvidia-alternative chips. It tests, debugs, and measures how fast the hardware performs with the code, and automatically rewrites the code if needed to improve performance. The system is self-optimizing, meaning it continuously learns and improves itself. It also adapts to different chip architectures, regardless of proprietary designs, Nixon says. The result is what Infinity claims is a CUDA-level software stack.

Customers include the AI chip maker (and would-be Nvidia challenger) D-Matrix, and Infinity is in talks with other big chip and cloud companies, Nixon said.

Humans are in the loop, however, providing high-level direction while the agent does more of the tedious grunt work. In one case study, the startup found the agent works much faster than a human alone, reducing what could have been a years- or months-long process to hours or days. Infinity doesn’t charge an upfront license fee; instead, it takes a cut of performance gains and cost savings, measuring changes in tokens per second.

Right now, Infinity has 26 employees, including those in design, operations, and engineering.

Inference startup Infinity raises $15M from Touring Capital, OpenAI and Anthropic researchers

Jeff Bezos, the founder of Amazon, along with the UK government, has made a significant investment in a British artificial intelligence startup known as CuspAI, which aims to revolutionize the search for rare materials. This ambitious initiative is backed by £2 billion in funding and seeks to act as the "search engine for rare materials," propelling technological advancements into a new era.

Based in Cambridge, CuspAI has successfully secured $450 million (£330 million) in funding from various investors, including Bezos and the UK government's sovereign AI fund, bringing the company's valuation to $2.6 billion in just two years since its inception.

The startup has introduced the AI Materials Foundry, a collaboration involving over 48 tech firms, industrial corporations, and research institutions focused on leveraging AI to discover and innovate new materials for chip manufacturers and other industries. CuspAI aims to shorten research timelines and potentially reduce or eliminate dependence on rare metals such as iridium and ruthenium in the supply chains of chipmakers.

The coalition features prominent companies like the chip manufacturer Nvidia, Meta Platforms (the owner of Facebook and Instagram), and Hyundai. Together, they plan to harness frontier AI technologies—the most sophisticated, general-purpose models—to identify new materials that could enhance energy networks, improve battery efficiency, and bolster the capabilities of semiconductors that are crucial for the ongoing AI revolution.

CuspAI co-founders Dr. Chad Edwards and Professor Max Welling emphasize the urgent need for progress: "The next 50 years of industrial advancement may be hampered by one pressing issue: the necessity for materials that are currently nonexistent. Our mission combines frontier AI, specialized expertise, proprietary data, and strong partnerships to address this need. It's evident that relying solely on known materials will not suffice for future advancements."

The recent Series B funding round will enable CuspAI to expand globally, with plans to establish a new office in Singapore and increase its workforce in locations such as Cambridge, Amsterdam, Berlin, Tokyo, and the United States.

Josh Coyne, a partner at Kleiner Perkins, the venture capital firm that co-led this funding round, stated, "Significant technological advancements are often tied to materials, and future innovations—ranging from affordable carbon capture solutions to semiconductors and cleaner water—are stalled due to undiscovered materials. CuspAI has created a search engine that addresses this gap."

This latest funding round follows a $100 million Series A investment secured by CuspAI last year. The company is the fourth to benefit from the UK government's fund, which targets early-stage British AI startups to help them scale for global competition, although the exact size of the government's investment remains undisclosed.

For a startup to qualify for the sovereign AI fund, it typically needs to be headquartered in the UK and have British founders. Liz Kendall, the secretary for science and technology, remarked, "CuspAI is harnessing the power of AI to tackle some of the most pressing challenges facing the UK and the global community today, including energy and climate issues. By supporting British companies like CuspAI, we are facilitating breakthroughs that will enhance our economy, safeguard the environment, and improve lives."

To strengthen its position, CuspAI has recruited John Giannandrea, a former executive from Apple and Google, who will assist in establishing its operations in the United States. Additionally, the company is collaborating with Abhi Talwalkar, a board member at chip maker AMD and the chair of Lam Research.

Jeff Bezos and the UK government invest £2bn in the British startup CuspAI.

A sunlit rooftop terrace overlooking the London skyline, where a bald silhouette of Jeff Bezos shakes hands with a suited British government minister beneath a large CuspAI logo banner. A Union Jack flag flutters beside them, and on a nearby table a pile of gleaming gold coins stamped “£2 bn” sparkles in warm natural light.

Jeff Bezos and UK government invest in £2bn British startup CuspAI - The Guardian

- Jeff Bezos and the UK government invested £2bn in the AI startup CuspAI.  
- CuspAI aims to accelerate discovery of new materials for chipmakers and industries.  
- The startup plans international expansion, opening offices in Singapore and several cities.  

Amazon’s founder, Jeff Bezos, and the UK government have invested in a £2bn British artificial intelligence startup that is aiming to become the “search engine for rare materials” that accelerates the next wave of technological breakthroughs.

The Cambridge-based CuspAI has raised $450m (£330m) in funding from investors including Bezos and the government’s sovereign AI fund, valuing the two-year-old business at $2.6bn.

The company has launched the AI Materials Foundry, a coalition of more than 48 tech companies, industrial firms and research facilities, with the aim of building software underpinned by artificial intelligence to discover and develop new materials for chipmakers and other industries.

