Jim Cramer put a number on something the market had been dancing around for months: Nvidia isn’t just building the fastest chips anymore, it’s buying the roads that lead to them. On September 3, 2026, Nvidia confirmed it is acquiring Hugging Face, the open-source AI model repository, for $12.9 billion, a figure Cramer rounded to “$13 billion” during a CNBC segment with David Faber the same day. His verdict, quoted by 24/7 Wall St., was blunt: Nvidia spent $13 billion to make sure customers never seriously shop a rival’s chip.
The deal is Nvidia’s second-largest acquisition on record, trailing only the roughly $20 billion purchase of assets from chipmaker Groq in December 2025, according to CNBC. It also lands weeks after Nvidia closed a combined roughly $40 billion in strategic stakes across Anthropic and OpenAI, moves this site covered in detail when Nvidia scaled back its AI lab investment strategy earlier this year. Taken together, the Hugging Face purchase reads less like an isolated bet and more like the latest layer in a stack Nvidia has been building since 2025: buy the labs, buy the distribution hub, and keep the CUDA software moat wide enough that switching to AMD, Broadcom, or a hyperscaler’s custom silicon costs more than it saves.
This is a news analysis of what Nvidia bought, what Cramer said about it, and whether the strategy is as airtight as Wall Street currently believes.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Cramer Actually Said on CNBC
Cramer’s comments, made during his September 3, 2026 conversation with David Faber and quoted by 24/7 Wall St., centered on a market asymmetry he says only Nvidia currently enjoys. His exact words: “If AMD were announced, they were buying Hugging Face, if Broadcom announced they were buying Hugging Face, we would send NVIDIA down. So you got to think of it like that. It’s a little asymmetrical.”
The point is narrower than “Nvidia bought a monopoly.” Cramer is describing investor psychology: the market has extended Nvidia enough benefit of the doubt, built on seven straight quarters of data center growth, that a $12.9 billion acquisition reads as routine capital allocation rather than a defensive scramble. If AMD or Broadcom tried the identical move, Cramer argues, investors would likely interpret it as an admission of weakness rather than strength, and punish the stock accordingly. Nvidia shares closed the session at $228.45, up 1.8% on the day and extending a 22.64% year-to-date gain, per 24/7 Wall St.’s market data.
Inside the $12.9 Billion Hugging Face Acquisition
Hugging Face is not a chipmaker or a cloud provider. It’s a repository and community where developers publish, discover, and fine-tune open-source AI models, datasets, and evaluation code. According to CNBC, the platform gives Nvidia direct access to more than 18 million AI developers, along with the talent behind it: Hugging Face CEO Clément Delangue and co-founders Julien Chaumond and Thomas Wolf are joining Nvidia as part of the deal.
Nvidia has said it will keep Hugging Face open to all silicon vendors, clouds, and model providers, and that it will not require users to run models on Nvidia hardware, a point CNBC’s coverage flagged explicitly. That commitment matters, because it’s the difference between an outright walled garden and something subtler: a distribution hub where the reference implementations, tutorials, and one-click deployment templates default to CUDA and NVLink unless a developer goes out of their way to target something else. Owning the hub doesn’t require locking the door. It just requires shaping which door people walk through first.
Jensen Huang’s Public Defense of the Deal
Nvidia CEO Jensen Huang framed the acquisition around openness rather than control. In comments carried by 24/7 Wall St., Huang said: “Our fundamental goal is just to make sure that AI advances as quickly as possible. And it’s really, really important right now as the open models are really accelerating that we make sure that we provide Hugging Face the platform to continue to scale and for the resource for them to scale and extend the open model ecosystem and community.”
In a blog post announcing the deal, quoted by CNBC, Huang wrote: “We build our own models, libraries and tools in the open so developers everywhere can use them, modify them and build on top of them.” Yet in the same CNBC segment, asked how Nvidia balances support for both open and closed models, Huang added a more revealing line: “I recommend that people use closed models as much as they can, you know, because it’s off the shelf, it’s incredibly good, it’s advancing very quickly.” Both statements can be true at once, and that’s the tension analysts keep circling back to: a company can genuinely champion an open ecosystem while still benefiting most when that ecosystem runs best on its own silicon.
