NVIDIA: AI GPUs, Infrastructure Deals and Investments

NVIDIA combines AI processors, networking and CUDA software. Explore its hardware, cloud partnerships, financing intentions, licenses and acquisition agreements.

kindCompany
foundedApr 1993

Updated

From graphics to accelerated computing

NVIDIA was incorporated in California in April 1993 and later reincorporated in Delaware. Co-founder Jensen Huang remains president and chief executive. Its GPUs support parallel computation for model training and inference, alongside graphics; CUDA, introduced in 2006, supplies a programming platform and libraries for that hardware.

The company reported $96.2 billion in revenue for the quarter ended July 26, 2026, including $89.0 billion from Data Center. Those period figures establish the scale of its infrastructure business without supplying a GPU market-share estimate.

Chips, networking and software as a system

NVIDIA’s AI platform combines GPUs, CPUs, interconnects, networking and software, including tools for model development, simulation and robotics. The March 2026 Vera Rubin introduction described an NVL72 rack with 72 Rubin GPUs and 36 Vera CPUs. Its advertised efficiency gains are vendor comparisons, rather than universal workload results.

Cloud access is another route to that hardware. The August 2026 AWS expansion planned two million additional GPUs for 2027 and 2028. That plan is distinct from completed installation and does not establish the fleet size of every other cloud.

Financing intentions and credit support

The September 2025 OpenAI announcement was a letter of intent: at least 10 gigawatts of planned systems, with NVIDIA intending to invest up to $100 billion progressively. The parties still expected to finalize details; the announcement did not establish that the maximum had been paid or became a finalized purchase commitment.

In August 2026, NVIDIA and six financial institutions signed memoranda targeting more than $500 billion of third-party infrastructure capital over time, subject to final agreements. Its separate SB Energy announcement described land, power and shell credit support and a planned $1.5 billion investment. Credit support, equity investment and capital-raising targets are different financial roles.

Investment, licensing and acquisition are different

The Anthropic partnership coupled model-and-hardware engineering with NVIDIA’s commitment to invest up to $10 billion. Microsoft separately committed up to $5 billion. The investment caps did not establish that those maximum amounts had already transferred.

The Groq agreement licensed inference technology nonexclusively and named Jonathan Ross, Sunny Madra and other team members who would join NVIDIA. Groq said it would remain independent. Newcomer separately reported a $6 billion nonexclusive Poolside license and $1 billion investment, citing an investor letter; those terms remain attributed reporting rather than a company-confirmed acquisition.

A proposed software-platform acquisition

NVIDIA agreed to acquire Hugging Face for approximately $12.9 billion in September 2026. Its announcement promised continued support for multiple clouds and accelerators without requiring NVIDIA compute. That is an announced transaction and a stated platform policy, rather than proof of completed control or future performance.

Market movements and enforcement boundaries

Reuters reported a roughly $593 billion market-value loss during the January 2025 DeepSeek selloff, and a first $5 trillion valuation that October. These are observable market milestones; they do not prove a single explanation for changes in long-term compute demand.

A separate August 2026 Taiwan indictment involved individuals including an NVIDIA employee and Supermicro Taiwan staff. Charges and alleged export diversions are distinct from a conviction of either corporation.

Sources

  1. [1]
    NVIDIA Corporation Form 10-K, fiscal year 2026

    U.S. Securities and Exchange Commission · Feb 25, 2026

  2. [2]
    Jensen Huang

    NVIDIA · retrieved Oct 23, 2026

  3. [3]
  4. [4]
  5. [5]
  6. [6]
  7. [7]
  8. [8]
  9. [9]
  10. [10]
  11. [11]
    NVIDIA to Acquire Hugging Face

    NVIDIA · Sep 3, 2026

  12. [12]
  13. [13]
  14. [14]
  15. [15]
    Nine indicted over AI server exports

    Taipei Times · Aug 25, 2026

NVIDIA history

2026

│
Jensen Huang admits an affair and NVIDIA's board opens an independent investigation

Jensen Huang admitted an affair in a memo to NVIDIA staff after The New York Times and the Financial Times reported that US intelligence believed a Megaspeed executive had been recruited by China to cultivate him. NVIDIA's board opened an independent investigation and said it had no reason to believe national security or company secrets were compromised.