CuspAI hopes to significantly cut research times and reduce or eliminate the use of rare metals – including iridium and ruthenium – in chipmakers’ supply chains.

The coalition, which includes the chipmaker Nvidia, the Facebook and Instagram owner, Meta Platforms, and the carmaker Hyundai, aims to use frontier AI, the most advanced, general-purpose artificial intelligence models, to discover new materials that could make energy networks and batteries more efficient and advance the power of semiconductors that underpin the AI revolution.

“If we don’t make progress fast, the next 50 years of industrial progress will be constrained by a single challenge: the world needs materials that don’t yet exist,” said the CuspAI co-founders Dr Chad Edwards and Prof Max Welling. “That’s what we’re on a mission to solve – combining frontier agentic AI with deep domain expertise, exclusive data access and close customer partnerships. The world simply can’t make the progress it needs to with known materials, that much is clear.”

The company said the round of series B funding would enable it to expand internationally, opening a new office in Singapore and increasing staff numbers in Cambridge, Amsterdam, Berlin, Tokyo and the US.

Josh Coyne, a partner at Kleiner Perkins, the venture capital firm that co-led the latest round of funding, said: “Most big leaps in technology come down to a material, and the next set, from cheaper carbon capture to semiconductors and cleaner water, is stuck waiting on materials nobody has discovered yet. CuspAI built a search engine that changes that.”

The latest fundraise followed a $100m series A investment last year.

CuspAI is the fourth company to receive funding from the government’s fund targeting investment in early-stage British AI firms to help them scale up to compete on the world stage. The government did not disclose the size of the investment.

Criteria for receiving funding from the sovereign AI fund, which makes a typical equity investment in a startup of between £1m and £10m, include having a main office in the UK and the company’s founders being British.

skip past newsletter promotion Free newsletter | Every weekday Sign up to Business Today Get set for the working day – we'll point you to all the business news and analysis you need every morning Preview latest Enter your email Sign up after newsletter promotion

Liz Kendall, the science and technology secretary, said: “CuspAI is putting the power of AI to work to help us tackle some of the biggest challenges facing the UK and the world today, such as energy and climate change. By backing British companies such as this we are paving the way for breakthroughs that are set to grow our economy, protect the natural world and improve everyone’s lives.”

CuspAI has brought on board John Giannandrea, a former Apple and Google executive, who is helping set up its US operations.

The company is also working with Abhi Talwalkar, a board member at chip maker AMD and chair of Lam Research.

Jeff Bezos and UK government invest in £2bn British startup CuspAI

At the beginning of 2023, Keith Peiris's startup Tome, which specializes in AI-enhanced presentations, was experiencing rapid growth. It became the fastest productivity tool to achieve 1 million users—a milestone that eventually soared to 25 million, as reported by Forbes. With support from esteemed Silicon Valley investors, including Lightspeed Venture Partners, Coatue, Greylock, and billionaire Reid Hoffman, Peiris secured $80 million in funding.

However, by late 2024, Peiris recognized that he was heading in the wrong direction with his company.

The primary issue was that many users, primarily made up of students and small business owners, were either on free plans or subscribed to the low-cost $10 monthly option. This subsistence was insufficient for sustaining the business. Attempts to penetrate professional markets, such as marketers and salespeople, fell short as the tool lacked integration with their data and the context necessary for effective presentations. Although user numbers grew, the company’s annual revenue stagnated at merely $3 million.

Peiris candidly reflects on Tome, stating, “Our technology thesis and our cultural thesis was very immature.”

With nearly half of Tome’s investment still available and significant GPU resources at his disposal, Peiris opted for a pivot. “We were optimistic that we hadn’t missed our chance so early in our journey. We have a talented team and ample capital—why should we not give it another shot?” he shares.

Peiris sought guidance from billionaire Stewart Butterfield, whose previous failures in gaming had led him to create renowned platforms like Slack and Flickr. Butterfield's advice was to downsize the team to one that could be fed by two large pizzas, focusing on building something new with strong customer traction that would keep investors and staff engaged. Thus, in March 2025, Peiris shut down Tome, laying off most of his 70 employees and retaining just six.

After eight months, he introduced Lightfield, an AI tool designed to assist sales professionals with administrative tasks such as call summarization, follow-up email writing, and client interaction management.

Initially, Peiris' investors were skeptical. Developing customer relationship management software was inherently more challenging, and there were doubts about Peiris and his co-founder's expertise in this arena. Nonetheless, it represented a superior business opportunity. “There were animated discussions, but ultimately, everyone came around,” he notes.

Currently, Lightfield's revenue is witnessing an impressive growth rate of 80% monthly, supported by 1,000 paying customers, including Substack, Goodfire, and IntentHQ, according to Peiris. Remarkably, Lightfield has been built entirely from the funds initially raised for Tome, and discussions for another funding round are underway—with previous investors showing renewed interest.

Pivoting is a fundamental aspect of entrepreneurship. When an initial concept falters, entrepreneurs often seek new directions. This phenomenon typically happens before a startup gains significant traction or investor backing. However, Lightfield is a part of the emerging wave of AI startups that secured substantial funding before moving away from their original goals. Some have undertaken sharp pivots, entirely discarding their initial products, while others have adapted to pursue new, profitable avenues. The rapid evolution in the AI sector presents additional challenges, as the pace of model improvements continuously reshapes the market landscape.

