Under the torrent of token reasoning, GPU dominators are striving to control the “AI model entrance”! Nvidia (NVDA.US) wants to devour Hugging Face for $14 billion this week
The Zhitong Finance App learned that Nvidia (NVDA.US), the “AI chip superleader” that announced its explosive performance report last Thursday, is in in-depth negotiations to acquire Hugging Face. The total transaction scale may reach about 14 billion US dollars, including about 12.9 billion US dollars in acquisition consideration, and may also include an employee retention plan of about 1 billion US dollars. The deal will enable Nvidia to control an important global open artificial intelligence model sharing platform, directly connect developers, models and enterprise customers, and reduce its high dependence on a few tech giants that contribute large amounts of revenue data while actively developing AI chips for self-developed data centers by supporting open models.
What Nvidia really wants to buy is not an ordinary AI application software company, but a “model distribution layer” for the AI era and a portal for global AI developers. Hugging Face is equivalent to “GitHub in the field of artificial intelligence models”, connecting models, data sets, development tools, enterprise customers and developers; after the acquisition, Nvidia can complete the “model discovery — evaluation — optimization — actual deployment” process on top of GPUs, NVLink networks, CUDA software stacks, and artificial intelligence systems, making the open model more naturally adapting to its hyperscale AI training and inference platform, while at the same time identifying which models and workloads are forming the next round of computing power requirements.
This is also Nvidia's structural hedging against chips developed by closed model giants and hyperscale cloud vendors: as long as the open model ecosystem remains prosperous, demand for artificial intelligence applications will not be completely locked in by a few closed platforms and their self-developed accelerators. The logic behind Stripe's acquisition of OpenRouter is similar — OpenRouter controls model selection, request routing, cost, and billing entry, while Hugging Face controls the entry point for discovery, sharing, development, and deployment; Stripe competes for “model call traffic,” and Nvidia competes for “how to transform model adoption into computing power requirements.”
Nvidia's 70% growth outlook breaks through the “AI peaking theory”, and global AI computing power infrastructure is still in full swing
Global investors remain strong in their risk appetite for the AI computing power industry chain after experiencing a sharp decline after experiencing AI deleveraging and de-crowding in July.
The Philadelphia Semiconductor Index once doubled its June high from the beginning of the year, then retreated at most about 29% in July; as of September 1, it closed at 11,288.61 points, and still rose more than 50% during the year. South Korea's KOSPI index fell from a high of about 9,385 points in June to 5,593.56 points on July 30. The maximum retracement was more than 40%, then a rebound of more than 20% meant re-entering the technical bull market; it closed at 6,835.80 points on September 1, and the increase was still about 62% during the year. The latest progress of these two major “AI computing power weather vane” indices shows that the July crash was more like an AI super bull market stress test caused by high leverage, crowded positions, and valuation compression, rather than a fundamental reversal in AI computing power demand.
Demand for AI computing power infrastructure is changing from the budget intentions of major cloud computing and AI application development companies to multi-year computing power capacity locking. Anthropic reportedly signed an artificial intelligence cloud computing agreement with Nvidia-supported Lambda worth 35 billion US dollars, corresponding to an artificial intelligence cloud computing agreement with a capacity of about 350 megawatts; not long ago, about 45 billion US dollars was used to lock Nscale's 460 megawatts of computing power over the next six years. The latter will deploy Nvidia's next-generation AI computing power cluster, the Vera Rubin platform, on a large scale. South Korea's exports in August increased 68.7% year-on-year to US$98.26 billion. Among them, exports of semiconductor products surged 209% to a record 46.65 billion US dollars, accounting for 47.5% of total exports.
AI GPU clusters and memory chips are still the most prominent bottlenecks in AI computing power systems in the world under the token inference torrent. TrendForce expects DRAM and NAND Flash contract prices to rise 13% to 18% and 10% to 15%, respectively, in the third quarter; the cumulative increase in server DRAM and enterprise solid-state drive prices may reach about 270% and 235%, respectively, in 2026, and HBM contract prices may rise another 70% to 140% in 2027. DRAM and NAND are expected to account for 47% of the capital expenditure of cloud computing service providers, rising further to 68% in 2027.
AI competition comes to “Big Model Entry Battle”: Nvidia wants to swallow Hugging Face
Nvidia is expanding competitive boundaries from chip performance to model distribution and developer relationships to grasp the source of future computing power demand. The successful acquisition of Hugging Face means that Nvidia is upgrading from a computing power vendor to a full-stack basic platform for artificial intelligence throughout model development, deployment, and commercialization.
According to people familiar with the matter, Nvidia is in in-depth negotiations to acquire artificial intelligence startup Hugging Face, and the potential size of this deal could reach around $14 billion.
One of the people familiar with the matter, who requested anonymity because the relevant details have not been disclosed, said that Nvidia may reach an agreement to acquire Hugging Face for $12.9 billion as soon as this week. The deal may also include a $1 billion retention plan for Hugging Face employees, the source said.
People familiar with the matter said that the two parties have yet to reach a final agreement, and the timing or details of the deal may still change. Representatives for both Nvidia and Hugging Face declined to comment.
The acquisition of Hugging Face will be the biggest move CEO Hwang In-hoon has taken to date to expand the scope of artificial intelligence applications and expand the customer base. The deal will allow Nvidia to control a key platform where developers can showcase and share artificial intelligence models.
