INTEL TRIED TO BUY THIS CHIP COMPANY FOR $1.6 BILLION. SEVEN MONTHS LATER IT IS WORTH $11 BILLION.
Source: TechCrunch — SambaNova raises $1B at $11B valuation, 5 months after last mega round
Intel tried to buy SambaNova last December at a valuation of roughly $1.6 billion. The deal fell through. Seven months later, SambaNova just raised $1 billion at an $11 billion valuation. Intel is now a minority investor in the company it could not figure out how to acquire. If there is a story that captures how fast the AI chip market is moving right now, it is probably this one.
SambaNova is a Palo Alto-based company that builds what it calls premium inference infrastructure. In plain English, that means it builds the hardware and software that lets large AI models run fast and at scale. The company’s edge, according to CEO Rodrigo Liang, is that it can fit multi-trillion-parameter models onto a single rack and run them quickly. As the frontier models from OpenAI, Anthropic, and Google keep growing, the infrastructure to run them efficiently becomes more valuable, not less.
The round was a Series F led by General Atlantic, with a second close expected in the coming weeks. Other investors include BlackRock, the Qatar Investment Authority, T. Rowe Price, Capital Group, Vista Equity Partners, and a long list of others that reads like someone cc’d their entire financial contacts list. Intel also participated, which remains one of the odder subplots in semiconductor investing: backing the company you once tried to swallow whole.
The really interesting detail is the JPMorgan partnership. JPMorganChase has chosen SambaNova as its inference-infrastructure partner, meaning its SN40L and SN50 chips will power AI inference inside the bank’s own walls rather than in the cloud. Liang says this is a signal to the entire banking industry: if JPMorgan is building private, on-premises AI infrastructure, others will follow. Banks deal in information that cannot live on a shared cloud server. The model has to come to the data, not the other way around.
SambaNova raised its Series E just five months ago and here it is raising its Series F already. That pace says something about demand. The company is using the new capital mostly to shore up its supply chain, securing the components and manufacturing capacity it needs to actually build and deliver hardware over the next 12 months. Supply chains for specialized AI chips are genuinely constrained right now. If you have orders and cannot fulfill them, the valuation is just a number on a slide.
The company says it is heading toward a public offering at some point, though the CEO was noncommittal on timing. The door to an acquisition remains open too, he said. At $11 billion, whoever wants to buy SambaNova better show up with a much bigger check than Intel was apparently ready to write back in December.
ALIBABA AND TENCENT BOTH INVESTED IN THE SAME AI VIDEO COMPANY. TENCENT ALSO SOLD $1.5 BILLION OF THE PARENT COMPANY STOCK. PICK A SIDE.
Source: CNBC — Kuaishou shares jump after Tencent joins $2.8 billion raise for Kling AI subsidiary
China’s two biggest tech rivals, Alibaba and Tencent, just backed the same company. The company is Kling AI, the generative video arm of Kuaishou Technology, and it just closed $2.8 billion in fresh funding. Both are listed as investors. These are companies that, in normal circumstances, are competing to eat each other alive. This week they apparently found common ground.
Kling AI makes AI-generated video. If you have seen a clip circulating online where someone’s face morphs seamlessly, or a product commercial is generated from a text prompt in about forty seconds, there is a reasonable chance Kling had something to do with it. The company’s parent Kuaishou is a short video platform that has spent years trying to establish itself as something more than China’s second-tier answer to TikTok. Spinning out Kling as a separate entity, raising $2.8 billion for it, and getting Alibaba and Tencent competing for a seat at the table is a pretty effective rebranding exercise.
The numbers are significant. The round values Kling AI at approximately $15 billion before the investment goes in, and the total raise could reach $3 billion if additional investors come in. The initial close brought in about $2 billion from investors including BlueFive Capital out of Abu Dhabi and Baidu, with Tencent’s $200 million arriving in a subsequent tranche.
Here is the detail that makes this interesting. Tencent invested $200 million into Kling AI while simultaneously moving to sell approximately $1.5 billion worth of Kuaishou stock. You could read that as Tencent saying: the parent is not the interesting play, but the AI subsidiary is. Either way, the move is deliberate and the timing is not an accident.
The underlying story is about China’s AI video race. Kling is competing with ByteDance’s Seedance, Alibaba’s own video models, and a growing list of other generators. The fact that Alibaba invested in Kling while running its own competing video AI products says something. When Alibaba backs a competitor’s product, the implicit bet is that the category is going to be much bigger than any single player can capture alone.
AI video generation is real and accelerating. The commercial applications in advertising, entertainment, and social media are substantial. The Chinese market for consumer video AI is among the largest on the planet. If you want to understand where the next wave of AI capital in Asia is flowing, watching who gets into Kling at what price is a reasonable place to start.