"As dominant models increasingly capture market attention, we’ll likely see many startups either pivot significantly or find value in elements of their existing infrastructure to carve out a new direction,” observes Aditya Agarwal, a general partner at South Park Commons, a venture capital firm collaborating with tech entrepreneurs.

For example, Pika raised $135 million to create an AI video generator but shifted focus to AI agents and avatars. Poolside raised $620 million for training AI coding models but, as of late 2025, planned to develop a large data center in West Texas named Project Horizon in partnership with CoreWeave—a venture that faltered after Poolside struggled to meet chip supply deadlines. Subsequently, Poolside restructured into two entities: one dedicated to infrastructure (PIC) and another focused on model building.

Character AI, another early AI success story, has undergone significant transformations since its inception. Founded by former Google DeepMind researchers Noam Shazeer and Daniel De Freitas in 2021, the startup raised around $200 million from prominent investors, aiming to craft AI characters inspired by real and fictional personalities. However, after substantial developments, including the acquisition of their founders by Google and a barrage of lawsuits tied to the app's safety, the company has since refined its business strategy. In early 2025, under new leadership, it shifted toward using open-source software and exploring novel monetization strategies, including advertising and in-app purchases.

We’ve refocused our efforts away from traditional fundraising and hypergrowth toward sustainable business development," explains Karandeep Anand, the new CEO.

Recognizing the need to diversify, the startup is now delving into various interactive entertainment formats, like AI-generated audio stories and microdramas, allowing users to engage with characters in more dynamic ways. Anand emphasized that rather than “pivoting,” it’s been more about evolving the existing concept to align with user needs.

Similarly, Patronus AI, an evaluation startup, adjusted its strategy to adapt to changing industry demands. Initially focused on developing models for error detection and copyright infringements, the company pivoted to create "digital world models"—replicas of internal systems that help AI agents complete tasks across various platforms. This shift now constitutes a significant portion of their revenue.

“Reinvention and reassessment of our purpose are critical for our growth,” asserts CEO Anand Kannappan.

For some startups, adapting was essential for survival. Wispr AI’s founders, Tanay Kothari and Sahaj Garg, experienced a significant setback after investing three years and $14 million in a neural-signal-driven headset that ultimately failed. Redirecting their efforts toward AI-driven voice dictation software, they transformed the company structure and strategy. Today, their Wispr Flow app boasts hundreds of thousands of daily users and a growth rate of 40% per month, marking a departure from previous, ambitious but infeasible projects. As reflected in their blog, “We ceased chasing the idealistic dream and instead focused on delivering what truly resonates with people.”

AI Startups Are Shifting Focus From Showy Demonstrations To Revenue-Generating Technology

A sunlit modern startup loft where two determined founders slide a colorful demo panel into a cardboard box labeled “Showcase,” while on a nearby wooden desk a laptop glows with a rising revenue graph, stacks of invoices and dollar‐marked papers bathed in warm morning light, whiteboard sketches and sticky notes hinting at the next product pivot.

AI Startups Are Pivoting From Flashy Demos To Tech That Pays The Bills - Forbes

- AI startup Tome grew rapidly but struggled to monetize its presentation tools effectively.  
- Keith Peiris pivoted to Lightfield, which aids salespeople and surged with 1,000 paying customers.  
- Other AI startups also pivoted for survival, adapting to evolving market demands and opportunities.  

I n early 2023, Keith Peiris’s AI-powered presentation startup Tome was growing quickly. It had become the fastest productivity tool to reach 1 million users, Forbes reported at the time, a number that eventually climbed to 25 million. He raised $80 million from top Silicon Valley investors like Lightspeed Venture Partners, Coatue, Greylock and billionaire Reid Hoffman.

But by late 2024, Peiris realized he was building the wrong company.

It turned out that most people didn’t want to pay for its tools, which generated beautiful slides within minutes. Tome’s users, mainly students and small business owners, were largely on free plans or paying $10 monthly subscriptions—not nearly enough to sustain the business. And the startup couldn’t quite crack the market for professionals like marketers and salespeople because it wasn’t connected to their data and did not have the context needed to make good presentations. Users were still growing, but the company’s annual revenue plateaued at around just $3 million.

“Being frank about Tome, our technology thesis and our cultural thesis was very immature,” Peiris admits now.

With about half of Tome’s funding still in the bank and lots of GPU capacity on reserve, Peiris decided to pivot. “We were looking at this being like, ‘There's no way that we've missed the boat in year two here. We've got a great team, a lot of capital. Why don't we try again?’” he says.

So Peiris turned to billionaire Stewart Butterfield— who had twice turned his failed attempts at starting gaming companies into iconic businesses like Slack and Flickr— for “great sage advice” on pivoting. Butterfield told him to start fresh with a team small enough that it could be fed with two large pizzas. More importantly, Butterfield said, Peiris needed enough customer traction on a new product to keep investors and employees too busy to talk about the old days. In March 2025, Peiris shut down Tome and laid off most of his 70 employees, leaving a team of six.

Eight months later, he launched Lightfield, AI software that helps salespeople with busywork like summarizing calls, writing follow up emails and keeping track of client interactions.

“We've got a great team, a lot of capital. Why don't we try again?” Keith Peiris, CEO of Lightfield

His investors weren’t thrilled at first. A customer relationship management software was inherently harder to build and VCs questioned Peiris and his cofounder’s experience in this space. But it was a better business opportunity. “I think that there was animated discussion, but I think eventually everyone came around to it,” he says.”

