Nvidia: AI deals, investors and valuation

Nvidia is negotiating to guarantee Nscale's debt for a 460 MW data center to be used by Anthropic.

Nvidia subscribes to $3 billion of the share offering with which Meta raises $95 billion for AI (AMD $1 billion); in return Meta expands Vera Rubin and Vera CPUs.

A SpaceX affiliate buys Nvidia Vera Rubin chips and leases them to the group, with ~$40 billion of Apollo-led debt; leasing from the first half of 2027.

Nvidia commits $1 billion in convertible notes in Nscale's $3.36 billion pre-IPO round; Nscale leases capacity to Microsoft and Anthropic.

According to Reuters, Nvidia is weighing becoming an anchor investor in Anthropic's IPO, which aims to raise up to $100 billion at a valuation of about $2 trillion. In March Huang had said the 2025 investment would probably be the last.

Korea's science ministry plans ₩3.85tn for 140 Nvidia Vera Rubin NVL72 racks (10,080 GPUs), ₩27.5bn per rack.

Signed on 2 September and confirmed in an 8-K; closing expected in the first half of 2027 pending approvals. The platform will remain open and not tied to Nvidia hardware.

Riding demand, Amazon adds 2 million Nvidia GPUs (Blackwell Ultra, Rubin, Rubin Ultra) for AWS, with deliveries in 2027-2028.

According to a letter to investors, Nvidia pays $6 billion for a non-exclusive license and part of the team: the same structure used with Groq.

Alongside the license, Poolside receives $1 billion of equity at a $12 billion pre-money valuation.

Nvidia takes a stake in SB Energy with $1.5 billion and provides credit support for land, power and facilities at the PORTS-Pike campus (Ohio); Reuters reports guarantees of up to $105 billion.

Nvidia signs memoranda of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR for financing platforms to mobilize over $500 billion for buyers of Nvidia compute.

Nvidia becomes Naver's third-largest shareholder to accelerate Korean AI factories; the deal is tied to $9 billion of project financing (Brookfield). Payment expected 06/11/2026.

Long-term partnership: Nvidia invests about $5 billion in Safe Superintelligence, which gets priority access to Vera Rubin. Much of the money will flow back to Nvidia in GPU purchases.

SSI gets priority access to the Vera Rubin platform, which should multiply its compute capacity roughly tenfold.

Nvidia and SK sign a multi-year initiative of up to ~$500 billion: HBM4 supply for Vera Rubin, memory co-development and a 2 GW SK Telecom data center with Nvidia systems from 2027.

Nvidia is among the founding investors of Helix Digital Infrastructure, KKR's new data center company.

In return, IREN will install up to 5 GW of Nvidia DSX infrastructure at its sites, starting with the 2 GW Sweetwater (Texas) campus.

Multi-year partnership for three new fiber and optical component plants in North Carolina and Texas.

Marvell links its custom chips (XPUs) and scale-up networks to the Nvidia ecosystem through NVLink Fusion.

At GTC, AWS announces more than 1 million Nvidia Blackwell and Rubin GPUs from 2026 to 2027, plus ConnectX and Spectrum-X networking.

At GTC Jensen Huang confirmed Samsung Foundry is making Nvidia's Groq 3 LPU on 4nm, shipping in H2 2026.

Nearly free warrants plus a partnership on AI factories and inference; the stake rises to 9.3% per a 13G.

Nvidia makes a significant, unquantified investment in Mira Murati's lab as part of a multiyear partnership.

Thinking Machines commits to at least one gigawatt of Nvidia Vera Rubin systems starting in 2027. Value undisclosed.

Nvidia is among the strategic backers of AMI Labs' $1.03B seed round.

Round led by Aker and 8090 Industries with Citadel, Dell, Jane Street, Nokia, Point72 and Nvidia.

Investment for a new US fab making lasers for data center optics, with a multibillion-dollar Nvidia purchase commitment and rights to future capacity.

Second half of the $4 billion optics package: lasers and optical networking components for AI factories.

$30 billion equity investment in the $110 billion round, replacing the $100 billion letter of intent of September 2025.