A SWEDISH STARTUP WITH 146 EMPLOYEES IS IN TALKS TO BE WORTH $13 BILLION. THE MATH IS INSANE AND ALSO POSSIBLY CORRECT.
Source: TechCrunch — Lovable reportedly in talks to double its valuation to $13.2B
Lovable is a Swedish company with 146 employees. It is in talks to raise money at a valuation of $13.2 billion. The company started less than three years ago. Its annual revenue run rate is currently around $500 million. You may need to read those sentences again and that is fine, go ahead.
The product is what the industry calls vibe coding. Instead of writing software code yourself, you describe what you want the software to do in plain language and an AI builds it for you. Lovable was one of the first companies to make this genuinely useful for non-developers, and it turns out a very large number of people have been waiting their whole lives for exactly that product. The company grew faster than almost any software startup in recent memory.
The numbers are extraordinary in their own right. The company reportedly added $100 million in new revenue in a single month earlier this year. With 146 employees, that works out to roughly $3.4 million in annual revenue per person. Even the most efficient software companies in the world tend to operate somewhere between $500,000 and $1 million revenue per employee. Lovable is running at three to seven times that ratio, which means either the headcount is going to grow substantially or the profit margins are something special.
Menlo Ventures is expected to lead the $300 million round. The new raise would exactly double the $6.6 billion valuation the company achieved last December, which itself came a few months after a $1.8 billion valuation in mid-2025. The company went from unicorn to decacorn territory in under eighteen months. The chart of its valuation over time looks like someone made a data entry error.
Enterprise customers include Workday, Asana, and Nvidia, which tells you this is not just a toy for hobbyists building weekend projects. Real companies with real budgets are paying Lovable to let their non-technical employees build internal tools and software prototypes without waiting for an engineering backlog to clear. That use case alone could justify the valuation if it scales across large organizations.
Whether $13 billion is the right number for a vibe coding company with 146 people is a reasonable thing to debate over a couple of drinks. Whether there is a real business here is much less debatable at this point. Half a billion in annual revenue from a company that basically did not exist three years ago tends to end that argument pretty quickly.
THE FIRST HUMANOID ROBOT COMPANY IS GOING TO WALL STREET. ITS CEO SAYS NOT TO EXPECT A ROBOT IN YOUR KITCHEN FOR AT LEAST 10 YEARS.
The first publicly traded humanoid robot company is almost here, and its CEO wants you to know upfront that she is not going to promise you a robot in your kitchen. Not for ten years at least. Which is either refreshing honesty or an unusual thing to lead with when you are trying to raise $620 million from public market investors, but here we are.
Agility Robotics announced it is going public via a SPAC merger with Churchill Capital Corp XI, valuing the company at around $2.5 billion. The deal still needs shareholder approval and SEC review, and is expected to close later this year. When it does, Agility will be the first pure-play humanoid robotics company trading on public markets, which means regular retail investors will finally be able to buy a piece of the robot industry directly without going through a VC fund.
The robot is called Digit. It stands about 5 feet 9 inches, weighs around 160 pounds, and it has reverse-bent knees that look like bird legs. The knees are not there to look interesting. They let the robot reach from floor level to high shelving without banging into warehouse racking. The hands have two thumbs and two fingers, optimized for gripping heavy plastic totes even when the contents shift in transit. Nothing about Digit is designed to seem humanoid for the sake of seeming humanoid. It is a warehouse worker, built to do warehouse work, in actual warehouses.
The customer list is real. GXO Logistics, Amazon, Toyota Motor Manufacturing Canada, Schaeffler, and Mercado Libre are already in the pipeline. The company has over $300 million in booked multi-year revenue representing roughly 1,000 robots operating on a subscription model where customers pay a monthly fee rather than buying the machines outright. That is a proper recurring revenue business, not a demo reel.
The investor base is notable. Amazon, SoftBank Vision Fund, Nvidia, and Foxconn all have stakes in Agility. Amazon being both a customer and an investor is the kind of alignment that either works extremely well or creates awkward moments when contract renewal time comes around. For now, everyone appears to be on the same page.
CEO Peggy Johnson previously helped engineer Microsoft’s $26 billion acquisition of LinkedIn and was the CEO of Magic Leap after that. She has seen both sides of the capital markets story up close. She says the company will avoid the post-SPAC volatility that destroyed many 2021-era companies by keeping its head down and executing robot by robot, customer by customer. At a $2.5 billion valuation compared to Figure AI’s $39 billion, Agility is the most conservatively priced bet in a sector that has completely lost perspective on scale. That is not the worst position to go public from.
THE COMPANY THAT TRAINS AI FOR A LIVING JUST ACQUIRED ANOTHER STARTUP — THREE MONTHS AFTER ITS OWN FOUNDER INVESTED IN THE TARGET. NOW IT IS IN TALKS AT $20 BILLION.