Now the company’s revenue is growing 80% every month thanks to 1,000 paying customers including Substack, Goodfire and IntentHQ, Peiris says. So far, he’s built Lightfield entirely using funding raised for Tome, but he’s now in talks to raise more money. His previous backers are all participating.

Pivoting is a classic part of entrepreneurship. You try something, it doesn’t work as well as you thought, so you try something else. But that usually happens before a company gains substantial traction and raises a significant amount of money. Not so in the AI funding frenzy. Lightfield is part of the first wave of AI startups that raised a ton of capital, only to move away from what they initially built. While some have made hard pivots that entirely scrapped the original product or vision, others have expanded their focus to newer, more lucrative areas. The fast-paced nature of AI, has made it even more challenging for startups to remain defensible: Models improve so fast that the market is constantly shifting.

"As frontier models end up taking more and more oxygen in the room, I think what you'll find is that a subset of these hot companies will either...do hard pivots or more likely they will find parts of infrastructure that they have built that they might want to spin out,” says Aditya Agarwal, a general partner at South Park Commons, a venture capital firm that works with tech founders.

Pika, for instance, raised $135 million to build an AI video generator but has evolved to building AI agents and avatars. Poolside raised $620 million to train AI coding models from scratch, but in late 2025 announced plans to construct a massive data center in West Texas called Project Horizon with CoreWeave — an effort that fell apart after Poolside wasn’t able to get chips online by CoreWeave’s deadline, the Financial Times reported. Poolside split itself into two companies, one focusing on infrastructure called PIC and another, to build models.

“I think it starts around a group of people that come together around a hypothesis that they may or may not reach,” says BCV partner Christina Melas-Kyriazi. “That plus capital has allowed these companies to take much more meandering paths than you would've otherwise seen in prior eras.”

Character AI, another early AI darling, has significantly transformed from the consumer chatbot app it first pitched to investors. Founded by Google Deepmind luminaries Noam Shazeer and Daniel De Freitas in 2021, the startup raised about $200 million from A-list backers like Andreessen Horowitz to generate AI characters inspired by real and fictional people that anyone could chat with, powered by its own large language models. The goal was to eventually “empower everyone with AGI,” Shazeer told Forbes in 2023. But in 2024, Google acqui-hired Shazeer and De Freitas and licensed Character’s technology for $2.7 billion. Then came a wave of lawsuits, which alleged that the company’s chatbots encouraged teens and minors to self harm, suicide and engage in sexually explicit conversations. In May 2025, board advisor and longtime Facebook executive Karandeep Anand took over as CEO.

That all seems to be distant history for the startup now. Character AI settled most of those lawsuits in January and banned under 18 users from chatting with characters on the app— a decision that cost it 4 million monthly active users. Instead of training models, it largely uses open source software. Instead of raising venture capital to release models or carry out research, the company is now fully bootstrapped and owned by its employees. Character AI has also been experimenting with new ways to monetize its products such as advertising and in-app purchases.

“A lot of the energy we've been focusing on is how do we not do the traditional ‘raise, raise, raise, raise money, grow, grow, grow.’ And we one day figure out whether this is a real business or not,’” Anand says. “Sustainable growth while also monetizing is very important to me.”

“We stopped chasing the sci-fi dream and started building what people truly needed.” Tanay Kothari, CEO of Wispr Flow

It’s now aggressively pushing into other non-text based forms of interactive entertainment, including releasing AI-generated audio stories, comics and microdramas, which are series of short video episodes produced with the help of AI. People can also chat with the characters featured in the stories. “Our goal has been on just expanding the ways people can experience this because not everyone is an avid typer,” Anand says.

But Anand says “pivot” is too strong of a word for the changes. “I would say this is more of an evolution,” he says.

That’s also how AI evaluation startup Patronus AI thinks of its recent change in strategy. The San Francisco-based startup got its start developing models to catch factual inaccuracies and flag responses that violated copyright, raising 20 million in funding from marquee backers like Lightspeed Venture Partners. Founded by Forbes 30 Under 30 alumni and ex-Meta AI researchers Anand Kannappan and Rebecca Qian, the San Francisco-based startup also built benchmarks to test how well AI models performed on financial, medical and legal tasks. But as the industry shifted towards AI agents, the startup began training so-called “digital world models” — replicas of internal systems and websites to test how well agents can complete tasks like ordering Doordash or booking a flight ticket. Patronus's simulations help improve agents by rewarding accuracy and punishing errors.

Patronus AI cofounders Anand Kannappan (left) and Rebecca Qian (right) Patronus AI

It now calls itself a “frontier lab.” Today, these replicas account for 70% of its revenue, says CEO Kannappan, attracting top AI firms as customers. After releasing the new product, Patronus closed a $50 million funding round at a $400 million valuation. “I think it's extremely important for us to continue to reinvent and rethink exactly what we do,” he says.