NVIDIA
Company
│
Cohere announces initial close of a Series E targeting C$7 billion

With Cohere, Government of Canada, Schwarz Group

Cohere announced the initial close of a Series E targeting up to C$7 billion. Canada committed up to C$1.6 billion in staged equity investment tied to Canadian jobs, computing infrastructure and headquarters commitments.

Cohere
Funding
■
Naver signs $9 billion Brookfield computing agreement

With Naver, Naver Cloud, Brookfield

Naver signed a $9 billion computing-infrastructure agreement with Brookfield, meeting a condition for NVIDIA’s strategic investment. The agreement supports a planned AI factory operated by Naver Cloud.

Naver
Deal
│
Salesforce puts its Koa CRM reasoning model into pilot

With Salesforce

Salesforce and NVIDIA introduced Koa, an Agentforce model post-trained from Nemotron 3 Super on synthetic CRM workflows. Already used inside Salesforce, it became available to select pilot customers, with general U.S. availability planned for winter 2026.

Salesforce
Model
■
NVIDIA agrees to acquire Hugging Face for $12.9 billion

With Hugging Face

NVIDIA is paying about $11.9 billion to Hugging Face's stockholders plus up to $1 billion in retention equity for staff who join NVIDIA, buying the platform used by more than 18 million developers, researchers and creators, with more than 3 million models, 500,000 datasets and 1 million applications. NVIDIA says the platform keeps its own name and stays open to non-NVIDIA hardware.

NVIDIA
Acquisition
■
AWS commits to 2 million more Nvidia GPUs for 2027 and 2028

With Amazon

Amazon committed to deploying 2 million more Nvidia GPUs across its infrastructure in 2027 and 2028, on top of the 1 million already pledged, taking the total past 3 million. The expansion covers Blackwell Ultra, Rubin and Rubin Ultra chips and sets aside 100,000 GPUs for secure US government workloads.

NVIDIA
Deal
│
Perplexity ships a fully local agent on the DGX Spark

With Perplexity

Portable Computer runs the orchestrator, subagents and harness on the desk, on a 27 billion parameter Qwen or Perplexity's own post-train of it, and asks permission before any step goes to a frontier model in the cloud. Local steps cost no credits, which is the whole argument.

Perplexity
Product
│
Taiwan indicts NVIDIA and Supermicro staff over Blackwell servers sent to China

With Super Micro Computer

Keelung prosecutors say the group ordered 130 Supermicro servers built on B300 GPUs, moved 74 of them to buyers in China through Indonesia, Japan and Hong Kong, and lost 56 to customs. The charges are breach of trust and forgery, which is what export control enforcement looks like when it reaches the people rather than the companies.

Taipei Times
Policy
■
NVIDIA licenses Poolside's model factory for $6 billion

With Poolside

A non-exclusive license to the system behind Poolside's Laguna coding models, plus a $1 billion investment at a $12 billion pre-money valuation and offers to more than 100 of its engineers. Poolside's founders told investors it was neither an acquisition nor an acquihire, since the company keeps running on its own.

Newcomer
Deal
■
Nvidia backs a $105 billion Ohio data center built for OpenAI

With SB Energy, OpenAI

Nvidia guaranteed up to $105 billion in land, power and shell financing for SB Energy's Pike County campus, plus a $1.5 billion stake in SB Energy itself. OpenAI signed a 20 year lease on capacity that reaches 8 gigawatts once built out, with the first phase due online in 2028.

NVIDIA
Deal
■
Nvidia turns compute into an asset class

With Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR

Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR signed memoranda to mobilize over $500 billion of third-party capital for AI data centers, underwritten against the GPUs themselves. Larry Fink compared it to the invention of mortgage-backed securities, a comparison with two endings.

NVIDIA
Funding
■
NVIDIA launches the Vera Rubin platform

Seven chips in full production, a rack pairing 72 Rubin GPUs with 36 Vera CPUs, and a claim of ten times Blackwell's inference throughput per watt. Jensen Huang told GTC he saw a trillion dollars of orders through 2027 behind it.

NVIDIA
Hardware
■
NVIDIA puts another $2 billion into CoreWeave

With CoreWeave

NVIDIA bought CoreWeave stock at $87.20 a share, deepening a circular relationship in which its biggest cloud GPU customer is also one of its own investments. CoreWeave gets early access to new NVIDIA chips and help funding more than 5 gigawatts of AI data centers by 2030.