The Microsoft-GitHub Playbook, Eight Years Later
CNBC’s own analysis drew a direct parallel to Microsoft’s 2018 purchase of GitHub for $7.5 billion. Then, as now, the buyer gained control over the leading discovery, collaboration, and distribution platform for its category of software, then leveraged developer loyalty into demand for its core paid business. For Microsoft, that meant steering GitHub users toward Azure. For Nvidia, the equivalent play is steering Hugging Face’s developer base toward CUDA and Nvidia’s accelerated computing stack.
The comparison also cuts the other way as a note of caution. GitHub, under Microsoft, largely retained its open character and its price of entry stayed free for most users, and Microsoft has still generated an outsized return in Azure loyalty over eight years. If Hugging Face follows the same slow-burn path rather than any abrupt CUDA-only pivot, the payoff for Nvidia is less about squeezing developers immediately and more about compounding a preference over the next several years, the same multi-year horizon Microsoft played on with GitHub.
Nvidia’s Acquisition Pattern: From Groq to Hugging Face
The Hugging Face deal doesn’t stand alone. It’s the fourth major move in roughly ten months where Nvidia has spent tens of billions to shape who builds AI models and on what hardware they run best.
| Deal | Amount | Date | Strategic Purpose |
|---|---|---|---|
| Anthropic stake | Up to $10 billion (plus $5B from Microsoft) | November 2025 | Tie frontier lab compute demand to Nvidia Grace Blackwell and Vera Rubin systems |
| Groq asset purchase | ~$20 billion | December 2025 | Nvidia’s largest acquisition on record; absorbed rival inference chipmaker’s assets |
| OpenAI stake | $30 billion (of a ~$110B round) | Finalized February 2026 | Secured equity in one of Nvidia’s largest GPU customers |
| Hugging Face acquisition | $12.9 billion | September 2026 | Control the open-source model distribution hub used by 18M+ developers |
Sources: TechCrunch, AI Funding Tracker, CNBC.
AI Funding Tracker’s review of the Anthropic and OpenAI checks notes that Huang told employees in March 2026 he expected those two stakes to be Nvidia’s last major direct investments in either lab, a comment that, in hindsight, undersold how much capital Nvidia would still deploy months later, just aimed at infrastructure and distribution rather than a third frontier lab. Combined, the Anthropic and OpenAI positions alone total roughly $40 billion in exposure to two of Nvidia’s biggest customers, a dynamic covered in depth when those deals were unwound and restructured earlier in 2026.
Data Center Revenue: The Numbers Behind Nvidia’s Confidence
Cramer’s asymmetry argument only holds because Nvidia keeps delivering the underlying growth that earns market patience. The data center segment, which now drives nearly all of Nvidia’s revenue, has compounded quarter over quarter through fiscal 2026 and into fiscal 2027.
| Quarter | Data Center Revenue | YoY Growth | Total Revenue |
|---|---|---|---|
| Q2 FY2026 (reported Aug 2025) | $41.1 billion | +56% | $46.7 billion |
| Q3 FY2026 (reported Nov 2025) | $51.2 billion | +66% | $57.0 billion |
| Q4 FY2026 (reported Feb 2026) | $62.3 billion | +75% | $68.1 billion |
| Q2 FY2027 (reported Aug 2026) | $89.0 billion | +117% | $96.2 billion |
Sources: Nvidia investor relations, Data Center Dynamics, Unite.AI.