For others, pivoting was crucial for survival. Wispr AI cofounders and Stanford graduates Tanay Kothari and Sahaj Garg spent six months building a pair of headphones that converted a person’s neural signals and silent speech into text and voice. Three years and $14 million in funding later, the duo finally decided that it wouldn’t work. Plus they were scared off by major, embarrassing failures in consumer AI hardware, like Humane’s AI pin. In 2024, they decided to kill the hardware device, laid off most of their 40 employees and refocused on the AI-powered voice dictation software they had built for the headphones. “We had a 22,000 square foot office with five people in it,” CEO Kothari wrote in a blog post detailing the pivot.

Today their AI voice dictation app Wispr Flow has hundreds of thousands of daily active users and is growing 40% month over month. “We stopped chasing the sci-fi dream and started building what people truly needed,” the blog reads.





More From Forbes

Forbes The World’s Largest Fast-Food Franchisee Is Still Hungry For More

Forbes Can’t Get A Piece Of OpenAI Or Anthropic? There’s A Booming Secondary Market For Their Swag

Forbes How Christopher Nolan Became The Biggest Movie Star In Hollywood

Forbes Israel’s Palantir Rival Is Selling $1 Million Spy Vans To U.S. Cops

AI Startups Are Pivoting From Flashy Demos To Tech That Pays The Bills

Artificial intelligence is transforming a sector in Silicon Valley that was once overlooked, turning it into a thriving area for investment. Venture capitalists are keenly investing in legal technology startups, eager to identify the next major player in the industry.

Historically, the legal tech space was viewed skeptically by investors due to its perceived limitations; the market was deemed too niche, with a reputation for sluggish clientele and protracted sales processes. 

The introduction of ChatGPT has dramatically altered this landscape. Where the sales process for legal tech could previously span several years, it has now been reduced to mere weeks as entrepreneurs scramble to harness advanced AI technologies to take on tasks traditionally performed by lawyers.

From San Francisco to London, investors are eager to fill a significant gap in their portfolios marked by the absence of substantial legal tech ventures. According to Crunchbase, global funding for legal startups soared to $2.1 billion in the first half of the year.

In light of this surge, Business Insider surveyed a selection of venture capitalists to highlight the most promising legal startups valued at under $1 billion this year. Here are their selections.

13 legal startups investors are keeping an eye on for 2026

A sunlit, high-end boardroom with large windows overlooking a city skyline, natural morning light washing over a polished mahogany table. Thirteen sleek leather portfolios, each labeled with a minimalist gold number, are arranged in a circle. In the center sits a gleaming brass scale of justice beside stacked law books. Well-dressed investors’ hands hover over the portfolios, poised to pick the next big legal innovators of 2026.

13 legal startups to watch in 2026, according to investors - Business Insider

- AI has transformed legal tech from a niche market to a booming investment sector.  
- Investors are drawn to legal startups aiming to innovate and reduce lengthy sales cycles.  
- Global funding for legal startups hit $2.1 billion in the first half of the year.

Artificial intelligence has turned one of Silicon Valley's least fashionable sectors into one of its hottest. Investors are pouring money into legal startups, hoping to find the next breakout winner before everyone else does.

For a long time, building software for lawyers felt like a contrarian bet. Investors saw legal tech as too small a category, with slow-moving customers and long sales cycles.

ChatGPT changed the equation. Sales cycles that once stretched for years began collapsing into weeks, as founders raced to apply frontier models to take on work done by attorneys.

Investors from San Francisco to London are now looking to fill a Harvey-sized hole in their portfolios. Funding to legal startups worldwide reached $2.1 billion in the first half of the year, per Crunchbase data.

Business Insider asked a group of venture capitalists to nominate the year's most promising legal startups valued at less than $1 billion. Here are their picks.

13 legal startups to watch in 2026, according to investors

In a recent regulatory submission, Netflix disclosed that it invested $587 million in cash to acquire InterPositive, a startup co-founded by acclaimed actor and director Ben Affleck.

The streaming giant first made the acquisition public in March, with Affleck expressing a desire to “protect the power of human creativity.” He explained that InterPositive’s AI technologies are designed to assist filmmakers in enhancing their footage during post-production, addressing common production hurdles such as missing scenes, background alterations, and lighting discrepancies.

At the time of the announcement, Netflix confirmed that the entire team from InterPositive would be integrated into the company, with Affleck stepping in as a senior advisor. However, specific financial details of the transaction were initially withheld. A later report from Bloomberg estimated the deal's value could be as high as $600 million.

In its latest earnings update, Netflix mentioned that approximately 300 of its titles have already incorporated generative AI technology.

Netflix acquired Ben Affleck's AI filmmaking startup for $587 million.

A warm, golden-hour Hollywood soundstage lit by natural sunlight. Front and center, a sleek AI-driven robotic arm holds a gleaming clapperboard stamped “$587 M.” Nearby sits a director’s chair labeled “Affleck.” Above it all, the Netflix logo looms large against a backdrop of film cameras, lights, and a subtle digital wireframe overlay hinting at AI innovation.

Netflix paid $587M for Ben Affleck’s AI filmmaking startup - TechCrunch

- Netflix acquired InterPositive, co-founded by Ben Affleck, for $587 million in cash.  
- InterPositive's AI tools assist filmmakers with post-production challenges like missing shots.  
- Netflix reported that approximately 300 titles have utilized generative AI technology.  

In Brief

In a new regulatory filing, Netflix revealed that it paid $587 million in cash for InterPositive, a startup co-founded by actor and director Ben Affleck.