Hwang In-hoon is committed to helping cultivate the so-called open source AI model to drive the long-term strong expansion of AI GPU demand and ecosystem to prevent the technology of a few large companies from dominating artificial intelligence — these companies currently contribute the largest portion of Nvidia's revenue, but they are also actively promoting their own chip projects.
Nvidia is already an investor in Hugging Face. Other investors include Alphabet's Google, Amazon, Intel, and Salesforce.
Hugging Face was founded in 2016 and operates an artificial intelligence model sharing platform where these models can be freely used by others. The startup was valued at $4.5 billion in a funding round three years ago. The well-known media “Business Insider” has previously reported Nvidia's negotiations with the company.
Hugging Face develops artificial intelligence software and provides software hosting services to other companies. The company was previously at the center of a cybersecurity incident: a model OpenAI was testing accidentally hacked the platform. This security breach has raised concerns about the security of cutting-edge artificial intelligence technology.
OpenAI said it could have responded positively earlier to stop attacks against the Hugging Face system.
Over the past year, Nvidia has reached a number of deals, including a $6 billion licensing agreement with startup Poolside in August, which also includes issuing job offers to many of the company's employees. Nvidia also paid around $20 billion to acquire most of the business of chip startup Groq.
Nvidia is the world's highest-capitalized listed company and the world's absolute leading manufacturer of artificial intelligence accelerators; these chips help train and run artificial intelligence models. To scale up the AI economy, the company gradually added a variety of other technologies, including software.
The Santa Clara, California-based Nvidia company last week gave a sales data forecast for the 2028 fiscal year that exceeded expectations. Hwang In-hoon's management expects overall revenue growth of about 70% for the 2028 fiscal year.
Global AI capital expenditure has entered the trillion-dollar era, and the bull market trajectory of AI chip superdominance is far from over
As global AI capital expenditure continues to expand strongly, Hugging Face can complement Nvidia's software distribution layer, further upgrading its growth logic from chip shipping to full-stack ecosystem penetration. If the deal is successful, Nvidia will further control the GPU vendor to discover, distribute and deploy the gateway to more efficiently transform developer and application ecosystem token traffic into CUDA and AI computing power requirements.
Deep collaboration between the model platform and GPU, CUDA, and Vera Rubin systems can enhance developers' stickiness and demand insight, and allow Nvidia to capture incremental opportunities for training and inference computing power earlier.
Based on the AI computing power industry chain and Nvidia's performance, and the positive signal of AI computing power demand released at the latest Hot Chips conference, the research department from Wall Street financial giant Goldman Sachs recently restored the “capital expenditure of American hyperscale cloud service providers (i.e. hyperscalers)” well-known in the market into a true “global AI investment.” Goldman Sachs said that the previously compiled $794 billion caliber not only omitted private AI companies, Asian companies, and non-US projects, but also mixed in traditional capital expenditure that was not entirely used for AI; Goldman Sachs's latest forecast for 2026 will be about 1 trillion US dollars of global AI investment, of which about 581 billion US dollars will actually occur in the US.
Another Wall Street financial giant, Morgan Stanley, predicts that by 2028, close to $3 trillion of AI-related infrastructure investment will flow through the global economy, and more than 80% of spending is still ahead. The core increase in AI computing power demand comes from AI moving from “answering questions” to “execution workflows”: agents require continuous planning, calling tools, verifying results, and retrying failures. The AI inference token consumption for a single task may reach 10 times, 20 times, or even 50 times that of traditional chat queries; the world model also extends the boundaries of computing power from text to robots, industrial simulations, and physical systems. Goldman Sachs officially predicts that by 2030, global token consumption may increase 24 times to 120 trillion per month; at the same time, the unit cost of AI inference tokens will drop by 60% to 70% each year, forming the Jevons Paradox (Jevons Paradox) effect of “cost reduction — application diffusion — increase in total demand”.
Nvidia's latest performance and outlook, combined with the latest tens of billion dollar cloud computing power resource agreement signed by AI application leader Anthropic and South Korea's strong semiconductor export data, highlights that global demand for artificial intelligence computing power is still far from peaking.
Nvidia's revenue for the second quarter of fiscal year 2027 increased 106% year over year to US$96.2 billion, and data center business revenue increased 117% to US$89 billion; revenue guidance for the third quarter reached US$108 billion. The company's management also expects revenue growth of about 70% for the 2028 fiscal year. This strong, and still supply-bound outlook, is the fundamental basis for Wall Street to continue to raise the company's profit forecast and target price.
Wall Street financial giants such as Citibank, Goldman Sachs, and Morgan Stanley are all optimistic about Nvidia's overall demand for Vera Rubin's next-generation computing power architecture, the growth visibility of GPU clusters in 2027 and beyond, and the expansion of AI infrastructure. Their unanimous bullish judgment is that AI computing power demand continues to be strong, and Vera Rubin's capacity and Nvidia's comprehensive software and hardware platform advantages are still being strengthened. Citi raised its target price from $300 to $315; Morgan Stanley raised its “holdings” from $288 to $300; Goldman Sachs raised it from $285 to $300; and Bank of America maintained its “buy”, global preferred stock position, and a target price of $350.