Deep Dive — Nvidia's quiet takeover of the AI stack
Nvidia just posted a $96.2 billion quarter, guided to another year of ~70% growth well into fiscal 2028, watched its stock climb 7%, and — hours later — entered serious talks to buy Hugging Face for over $13 billion. Read separately, those are two feel-good headlines about an AI industry that can't stop spending. Read together, they are something more pointed: the company that powers almost every AI datacenter is moving to own the layer where models are distributed, hosted, and served — not just the silicon underneath them.
We flagged the quarter's scale this morning in Nvidia's $96B quarter puts it on the edge of $100B. The deeper story is what Nvidia does with that momentum, and the Hugging Face talks are the clearest tell yet that Jensen Huang's company no longer thinks of itself as a chipmaker selling shovels to the AI gold rush. It is trying to become the town itself.
The quarter that turned the scare around
For most of 2026, the dominant worry in markets has been that AI capital expenditure is a fragile loop — hyperscalers and startups borrowing billions to buy GPUs to rent back to each other, with revenue that fails to materialize. A July sell-off knocked a trillion dollars off chip stocks before they recovered. Wall Street arrived at Nvidia's late-August report braced for the narrative to crack.
It did the opposite. Revenue for the second quarter came in at $96.2 billion, with a forecast that chief financial officer Colette Kress framed as roughly 70% growth for fiscal 2028, the period running from February 2027 through January 2028. Huang went further: actual demand "is much greater than 70%," he said, with the company limited by what it can physically supply rather than by what customers want. The bottleneck is on Nvidia's side — TMSC, its main manufacturer, is still capacity-constrained, and high-bandwidth memory chips remain in short supply industry-wide.
Equally important was the claim that Nvidia's customer base is spreading out. Sales to its AI-clouds, industrial, and enterprise customers hit $40.3 billion in the quarter, up 138% year over year. That is the rebuttal to the "one lab drives the market" thesis that has haunted the trade: Huang framed it as an inflection point, arguing that a year ago "one lab alone was driving the buildout" and now a wave of frontier labs, startups, an open-model ecosystem, and physical AI are scaling in parallel.
It was enough to calm the skeptics. Siddy Jobe of Econopolis called the valuation "cheap"; Paul Meeks of Freedom Capital Markets said he sees no real threat of a slowdown into 2028. The immediate read across the industry was bullish. AWS underscored the point the next day by committing to another 2 million Nvidia GPUs through 2027 and 2028 — a hyperscaler, with in-house Trainium and Graviton chips of its own, still placing an order that TechCrunch sized in the tens of billions of dollars.
The $13 billion pivot into the middle of the stack
If the earnings say the compute buildout has runway, the Hugging Face talks say Nvidia wants a bigger share of what runs on top of it. Business Insider reported Wednesday that Nvidia is in serious talks to acquire Hugging Face at a valuation north of $13 billion. Microsoft met with the company too, but those conversations are no longer ongoing, per the report — positioning Nvidia as the lead suitor. Neither company has commented, and Reuters said it could not verify the report, so nothing is signed. We first tracked this when Hugging Face explored a sale at a $13B valuation; Nvidia as the buyer is the next chapter.
The price would roughly triple the $4.5 billion post-money valuation Hugging Face set in a 2023 funding round led by Salesforce Ventures, with Alphabet, GV, and IBM Ventures participating. What Nvidia would be buying is less a startup than a shared utility: the hub where most of the open-model world hosts its models, datasets, and inference code, and where builders actually run and test the things Nvidia's GPUs accelerate.
That is why the deal is so loaded. Nvidia does not need Hugging Face to sell chips — the hub's users already depend on its hardware. It already collaborates with Hugging Face deeply, wiring its NIM inference software and DGX training cloud into the platform. Buying the company outright would take a partner it already leverages and make it a wholly owned distribution channel, giving the chipmaker a stranglehold over how open models are discovered, shared, and served. In the emerging "AI factory" era, where the economics shift from selling boxes to selling an entire managed platform, owning the front door of the open ecosystem is a strategic position money can't easily replicate afterward.
The tension is obvious and was raised in the earlier brief. Hugging Face has built its reputation as a neutral, community-first steward of models and data — its founder Clément Delangue has repeatedly cast the company as a keeper of the ecosystem rather than a merchant in it. A union with the supplier that almost every lab already depends on puts the cop in the middle of the town square. The open question is whether Nvidia would preserve Hugging Face's model-agnosticism — including its partnerships with AMD, Intel, and Cloudflare for inference — or quietly turn the platform into an on-ramp for its own stack.
Why the middleware suddenly looks like the prize
To see why Nvidia would pay $13 billion for a platform it already relies on, it helps to reconsider how money is made in AI right now. Raw GPU demand was never really in doubt after the earnings. What is contested is the margin that sits above the hardware: who controls the models, the hosting, the inference runtime, the developer relationship. The chip is the tollbooth; the middleware is the whole town beyond it.
Nvidia has been quietly methodical here. Beside its hardware, its NIM inference stack, DGX Cloud training service, and partnerships that let developers spin up GPU clusters on demand have all been extending its fingers into the software layer for years. Its open-weights push is part of the same logic — we explored that strategy in detail in Nvidia's $6 billion plan to beat China at open weights, where giving away capable models is a way to keep inference running on Nvidia silicon rather than a rival's. Buying Hugging Face would fuse both moves: an open-model distribution monopoly that is also, by construction, an Nvidia-adjacent one.
The danger Nvidia is positioning against is real. Hyperscalers are building custom silicon at scale — Amazon's Trainium and Graviton, and now OpenAI's in-house inference chips, which we covered when the accelerator was unveiled. The more of the open ecosystem that gets consolidated inside an Nvidia-controlled platform, the harder it becomes for those challengers to build software momentum of their own. Control of distribution is a moat that competes with architectural advantage.
Skeptics will note the flip side: Hugging Face is, at bottom, a platform people use because it is free and open. Charging for it, or forcing it to favor Nvidia's own models and tooling, could erode the very community that gives it strategic value. There is a real argument that Nvidia should simply keep Hugging Face as an independent ally rather than spend $13 billion to make its repeated independence into a marketing liability — regulators in both the US and EU have recently shown appetite for scrutinizing exactly this kind of consolidation at the heart of the open AI ecosystem.
What to watch next
Three things decide whether this becomes the defining acquisition of the AI buildout. First, whether the deal is actually confirmed and at what price — reports of talks are not a signed term sheet, and a company as regulatory-conscious as Nvidia knows a $13 billion purchase of the neutral open-model hub will draw attention. Second, what Nvidia promises about openness: the clearest signal of intent will be whether it grants Hugging Face structural independence, or whether it folds the platform's roadmap into its own. Third, how Microsoft and OpenAI respond — both are actively building competing model and compute layers, and a Nvidia-owned Hugging Face would materially change the map.
The near-term read is simpler. The record quarter and the 70% guidance quiet the "AI has peaked" camp for now, and the $13 billion bid says the man who built the silicon supercycle now believes the bigger prize sits in the software it runs. Nvidia may soon not just sell the shovels — it may own the claim on the goldfield too.