The streaming company announced the acquisition in March, with a statement from Affleck saying he wanted to “protect the power of human creativity.” According to Affleck, InterPublic’s AI tools help filmmakers improve their footage in post-production, particularly when it comes to making up for “real-world production challenges such as missing shots, background replacements or incorrect lighting.”

At the time, Netflix announced that the entire InterPositive team would be joining the company, with Affleck joining as a senior advisor, but it didn’t disclose the financial terms of the deal. A subsequent report in Bloomberg suggested that the deal could be worth up to $600 million.

In its most recent earnings report, Netflix said that around 300 of its titles have already used generative AI.

Netflix paid $587M for Ben Affleck’s AI filmmaking startup

A notable figure in the AI landscape making headlines is not an American tech magnate like Elon Musk or Sam Altman but rather Yang Zhilin, a 34-year-old researcher and entrepreneur. Yang is at the helm of Moonshot AI, a Chinese firm that recently unveiled its innovative open-weight AI model, Kimi K3, which has captured attention in Silicon Valley.

The Kimi K3 model stands out as China’s latest advanced open-weight AI, boasting coding and agentic features that compete with leading models from American companies such as OpenAI and Anthropic, all while being significantly more affordable. This development signifies a remarkable advancement for China's position in the global AI race.

Born in 1992 in Shantou, Guangdong province, Yang Zhilin pursued his education at Tsinghua University before obtaining a Ph.D. at Carnegie Mellon University in Pennsylvania. Here, he studied under respected AI figures like Ruslan Salakhutdinov and William Cohen. Salakhutdinov praised Yang's brilliance in a recent post on X, noting that, despite attractive offers from major tech firms, Yang was passionate about launching his own enterprise.

During his time at Carnegie Mellon, Yang interned with both Google Brain and Meta, contributing to several research papers addressing various aspects of language models. After his academic pursuits, Yang returned to China and played a significant role in key AI initiatives, which included Huawei's PanGu model and the large-scale multimodal AI project, Wu Dao, developed by the Beijing Academy of Artificial Intelligence. He was also a cofounder of Recurrent AI, a startup leveraging AI to evaluate sales conversations to enhance business outcomes.

When it comes to his approach to AI development, Yang emphasizes the importance of scale. He articulated, "If you can solve it with scale, don't solve it with a new algorithm. The new algorithm's value is to enable better scaling," during an interview with journalist Xiaojun Zhang.

Yang co-founded Moonshot AI in early 2023, alongside a team experienced in developing cutting-edge AI technologies like Transformer-XL and ShuffleNet. Moonshot AI first gained recognition through its Kimi K2 foundation model, which featured an impressive trillion-parameter context window capable of processing extensive documents. The company has expanded its offerings to include an AI assistant named Kimi, alongside advancements in coding, research, and autonomous agents, gaining support from major Chinese backers such as Alibaba and Tencent.

The recent Kimi K3 launch has propelled both Yang and Moonshot AI into the global spotlight. Vercel CEO Guillermo Rauch noted that this model positioned itself "ahead of all proprietary ones for this comprehensive web engineering benchmark," while Wharton professor Ethan Mollick referred to it as “closest to the frontier yet.”

In an interview, Yang expressed his belief in the future of artificial general intelligence (AGI), suggesting that the next leading company in this domain could overshadow current giants with significantly larger scales.

In the wake of Kimi K3's announcement, some prominent figures in technology voiced regret over Yang's choice to return to China instead of continuing his work in the U.S. Legendary investor Vinod Khosla highlighted restrictive immigration policies instated during the Trump administration as a critical factor in this trend, emphasizing that the U.S. risks losing exceptional talent.

Yang’s decision to leave the U.S. has been linked to the complicated and uncertain nature of the American immigration system. Although some suggested that these policies may deter brilliant individuals from remaining in the country, Salakhutdinov, Yang's former advisor, highlighted that Yang had always envisioned establishing his own startup back in China. He noted Yang's determination not to let the opportunity to launch his own venture slip away, a sentiment he respects and agrees with wholeheartedly.

Introducing Yang Zhilin, the founder and CEO of China's trending Kimi K3 AI model.

A confident Yang Zhilin standing in a sunlit, minimalist innovation lab, arms lightly folded beside a transparent OLED screen labeled “KIMI K3” with warm orange data streams flowing across it; natural daylight floods through floor-to-ceiling windows onto a sleek desk where a small Chinese flag stands, modern servers and polished equipment softly receding into the bright background—an upbeat, forward-looking tech portrait.

Meet Yang Zhilin, the CEO and founder behind China's buzzy new Kimi K3 AI model - Business Insider

- Yang Zhilin, founder of Moonshot AI, unveiled the Kimi K3 AI model this week.  
- Kimi K3 competes with American AI models while being significantly more affordable.  
- Yang's return to China was driven by a desire to launch his own startup.  

There's a founder making waves in the AI industry that's not an American tech billionaire like Elon Musk, Sam Altman, or Dario Amodei.

Yang Zhilin is the 34-year-old researcher and entrepreneur behind the Chinese company Moonshot AI and its new open-weight Kimi K3 AI model, which seized Silicon Valley's attention this week.

Kimi K3 is the latest buzzy open-weight model from China. Its coding and agentic capabilities rival those of leading American models from labs like OpenAI and Anthropic and come at a much lower cost, marking a new milestone for China in the AI race.