By the fiscal second quarter of 2027, data center sales alone reached $89.0 billion, up 117% year over year, according to Unite.AI’s earnings recap. Net income for that quarter hit $59.69 billion, per 24/7 Wall St.’s coverage of the same results, framing the $12.9 billion Hugging Face price tag as, in the outlet’s words, “a rounding error in cash and a meaningful move in strategy.” Nvidia also had roughly $99 billion remaining under its buyback authorization at the end of that quarter, out of a $1 trillion program, leaving ample room to keep making acquisitions of this size without touching its balance-sheet flexibility. Readers tracking the hardware side of that spending can see it reflected in Nvidia’s GB300 Blackwell Ultra shipments, which are the physical product driving the same data center numbers.
Competitive Landscape: AMD, Broadcom, and Custom Silicon
The real audience for the Hugging Face deal isn’t developers, it’s the hyperscalers building their own chips. Amazon’s Trainium is already disclosed as a multibillion-dollar business, and Google’s TPU line has been in production for years, both aimed at the same workloads Nvidia’s GPUs currently dominate. If the open models developers pull from Hugging Face keep shipping CUDA-optimized first, those custom chips underperform on the exact workloads companies actually deploy, regardless of their raw specs on paper.
| Company | Ticker | Sept 3-4, 2026 Move | Competitive Position vs. Hugging Face Deal |
|---|---|---|---|
| Nvidia | NVDA | $228.45, +1.8% on the day | Now owns the open-source distribution hub outright |
| AMD | AMD | ~$477, up nearly 4.7% | Most exposed if CUDA-first defaults persist on Hugging Face |
| Broadcom | AVGO | ~$358, roughly flat | Issued soft Q4 guidance on custom-silicon lumpiness, per CNBC |
| Amazon (Trainium) | AMZN | Not directly reported | Disclosed multibillion-dollar custom chip business, no Hugging Face equity |
| Google (TPU) | GOOGL | Not directly reported | Years of in-house TPU production, still outside the Hugging Face ownership structure |
Source: 24/7 Wall St. stock data and companies-mentioned figures, published September 4, 2026.
24/7 Wall St. also noted that Broadcom’s own results that week paired strong AI chip demand with soft fourth-quarter guidance, a reminder that custom-silicon revenue runs lumpier than Nvidia’s platform-driven sales even during an AI infrastructure boom. That contrast is part of why Nvidia’s broader AI chip pricing power has held up even as rival stock reactions to Nvidia hardware launches have gone the other direction in past announcements this year.
The Falsifiable Test Cramer Didn’t Spell Out
What separates a defensible $12.9 billion bet from an overpriced vanity purchase is a test that’s actually falsifiable, and 24/7 Wall St.’s coverage lays it out clearly: if the next wave of leading open models published on Hugging Face ships optimized for AMD’s MI-series accelerators or a hyperscaler’s custom chip first, with CUDA support arriving only weeks later, the strategy will have failed at its stated purpose. Until that pattern shows up, the offensive-and-defensive framing that both Cramer and Huang described holds.
That test also gives outside observers, developers, competitors, and regulators alike, something concrete to watch instead of relying on Nvidia’s own characterization of “openness.” Nvidia’s public commitment to keep Hugging Face neutral across silicon vendors is the kind of promise that’s easy to make in a press release and harder to verify without watching actual model releases over the next several quarters.
What This Means for AI Developers and Enterprises
For the 18 million-plus developers already on Hugging Face, day-to-day workflows aren’t expected to change immediately. Nvidia has said the platform stays open to all clouds, silicon vendors, and model providers. The more gradual shift to watch is in defaults: which inference runtime a new model repo recommends out of the box, which quantization format gets first-class documentation, and which deployment templates appear first in a model card’s “how to run this” section. None of that requires Nvidia to close anything off. It just requires Nvidia to make the CUDA path the path of least resistance, which is a much harder thing for competitors or regulators to point to as a violation.
For enterprises evaluating AI infrastructure procurement, the acquisition adds a new variable to vendor diversification strategy. Teams that had planned to hedge Nvidia dependence by testing workloads on AMD’s Instinct line or a hyperscaler’s custom silicon may now find that the reference implementations and tuning guides they’d lean on for that migration are, by default, still built around Nvidia’s stack first.
