Build7 distinct publishers3 min readPublished
Huang promises Nvidia compute will never be required to build or deploy on the Hub, and I believe him. The pledge says nothing about pricing, rate limits or defaults, which is where a registry actually steers work.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
Permission is what the pledge covers. Multi-cloud and multi-accelerator support is promised to continue [7]. Pricing, rate limits and which artifact a developer sees first are not mentioned anywhere in the post [23]. The word "open" appears 19 times in it, according to The New Stack [8], which is a decent proxy for how much of this deal's risk Nvidia thinks lives in that one word.
The GitHub comparison is the right one, used the wrong way round. Microsoft paid $7.5 billion in 2018 on similar promises of independence, and The New Stack notes GitHub largely kept its openness; the backlash came when Microsoft trained the proprietary, paid Copilot on public GitHub code [9]. Nvidia is paying about 1.7 times that in nominal dollars [21]. The platform promise held, but no one had written a promise covering the corpus, and that is where the pressure landed. Hugging Face holds more than 3 million models and 500,000 datasets [5], and The Decoder reports Nvidia already ships NVFP4 builds of Kimi K2.6 and GLM-5.1 tuned for its own chips [10].
The purchase economics are legible. They mostly cut toward neutrality. In late July Nvidia reported $36 billion in commitments from agreements with AI cloud partners; those commitments shrink when partners sell capacity to third parties, and Nvidia partly insures partners against unused capacity, per The Decoder [11]. Hugging Face rents compute as well as distributing weights, and it plugs into Nvidia's Lepton marketplace, which routes jobs to CoreWeave, Lambda and Nebius rather than into the DGX Cloud that Nvidia scaled back to avoid poaching its buyers' rental customers [12]. A job the Hub routes to a Lepton partner is third-party capacity sold, which trims both the commitment and the insurance exposure. Google, Amazon, OpenAI and now Anthropic design their own accelerators, while open models run inside companies, universities and public agencies that do not [13]. Huang gave Axios the thesis in one line: "Free AI should be great for hardware" [14].
The failure mode worth planning for is convenience, a slow drift in which artifacts stay maintained and visible, well short of outright exclusion. For the neutrality pledge to transfer to a non-Nvidia fleet, AMD and Intel artifacts would have to stay first-class in the same listings, at the same freshness, with the same deploy paths tested. That is not something a blog post can establish, and it is not what the promise says.
Huang announced the deal on September 3, 2026 after reports surfaced in late August [3], and the SEC filing puts close no earlier than the first half of 2027 [4]: four to ten months of runway [22]. The retention stock is up to roughly 8 percent of the headline number [20], which is about right when the asset is a team and a community. Delangue says Nvidia has committed to keeping the platform open, independent and compute-agnostic, with the founders and existing team staying [15], and his own argument for selling is that open source now needs more compute than Hugging Face can fund alone [16]. I take both at face value. I would still mirror the weights I ship and pin the revisions, because a registry with one owner is a single point of control regardless of who the owner is. That was true last week too. The acquisition just makes the budget conversation shorter.
Ranked by verification strength, evidence, and original report placement.
Jensen Huang announced in an NVIDIA blog post that NVIDIA has agreed to acquire Hugging Face for $12,930,300,000.
The roughly $12.9 billion total breaks down into approximately $11.9 billion in purchase price plus a stock program of up to $1 billion to retain staff.
Huang's announcement states that developers will choose the models, frameworks, clouds and inference service providers and computing platforms they want, and that "NVIDIA compute will not be required to build on or deploy through Hugging Face."
Huang writes that Hugging Face will continue to support open source and open weight models from every model builder and will continue to support multi-cloud and multi-accelerator development and deployment.
Huang announced the deal on September 3, 2026 in a blog post, after reports of the bid surfaced in late August.
According to the SEC filing, the deal will not close until the first half of 2027 and still needs regulatory approval.
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Hugging Face sells its independence for 1.85 times the valuation it turned down1 distinct publisher
invest
Nvidia's Perplexity talks move its money one layer further from its own chips1 distinct publisher
leadership
Nvidia's $279bn of reserved capacity now sets the price of everyone else's compute5 distinct publishers
product
The open-weight default is for sale at $13B, and its buyer pool ships models too4 distinct publishers
Distinct publishers with included, body-backed reporting in this cluster.
economictimes.indiatimes.com
1 article · September 3, 2026
mezha.net
1 article · September 3, 2026
nvidianews.nvidia.com
1 article · September 3, 2026
phoronix.com
1 article · September 3, 2026
the-decoder.com
2 articles · September 3, 2026
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Two documents, one written by the buyer
Almost everything firm here comes from Huang's announcement — the exact $12,930,300,000, the usage figures, the neutrality language — and the only independent paper trail is the SEC filing, which The New Stack alone appears to have opened. Mezha's Reuters sourcing corroborates the price split; The Decoder supplies the channel economics nobody else has. The deal facts are solid because deal facts are checkable. The characterisations around them are not.
The hub is entrenched; the ownership isn't in force
What is genuinely deployed is Hugging Face itself — 200,000 companies deploying from it and a Lepton integration that already routes jobs to CoreWeave, Lambda and Nebius. What is not deployed is anything this deal changes: nothing closes before 2027, so every neutrality commitment is currently a sentence rather than an observed behaviour. The measurable adoption belongs to the asset, not to the transaction.
Reassurance runs ahead of enforcement
The gap here isn't exaggeration about what happened — the price and the pledge are real and precisely stated. It's that a promise repeated 19 times covers only what developers are permitted to do, while the levers a registry uses to steer work are pricing, quotas and defaults, and none of those are mentioned in the announcement or the filing language. Five of our six sources reproduce the assurance without probing its scope; the GitHub precedent, where openness survived but Copilot still turned public code into a paid product, sits in one story out of six.
The buyer authored the primary source
Huang told Axios that free AI should be great for hardware, and that is the cleanest statement of interest anyone could ask for. Behind it: $36 billion in cloud-partner commitments that ease as partners resell capacity, a Lepton channel looking for demand, and rivals designing their own accelerators. On the other side, Delangue is arguing the case for a sale he is party to, and up to $1 billion of NVIDIA stock is set aside to keep the staff who will decide day to day what neutrality means in practice.
Facts firm, consequences unresolved
We can be confident about the transaction — price, structure, timing, pledge text, and the strategic motive stated on the record. We cannot be confident about what follows, and the story does not pretend otherwise: the deal needs approvals that The New Stack rightly says are no formality, the principals disagree about how it began, and the mechanisms most likely to shift developer behaviour were never put in writing.
thenewstack.io
1 article · September 3, 2026
tomshardware.com
1 article · September 3, 2026