Who is Yang Zhilin?

Yang was born in 1992 in Shantou, a city in China's Guangdong province. He attended Tsinghua University before earning a Ph.D. at Carnegie Mellon University in Pennsylvania, where he studied under prominent AI researchers Ruslan Salakhutdinov and William Cohen.

"He is absolutely brilliant," Salakhutdinov said of Yang in a post on X on Friday. In another post, Salakhutdinov said that after his graduation, Yang was heavily recruited by Big Tech but was committed to starting his own company.

Related video LinkedIn marketing leader Jessica Jensen said flexibility will help companies attract and retain top talent

While at Carnegie Mellon, Yang interned at Google Brain and Meta. He's also coauthored several research papers on topics ranging from limitations in how language models handle context to prompt-tuning advice.

Before launching Moonshot AI, Yang returned to China and contributed to several major AI projects, including Huawei's PanGu model and the Beijing Academy of Artificial Intelligence's Wu Dao, a large-scale multimodal AI model. He also cofounded Recurrent AI, a startup that used artificial intelligence to analyze sales conversations and help companies improve their performance.

One of Yang's principles in building AI is to lean toward scale.

"If you can solve it with scale, don't solve it with a new algorithm. The new algorithm's value is to enable better scaling," he told journalist Xiaojun Zhang in an interview published on LinkedIn.

He helped start Moonshot AI in early 2023. The team behind the company had developed multiple AI technologies, including Transformer-XL, RoPE, Group Normalization, ShuffleNet, MuonClip, and Mooncake.

Moonshot AI's first claim to fame was giving its Kimi K2 foundation model an unusually large context window — a trillion parameters — to process lengthy documents. Moonshot also has an AI assistant product, Kimi, and has since expanded into coding, research, and autonomous AI agents, attracting backing from Alibaba, Tencent, and other major Chinese investors.

Kimi K3 has now catapulted both Moonshot AI and its founder onto the global stage.

Vercel CEO Guillermo Rauch said Kimi K3 marked "the first time that an open model is ahead of all proprietary ones for this comprehensive web engineering benchmark," in a post on X. While Wharton professor Ethan Mollick called it "closest to the frontier yet."

"The ultimate AGI company will dwarf today's giants — double, triple the scale. Not necessarily OpenAI, but such a company will exist," Yang said in his interview with Zhang.

Why Yan Zhilin left the US

Some leading tech figures lamented after Kimi K3's release this week that Yang didn't remain in the United States to work at an American AI lab. Legendary billionaire investor Vinod Khosla blamed the Trump administration's restrictive immigration policies.

"Even bigger issue is the brilliant talent we are scaring away from other countries with our immigration policies for great talent," he said in response to the news of Kimi K3's success.

The Trump administration has moved to tighten immigration restrictions, including for student visas. Last year, the government introduced a $100,000 fee for employers sponsoring some new H-1B applications for foreign workers. The ruling was later struck down by a federal judge and remains in litigation.

Then, in May, a US Citizenship and Immigration Services memo implied that people who could previously apply for a green card from inside the US may now have to leave the country while their case is being processed.

And this week, the administration introduced a new rule that puts an expiration date on how long people on student visas can initially stay in the US.

Salakhutdinov, however, said that while the US immigration process can be intimidating and "uncertain," Yang was always set on going back to China and building his own startup.

"I remember him telling me that if he didn't at least try starting his own company, he would regret it for the rest of his life. I respect that, and he was right," Salakhutdinov wrote on X.

Meet Yang Zhilin, the CEO and founder behind China's buzzy new Kimi K3 AI model

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

Instacart, the prominent grocery technology leader in the U.S., is set to acquire Israeli startup Arpalus, a firm known for its innovative artificial intelligence and computer vision solutions aimed at enhancing visibility on supermarket shelves. While the financial terms of the acquisition remain undisclosed, industry estimates suggest the deal is valued in the tens of millions. This acquisition represents Instacart’s first venture into the Israeli tech market.

Arpalus tackles a significant challenge prevalent in online grocery shopping: the discrepancy between the inventory that retailers present digitally and what is truly available on store shelves. Inaccurate or missing inventory data often frustrates customers, leading to substitutions, canceled orders, and waning trust in online grocery services.

Established in 2019 by Ofir Zilberberg, who has a background in the Israeli Air Force and Elbit Systems, alongside Dr. Rotem Bennet, an expert in computer vision formerly associated with Microsoft Research, Arpalus has developed sophisticated AI models that analyze supermarket shelves in real time. The startup, which emerged from the Nielsen Innovate incubator, operates out of Netanya and employs around 20 individuals.

Since its inception, Arpalus has successfully raised about $5 million from various investors, including the Israel Innovation Authority, True Global Ventures, Duo Partners Fund, and the Techstars Accelerator, as well as private stakeholders from the retail sector.

The core technology of Arpalus is capable of transforming a brief video scan of a store shelf into a comprehensive digital overview of available products. Designed to tackle the unique challenges faced in grocery retail, such as variable lighting, unreliable Wi-Fi, and densely packed merchandise, Arpalus claims its system can recognize individual items with an impressive 95% accuracy on average.

This technology is compatible with standard smartphones and other camera-equipped devices. Post-acquisition, Instacart aims to leverage its network of around 600,000 shoppers to scale shelf intelligence across the platform. While fulfilling customer orders, these shoppers will have the capability to scan the shelves via the Instacart app, thereby providing a more precise and continually updated inventory of product availability.