Historical Context: How Platform Ownership Became a Chip Strategy
Nvidia’s CUDA moat didn’t start with Hugging Face. It started in 2007 when Nvidia released CUDA as a proprietary parallel computing platform, years before deep learning made GPUs indispensable. That early bet meant that by the time AI research exploded in the 2010s, nearly every academic paper, framework, and tutorial already assumed Nvidia hardware, and switching costs for researchers and companies alike were built in from the start rather than added later. The Hugging Face acquisition is best read as an extension of that same 2007-era logic applied to a 2026-era distribution problem: own the layer where defaults get set, not just the chip that eventually runs the workload.
The Microsoft-GitHub parallel that CNBC drew is instructive here too. Microsoft’s 2018 GitHub purchase for $7.5 billion didn’t close off open-source development, but it did make Azure the natural next step for millions of developers already living inside GitHub’s workflow. Eight years on, that bet is widely regarded on Wall Street as one of Microsoft’s better-timed acquisitions of the decade, which is part of why Cramer’s framing of the Hugging Face deal as “offensive and defensive” landed as a compliment rather than a criticism.
Wall Street’s Reaction: Stock Moves and Price Targets
Beyond the day-of stock move, 24/7 Wall St. reported that Nvidia carries an average analyst price target of $325.99, well above where shares traded after the announcement, with a forward fiscal 2028 EPS consensus of $13.13. That gap between the trading price and analyst targets suggests Wall Street had already priced in continued platform expansion before the Hugging Face deal was even announced, treating it as confirmation of an existing thesis rather than a surprise catalyst.
Palo Alto Networks CEO Nikesh Arora publicly congratulated Huang on the deal via a post on X, according to CNBC’s coverage, specifically citing the importance of balancing open-source and open-weight approaches with frontier language models. That kind of public reaction from an adjacent enterprise software CEO signals how closely the broader tech industry is watching Nvidia’s platform moves, even outside direct chip competitors.
Predictions: Where This Goes From Here
- CUDA-first defaults become the visible battleground. Expect scrutiny to shift from deal size to which inference runtimes and quantization formats new Hugging Face model cards recommend first over the next two to three quarters.
- AMD and hyperscalers push harder on Hugging Face parity. AMD, Amazon, and Google have clear incentive to lobby publicly, and technically, for equal footing on model card documentation and one-click deployment templates.
- Regulatory attention increases but enforcement lags. With Nvidia’s data center revenue now over 90% of company sales in some recent quarters, expect more commentary from policymakers about market concentration, even if no formal antitrust case materializes in the near term.
- Nvidia’s acquisition pace slows on lab investments, not infrastructure. Following Huang’s earlier comments that the Anthropic and OpenAI stakes were likely his last major lab checks, further large deals are more likely to target distribution and tooling platforms like Hugging Face rather than additional equity stakes in frontier labs.
- Cramer’s falsifiable test becomes the industry’s benchmark. Watch for competitors and analysts to explicitly track whether leading open models ship AMD- or custom-silicon-optimized versions before or after their CUDA-optimized release, using that gap as a public scorecard for the deal’s success.
Market Impact Beyond the Chip Sector
The ripple effects extend past semiconductor stocks. Cloud providers that had been quietly building custom silicon partly to reduce Nvidia dependence now face a platform-level reason developers might still default to Nvidia-optimized deployments regardless of what chip sits underneath a given cloud instance. That matters for AWS, Azure, and Google Cloud pricing strategy around AI compute, since a persistent developer preference for CUDA-tuned models can sustain demand for Nvidia-based instances even when a cloud provider would rather steer customers toward its own cheaper, in-house silicon. It also reinforces why AWS continues placing massive Nvidia GPU orders even as it develops Trainium in parallel, since hedging against Nvidia dependence and serving developer demand for Nvidia-first tooling can pull a cloud provider’s procurement in opposite directions at once.
