TheChek News Flow

Gapping up
In reaction to earnings/guidance:

Other news:

Nvidia has agreed to buy Hugging Face, a company known for its GitHub-like repository of open-source AI models, for $12.9 billion, according to a person with knowledge of the agreement.

The move will put Nvidia in charge of a strategic asset in the race among open-source model developers to catch up to closed-source AI models from Anthropic and OpenAI. Nvidia leaders believe that a plethora of successful open models will help preserve its AI hardware dominance because they will be a counterweight to the closed-source AI developers, each of which is trying to develop competing AI server chips and lessen their reliance on Nvidia’s.

For that reason, Nvidia has been boosting its own effort to develop Nemotron-branded open-source AI models and has committed to spend tens of billions developing its own open models.

Other AI server chip makers such as Google, Microsoft, Amazon previously invested in Hugging Face, as did Nvidia.

Deal talks between Hugging Face and Nvidia began after the startup received acquisition interest from another suitor. Hugging Face had several potential suitors among its investors, including Salesforce, said a person with knowledge of the interest.

Besides storing models, Hugging Face helps AI app developers find and download audio, visual and text-based AI models and optimize them to run on different server chips. It also offers cloud services to power the models in AI applications.

Despite being the dominant repository of open models, the 10-year-old Hugging Face isn’t generating a lot of revenue. It recently was generating $150 million in annualized revenue, typically a figure that multiplies the prior month’s revenue by 12. That means Nvidia is paying about 80 times the forward revenue of the startup, which is high by today’s standards.

But Hugging Face revenue has grown relatively quickly of late; it was generating about $100 million in annualized revenue a couple of months ago. Interest in open source AI has surged as the capabilities of Chinese open-source models from Z.ai, Moonshot and DeepSeek improved to the point where more American businesses said they are using them.

Hugging Face customers pay flat-fee monthly subscriptions and added fees tied to the amount of computing power they use and data they store. In June, co-founder and CEO Clem Delangue told The Information that the startup had doubled its number of paying subscribers in the first half of 2026. Delangue said on a recent podcast appearance that the company is “close to profitability.”

Nvidia, the Cloud Provider?

Nvidia, by buying Hugging Face, would in effect be resuscitating its AI cloud business, a business it backed away from a year ago. The business, DGX Cloud, had put Nvidia in quasi-competition with Amazon Web Services. Nvidia had committed to spend $13 billion to rent back its own AI chips from bigger cloud providers that had purchased them, including AWS. Nvidia has then turned around and rented out some of those chips to companies such as Amgen and ServiceNow that develop their own AI applications. Nvidia believed the cloud business could one day help it generate $150 billion in revenue.

Hugging Face could also serve as a contingency plan. Nvidia said during earnings that it had committed to $36 billion in cloud services contracts as part of credit support agreements with its customers. If market conditions shift and customers aren’t able to lease that capacity out to third parties, Nvidia would end up responsible for some of that compute. And if Nvidia finds itself with more compute than its own engineers can use, it could use the capacity to power AI for Hugging Face customers while expanding that business.

The Hugging Face deal caps a frenetic series of investments for Nvidia CEO Jensen Huang in Nvidia customers and users of its chips and data center firms that work with Nvidia customers, in order to lock in the company’s hardware bundle. Nvidia has put tens of billions of dollars in everything from AI application developers such as Perplexity and OpenAI to companies that develop land and power for data centers.

Last week Nvidia raised eyebrows by signing a $6 billion deal to license AI development tools from a Poolside, which develops open source AI models, and made offers to hire more than 100 of its employees. Nvidia earlier this summer acquired Kumo AI and Essential AI to help develop model training tools as well.

Nvidia disclosed Wednesday that it holds $99 billion in equity investments and had as of the end of July committed to invest another $25 billion. Nvidia previously committed to make a $30 billion equity investment in OpenAI, and it also disclosed $108 billion in credit support on behalf of OpenAI, which plans to lease a large data center facility in Ohio, filled with Nvidia chips, likely starting in 2028.

Kress said that while “some will call circular financing, we see it differently” because Nvidia is aiding companies that will “become the largest technology companies in history.”

Customers Launch Their Own Chips

She said that a quarter of Nvidia’s revenues next year would be related to AI labs that Nvidia is supporting financially.

That could be problematic if OpenAI as well as Anthropic end up succeeding in their efforts to develop their own AI server chips while dominating the AI model and application market. OpenAI on Tuesday released an evaluation of its new Jalapeño AI server chip, claiming it is faster and more powerful than Nvidia’s flagship Blackwell chips. But the chips haven’t yet been tested as thoroughly against Nvidia’s latest chip, Rubin, nor have they been used at scale.

Huang told CNBC Wednesday that he wasn’t worried about such efforts and had confidence that Nvidia’s technology would be the most practical hardware for AI developers, no matter the alternatives.

However, Nvidia faces a more serious rival in Google, which has been trying to sway Nvidia’s most loyal customers to use Google’s tensor processing units in addition to Nvidia’s hardware.

Early investors in Hugging Face include Ronny Conway, Lux Capital and Addition.

Business Insider earlier reported that Nvidia was in talks with Hugging Face over a potential merger.