The acquired data will be incorporated into Instacart’s extensive retail technology framework, enhancing the reliability of online grocery orders while also bolstering in-store solutions like smart carts and computer vision applications.

Instacart has stated that Arpalus’ technology will enhance its existing retail intelligence capabilities, which rely on insights garnered from over 1.6 billion orders and real-time inventory data from nearly 100,000 stores across North America, in addition to more than 10 million daily data points generated from shopper activities.

This acquisition reflects Instacart's strategic transition from being primarily a grocery delivery service to a tech-driven retail company focused on artificial intelligence, data analytics, and automation. The company has been investing in solutions that bridge the gap between physical stores and the digital shopping experience, as evidenced by its innovations like Caper smart carts and the Store View technology that helps retailers track shelf availability.

Founded in 2012 and based in San Francisco, Instacart went public on the Nasdaq in 2023 and currently holds a market valuation of around $11 billion. It connects approximately 1,800 retail chains with hundreds of thousands of shoppers and consumers throughout the United States and Canada.

Instacart purchases Israeli AI company Arpalus to enhance real-time shelf intelligence

A bright, sunlit grocery aisle bathed in warm morning light, packed shelves neatly organized. Center frame: a sleek digital scanner mounted on the shelf edge displaying live stock counts in green accents. In the foreground, a hand holds a tablet screen boldly featuring the Instacart logo, scrolling through inventory data. A small Israeli flag sticker is affixed to the scanner device, hinting at the Arpalus connection.

Instacart acquires Israeli AI startup Arpalus to bring real-time shelf intelligence to supermarkets - calcalistech.com

- Instacart acquires Israeli startup Arpalus to enhance grocery inventory visibility and accuracy.  
- Arpalus developed AI technology for real-time analysis of supermarket shelf availability.  
- The acquisition supports Instacart's shift towards retail technology utilizing AI and data integration.  

The American grocery technology giant Instacart is acquiring Israeli startup Arpalus, which develops artificial intelligence and computer vision technology designed to provide real-time visibility into supermarket shelves. The value of the transaction was not disclosed, but according to Calcalist’s estimate, it is in the tens of millions of dollars. The deal marks Instacart’s first acquisition of an Israeli company.

Arpalus’ technology addresses one of the biggest challenges facing online grocery shopping: the gap between what retailers show as available digitally and what is actually sitting on store shelves. Missing or inaccurate inventory information is a major source of customer frustration, leading to product substitutions, cancelled orders and declining trust in online grocery platforms.

1 View gallery Instacart. ( Photo: Courtesy )

Founded in 2019 by Ofir Zilberberg, who serves as CEO and previously worked at the Israeli Air Force and Elbit Systems, and Dr. Rotem Bennet, a computer vision expert and former Microsoft Research employee who has since left the company, Arpalus developed AI models that analyze supermarket shelves in real time. The company emerged from the Nielsen Innovate incubator, is headquartered in Netanya and employs approximately 20 people.

Arpalus has raised around $5 million since its founding from investors including the Israel Innovation Authority, True Global Ventures, Duo Partners Fund, Techstars Accelerator, Noam Lanir, the American Playhouse fund and private investors from the retail industry.

The company’s technology converts a short video scan of a store shelf into a digital view of available products. Built specifically for the difficult conditions of grocery retail, including changing lighting, unreliable Wi-Fi connections and thousands of similar-looking products packed tightly together, Arpalus says its system can identify individual items with more than 95% accuracy on average.

The technology runs on standard smartphones and other camera-equipped devices. Following the acquisition, Instacart plans to use its network of approximately 600,000 shoppers to generate shelf intelligence at scale. While completing customer orders, shoppers will be able to scan shelves through the Instacart app, creating a more accurate and continuously updated picture of product availability.

The data will be integrated into Instacart’s broader retail technology platform, improving the accuracy of online grocery orders while also supporting in-store solutions such as smart carts and computer vision tools.

Instacart said Arpalus’ technology will complement its existing retail intelligence infrastructure, which is built on more than 1.6 billion lifetime orders, real-time inventory information from nearly 100,000 stores across North America and more than 10 million daily data points generated from shopper activity.

The acquisition is part of Instacart’s broader shift from a grocery delivery marketplace into a retail technology company focused on artificial intelligence, data and automation. The company has invested in technologies designed to connect physical stores with digital commerce, including its Caper smart carts and Store View, a computer vision product that helps retailers monitor shelf availability.

Founded in 2012 and headquartered in San Francisco, Instacart went public on Nasdaq in 2023 and currently has a market value of approximately $11 billion. The company connects roughly 1,800 retail chains, hundreds of thousands of shoppers and consumers across the US and Canada.

Instacart acquires Israeli AI startup Arpalus to bring real-time shelf intelligence t

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.

Jeff Bezos' family office invested in five AI startups in June.

More from AI Startups & Investments

OpenAI considers expanding in Israel by hiring employees and exploring a business center.

Inference startup Infinity secures $15M from Touring Capital along with researchers from OpenAI and Anthropic.

Jeff Bezos and the UK government invest £2bn in the British startup CuspAI.

AI Startups Are Shifting Focus From Showy Demonstrations To Revenue-Generating Technology