For component suppliers further down the chain, the deal also underscores how much of Nvidia’s competitive position now runs through partnerships and acquisitions rather than raw chip specs alone, a dynamic visible in Nvidia’s separate $3.5 billion investment in MediaTek to extend NVLink Fusion into more of the supply chain. Each of these moves, taken individually, looks like ordinary corporate development. Stacked together across less than a year, they describe a company treating software and distribution control as seriously as it treats transistor design, at a moment when Intel’s competing AI GPU roadmap has slipped and left Nvidia holding an outsized share of the data center AI accelerator market with little near-term pressure to change course.
Frequently Asked Questions
How much did Nvidia actually pay for Hugging Face?
Nvidia’s acquisition of Hugging Face was reported at $12.9 billion, according to CNBC’s coverage of the September 3, 2026 announcement. Jim Cramer and several outlets rounded the figure to $13 billion in their commentary.
What did Jim Cramer say about the Nvidia-Hugging Face deal?
Speaking with David Faber on CNBC, Cramer argued the market treats Nvidia’s acquisitions differently than it would treat an identical move from AMD or Broadcom, calling that pricing gap “a little asymmetrical.” He described the deal as both offensive, extending Nvidia’s ecosystem, and defensive, keeping the platform out of rivals’ hands.
Is Hugging Face still open to AMD and other chipmakers after the acquisition?
Nvidia has stated publicly, per CNBC, that Hugging Face will remain open to all silicon vendors, clouds, and model providers, and that it will not require users to run models on Nvidia hardware. Whether that openness translates into equal treatment for non-Nvidia optimized models in practice is the open question analysts are watching.
How does this compare to Microsoft’s GitHub acquisition?
CNBC drew a direct parallel to Microsoft’s 2018 purchase of GitHub for $7.5 billion, noting both deals gave the acquirer control over the leading discovery and distribution platform in their category, which each company then used to drive demand for its core paid business, Azure for Microsoft and CUDA-based hardware for Nvidia.
What other AI investments has Nvidia made recently?
In the ten months before the Hugging Face deal, Nvidia committed up to $10 billion to Anthropic in November 2025, spent roughly $20 billion acquiring assets from chipmaker Groq in December 2025, and finalized a $30 billion stake in OpenAI in February 2026 as part of a roughly $110 billion funding round, according to TechCrunch and Reuters reporting.
How is Nvidia’s data center revenue trending in 2026?
Nvidia’s data center revenue climbed from $41.1 billion in the quarter reported August 2025 to $89.0 billion in the quarter reported August 2026, an increase of 117% year over year, according to Nvidia’s own investor relations disclosures and coverage from Unite.AI and Data Center Dynamics.
What would prove Nvidia’s Hugging Face strategy has failed?
According to 24/7 Wall St.’s analysis, the clearest failure signal would be leading open models on Hugging Face shipping optimized for AMD’s MI-series chips or a custom hyperscaler chip before CUDA support arrives. If that pattern becomes common, it would suggest Nvidia’s ownership of the platform isn’t translating into the software lock-in the acquisition was designed to create.
Who is joining Nvidia as part of the Hugging Face deal?
Hugging Face CEO Clément Delangue and co-founders Julien Chaumond and Thomas Wolf are joining Nvidia along with the rest of the platform’s executive team, according to CNBC’s report on the acquisition.
Related Coverage
- Nvidia Brings DLSS 5 to RTX 40, Skips Full Control [2026]
- AMD Ryzen 5 7500 Zen 4 CPU Fills iGPU Gap in Budget Lineup [2026]
- AMD Ryzen 5 7500 Leak: Zen 4, 6-Core, RDNA2 Graphics [2026]
- HPE Stock Surges 7% as Nvidia Vera CPU Server Debuts [2026]
- AMD RDNA 5 Leak Targets 3.5 GHz GPU Clocks [2026]