Hugging Face gained more notoriety last month when OpenAI said its AI broke out of an internal testing software and hacked not only OpenAI’s systems but also those of Hugging Face and other firms. Hugging Face said it used AI, including open source models, to combat the hack.

The latest development will come as a welcome signal about the business model ahead of its possible IPO

DeepSeek is set to soon secure new funding that would value the Chinese artificial-intelligence startup at $74 billion, building its war chest as it targets an initial public offering next year.

The Hangzhou-based company is seeking to raise $7.4 billion to bankroll research and development, and its expansion into computing infrastructure, according to people familiar with the matter.

The fundraising would mark a step-up from DeepSeek’s previous valuation in June, when investors including Tencent valued the startup at more than $50 billion, The Wall Street Journal reported previously.

In the latest round, existing shareholders—including investment firms Monolith Management and Shixiang, as well as Chinese battery giant Contemporary Amperex Technology—are set to inject new capital into the AI startup, the people said.

Several funds backed by Chinese local government vehicles are also planning to invest, some of the people said.

While DeepSeek told most investors in its previous funding round that they had to put money into a limited partnership managed by founder Liang Wenfeng, the company will allow more direct investment in the latest round, some of the people said.

DeepSeek’s annual recurring revenue—a key metric for subscription-based services—has reached $500 million, according to people familiar with its financials.

That will come as a welcome signal about the company’s business model ahead of its possible IPO. DeepSeek has hired banks and advisers in preparation of a planned listing in Shanghai next year, The Wall Street Journal has reported.

DeepSeek’s efforts are part of the broader rush across China’s AI sector to raise capital.

AI model developer Moonshot AI is wrapping up a fundraising round that values the company at around $50 billion ahead of a potential IPO in Hong Kong, people familiar with the matter said. Earlier this week, Alibaba raised $10.2 billion in a new share placement to invest in AI.

DeepSeek attracted global attention in early 2025 with a model that rivals leading American AI systems at a fraction of the cost. It remains among the cheapest AI model providers even after it raised prices on its V4 models earlier this month.

South China Morning Post and The Information earlier reported DeepSeek’s latest fundraising efforts.

Or possibly more than one galaxy, in order to “make his mark on the universe”, as 24-year-old Aschenbrenner has apparently said at several parties. The media coverage does not record how he might have dealt with follow-up questions like “what would you do with one if you owned it?”, “do you know what a light year is, and how long it takes to travel 25,000 of them” or even “there’s already a website that lets you do that, it only costs $19.95”.

We have all met people like this; they seem to be particularly drawn to the investment banking industry as well as to tech. People who are very clearly, very obviously incredibly intelligent – Aschenbrenner went to Columbia University at the age of 15, graduated as valedictorian and was described by professors as “one of the most brilliant undergraduates they ever met”.

But, people whose intelligence and original thinking seem to regularly lead them to do and say things which are … kind of dumb. Not just dumb in the “ignoring both the legal reality of ownership and general relativity” kind of way, but also dumb in the “making horrible career moves” kind of way. Aschenbrenner seems to have been one of those.

One might have said that it wasn’t really his fault that his first job out of university was with FTX – plenty of otherwise street-smart people were also taken in by Sam Bankman-Fried, and he got out early (apparently after an unrelated argument with Sam). But when he walked into his next job, at OpenAI, it didn’t take him long to get involved in the internal drama, pick the wrong side and get fired (with a cloud of accusations and counter-accusations about sensitive company information).

And then he sat down and wrote a 160 page blog post, called “Situational Awareness”, which was so good that people gave him, at the age of 24, his own $45bn hedge fund to manage. Which he then levered up by a factor of three, “hedging” his long positions in AI stocks with short positions in stocks that he thought would be vulnerable to AI. In other words, doubling up the same investment thesis rather than de-risking it. Which brings us roughly up to date.

People like this are fascinating company, and they will never be poor for long, because they keep getting job offers. But the lack of common sense can make them somewhat exhausting to be around, particularly if any of your own financial security or wellbeing depends on their judgement. If you ever wonder why investment banking is such a heavily regulated industry, guys like this are a big part of the reason.

Elsewhere, when life hands you lemons, make lemonade. Or, as they might have said in the Luxembourg office of Gazprombank, when life hands you sanctions, make sanctionsade.

It was quite a simple trade; in the aftermath of the invasion of Ukraine, a lot of Gazprom’s foreign currency debt was “stranded”, in the sense that although the company wasn’t subject to sanctions itself, most of the financial institutions involved in handling its interest payments were, so it became extremely difficult for non-Russian holders to get paid. For this reason, the bonds traded at a hefty discount.

Vladimir Putin’s government decided to solve the problem by allowing them all to be exchanged for new rouble-denominated bonds, held in Russia at full face value. So if you were a Russian citizen who was able to predict which bonds were about to be exchanged, and you had an employer who was willing to lend you lots of money, you could get rich.

It seems a bit too good to be true, and indeed, the news story is full of people “denying all wrongdoing” and pointing to regulatory investigations which didn’t end in action. All four bankers have now left Gazprombank and are doing other things, presumably with fond memories of the day that an opportunity crossed their desk.