NVIDIA
Funding
■
NVIDIA puts the six-chip Rubin platform into full production

Jensen Huang said Rubin, the successor to Blackwell, is shipping in volume, with Vera Rubin racks cutting the cost per token to roughly a tenth of the prior platform. Cloud partners including AWS, Google Cloud, Microsoft and Oracle are due to deploy it later in 2026.

NVIDIA
Hardware

2025

■
Trump clears Nvidia to sell H200 chips into China

With White House

A reversal of the export limits the Biden administration imposed, letting approved Chinese customers buy Nvidia's H200 in exchange for 25 percent of the sales going to the US government. Nvidia's newer Blackwell chips stay restricted.

Al Jazeera
Policy
│
SoftBank sells all of its Nvidia to pay for OpenAI

With SoftBank, OpenAI

All 32.1 million shares went in October for $5.8 billion, and SoftBank's finance chief gave the reason outright: more than $30 billion had to go into OpenAI this year, so existing assets had to be sold to find it. The projected total by year end was $34.7 billion.

Fortune
Funding
■
Nvidia becomes the first $5 trillion company

Nine months after losing $600 billion in a day over the theory that efficient models need fewer chips, Nvidia closed above five trillion dollars. The market had settled on cheaper inference meaning more of it, not less.

CNBC
Hardware
│
Nvidia invests $1 billion in Nokia for AI-RAN hardware

With Nokia

NVIDIA agreed to buy $1 billion of Nokia stock at $6.01 a share, becoming the telecom equipment maker's second largest shareholder, as the two build AI-RAN radio hardware for 6G networks with T-Mobile field trials planned for 2026.

NVIDIA
Deal
■
NVIDIA and OpenAI announce a 10-gigawatt letter of intent

With OpenAI

NVIDIA and OpenAI announced a letter of intent for at least 10 gigawatts of NVIDIA systems. NVIDIA intended to invest up to $100 billion progressively as systems were deployed; the parties still expected to finalize details, rather than announcing completed disbursement.

NVIDIA
Funding
│
Nvidia puts $5 billion into Intel

With Intel

An investment and a joint roadmap for x86 processors with Nvidia graphics built in, between two companies that had spent thirty years as opposites. Intel's stock more than tripled over the following eight months.

NVIDIA
Hardware
■
China bars its tech companies from buying Nvidia AI chips

The Cyberspace Administration of China told firms including ByteDance and Alibaba to stop testing and ordering Nvidia's RTX Pro 6000D, escalating from informal discouragement in August to a direct ban. Nvidia still dominates AI chips domestic makers have not matched.

TechCrunch
Policy
│
Nvidia's Jetson Thor robotics computer goes on sale

A Blackwell-based developer kit for robots starting at $3,499, with 7.5 times the AI compute of its predecessor. Aimed at humanoid and industrial robotics startups building on Nvidia's stack rather than at data centers.

NVIDIA
Hardware
■
Nvidia says it will resume H20 chip sales to China

Nvidia said the US government had assured it that export licenses for the H20, the most powerful AI chip it can still legally sell into China, would be granted after CEO Jensen Huang met officials in Washington and Beijing. The stock jumped on the reversal.

TechCrunch
Policy
│
Nvidia touches $4 trillion first

Shares hit an intraday high of $164 and the market capitalization crossed four trillion dollars, a first for any company, before closing just below it. Three and a half months later the same milestone read like a waypoint.

Al Jazeera
Hardware
■
OpenAI announces Stargate UAE, a 1-gigawatt Abu Dhabi data center

With OpenAI, G42, Oracle, Cisco, SoftBank

Built with G42, Oracle, Nvidia, Cisco and SoftBank, the cluster is the first deal under OpenAI's for-Countries program and is meant to make the UAE the first country where ChatGPT is available nationwide. The first 200 megawatts are due online in 2026.

TechCrunch
Deal
■
Nvidia unveils Project DIGITS, a desktop AI supercomputer

At its CES keynote, Nvidia introduced Project DIGITS, a desktop unit built around the GB10 Grace Blackwell superchip that runs models up to 200 billion parameters from a standard outlet. Pricing starts at $3,000, with shipments planned for May 2025.

NVIDIA
Hardware

2024

│
NVIDIA cuts the price of its Jetson Orin Nano developer kit

A software update unlocked up to 1.7 times more generative AI performance on the existing Jetson Orin Nano board, which NVIDIA renamed the Orin Nano Super and dropped in price from $499 to $249, aimed at hobbyists running small models at the edge.

NVIDIA
Hardware

2023