Source: TechCrunch — Mercor is in talks for a $20B valuation
Mercor is in talks to raise money at a $20 billion valuation. Nine months ago it was worth $10 billion. This week it also acquired a startup called Deeptune. The founder of Mercor personally invested in Deeptune just three months before his own company bought it. At $20 billion, Mercor would be worth more than most professional sports leagues combined. The week has been eventful for them.
Let us start with what Mercor actually does. The company runs a network of over five million domain experts, including lawyers, doctors, engineers, and software developers, who build training data for AI models. When a frontier lab like Anthropic or OpenAI needs to know whether an AI agent correctly used a spreadsheet, filled out a Salesforce form, or navigated a complex legal document, Mercor’s human experts build the tasks, run the agents through them, and produce the feedback that makes the models smarter. It is the part of AI training that nobody in consumer media talks about but that every serious lab depends on.
Deeptune does something complementary. It builds flight simulators for AI agents, digital replicas of enterprise software environments like Salesforce, Excel, and Slack, where AI agents can practice tasks in realistic simulations before they touch real production systems. The combination makes sense: Mercor has the human experts and the task-building pipeline, Deeptune has the simulated environments where those tasks run. Together they cover more of the AI training stack than either could cover alone.
The founder conflict is the detail worth sitting with. Brendan Foody, a Mercor founder, was listed as a personal angel investor in Deeptune’s $43 million Series A just three months before Mercor announced the acquisition. Financial terms of the deal were not disclosed. What you can observe is that a founder invested personally in a startup, and then his company acquired it three months later. The acquisition could be entirely justified on the merits, and probably is given how well the products fit together, but the timeline is the kind of thing that generates conversations in boardrooms.
The $20 billion talks arrive on top of last October’s Series C that valued Mercor at $10 billion. Doubling in nine months reflects either real momentum or a market that is still pricing AI training infrastructure very aggressively. The company reportedly runs at a $450 million annual revenue rate. At $20 billion, you are paying roughly 44 times forward revenue. That is a large multiple. But the five-million-person expert network is not something any competitor can build quickly.
The bigger signal is about where AI development is going. Labs are no longer just training text models on internet data. They are training agents to take complex actions in software environments, which requires realistic simulations and human experts to judge whether those actions were correct. Mercor is building infrastructure for both sides of that problem. The market, at $20 billion, is saying that is a very good place to be right now.
AN AI BUILT A LAW FIRM, RAISED $120 MILLION, AND GOT THE FORMER CHAIRMAN OF KIRKLAND AND ELLIS TO INVEST. THE BILLABLE HOUR IS STARTING TO NOTICE.
Source: TechCrunch — AI law startup Norm raises $120M, hits unicorn valuation
A startup just built an AI-powered law firm, raised $120 million at a $1.2 billion valuation, and got the former chairman of Kirkland and Ellis to write a personal check into it. Kirkland and Ellis billed over $7 billion in legal fees last year. When the guy who ran the machine starts backing the thing that is supposed to replace it, you pay attention.
Norm is not selling AI tools to lawyers. That story is two years old and mostly involves expensive subscriptions that partners use to summarize documents slightly faster while billing the same amount. Norm built an actual law firm called Norm Law, staffed with AI agents supervised by human attorneys, that provides legal services to enterprise clients. The billing model is the real innovation: instead of charging by the hour, which is how virtually every law firm in the world has operated since the profession existed, Norm charges based on outcomes. You pay for the result, not the time someone spent producing it.
The business case is obvious to anyone who has ever received a legal invoice and stared at it in quiet disbelief. Hourly billing exists because it benefits the firm, not the client. It survived this long because until recently there was no credible alternative if you needed serious legal work done at scale. AI is creating that alternative, and Norm is among the first companies to build an actual law practice around it rather than just a software product that law firms can ignore.
Khosla Ventures led the round. Other investors include Bain, Coatue, Craft Ventures, Vanguard, New York Life, and TIAA. That last pair, Vanguard and TIAA, are institutions managing trillions in assets. They are not typically writing checks into legal tech startups. Their presence here suggests this is being treated less as a technology bet and more as a financial industry infrastructure play, the kind of investment you make when you believe a fundamental operating cost across the entire economy is about to drop significantly.
Jeff Hammes, the former chairman of Kirkland and Ellis, is personally backing Norm. That is roughly equivalent to the former head of Blockbuster showing up to invest in Netflix in 2005. The man who ran one of the most powerful law firms on the planet thinks this startup is credible enough to put his reputation and his money behind it. That is a signal that is hard to dismiss regardless of what you think about the underlying technology.
The company has raised over $260 million in total funding across its roughly three-year life. At $1.2 billion, the valuation is modest by current AI startup standards, which means either there is substantially more upside here than the market has priced yet, or legal services turns out to be a harder market to crack than these investors believe. Given who those investors are, the smart money appears to be betting on the former.