Alongside the investment, OpenAI commits to 3 GW of dedicated inference capacity and 2 GW of training on Vera Rubin systems.

The British autonomous driving startup raises with Mercedes, Stellantis, Nissan and existing backers Uber, Microsoft and Nvidia.

India's Yotta is investing over $2B in a 20,736-GPU Nvidia Blackwell Ultra supercluster in Greater Noida; over 10,000 B300s go to the IndiaAI Mission.

Inside the same supercluster Nvidia will run one of Asia-Pacific's largest DGX Cloud clusters under a four-year deal worth over $1B - the GPUs it sold come back as a contract.

Multiyear partnership: Meta buys millions of Blackwell and Rubin GPUs, adopts Grace CPUs at large scale (Vera from 2027) and Spectrum-X networking. No amount disclosed.

Private placement that doubles Nvidia's stake; target of 5 GW of new data centers with Rubin chips, with Nvidia helping to find land and power.

Beijing told Alibaba and other tech firms to prepare H200 orders; Alibaba told Nvidia it wants more than 200,000 units, paid upfront.

xAI closes $20 billion (target $15 billion) to expand Colossus, with Nvidia and Cisco as strategic investors. After the SpaceX-xAI merger (02/02/2026) the stake became 122.8 million SpaceX shares, about $21 billion in June 2026.

ByteDance plans to spend about 100bn yuan on Nvidia chips, mostly H200s, in 2026, up from about 85bn in 2025, if Beijing allows imports.

Non-exclusive license to the LPU inference technology, asset purchase and move to Nvidia of founder Jonathan Ross and president Sunny Madra. Groq remains independent. It is the largest deal in Nvidia's history.

Nvidia is in advanced talks to lease a data centre under construction at Mevo Carmel near Yokneam, a ~$1.5B project shared with Mega Or.

Trump cleared H200 sales to approved Chinese buyers with 25% of revenue going to the US government; Beijing approved imports in March 2026.

Stake in the chip design software leader, with a partnership on CUDA-X and agentic AI; Synopsys says it is not obliged to spend the funds on Nvidia GPUs.

At the US-Saudi forum HUMAIN announces up to 600,000 Nvidia GB300 GPUs in Saudi Arabia and the US over three years; xAI is the first customer.

Nvidia is a founding partner of Brookfield's AI program (BAIIF fund with $10 billion of equity, up to $100 billion with debt) that will build AI factories on Nvidia designs.

Three-way deal with Microsoft (up to $5 billion): Anthropic adopts Nvidia systems for the first time, up to 1 GW. Part of the funds went into the $30 billion Series G of February 2026.

In the three-way deal with Microsoft, Anthropic commits to up to 1 GW of capacity on Grace Blackwell and Vera Rubin, in exchange for Nvidia's $10 billion.

Deutsche Telekom and Nvidia are building an industrial cloud in Munich with up to 10,000 Blackwell GPUs (DGX B200 and RTX PRO), +50% of German AI compute.

At APEC Nvidia announces 260,000 Blackwell GPUs for Korea; 50,000 go to Samsung for an AI semiconductor megafactory.

The SK group (SK Hynix, SK Telecom) receives 50,000 Nvidia GPUs for an AI cloud for manufacturing and chip digital twins.

Hyundai Motor Group deploys 50,000 Blackwell GPUs for autonomous driving, robotics and factories, with a joint investment of about $3 billion with Nvidia in Korea.

Naver Cloud will operate 60,000 Nvidia GPUs for its AI services and a physical AI platform.

The Seoul government receives 50,000 GPUs for the national AI computing center and sovereign models.

Bloomberg reports an initial investment of $500 million, up to $1 billion, in a $2 billion round at a $12 billion pre-money valuation. The round did not close and in January 2026 Poolside lost the GPU cluster it had reserved.

Stake of about 2.9% tied to the AI-RAN partnership for 5G/6G networks based on Nvidia's ARC-Pro platform.

Hon Hai's board approved NT$42bn for a supercomputing center with 10,000 Nvidia GB300 chips (27MW first phase), online in H1 2026.

Bloomberg: a dedicated vehicle buys Nvidia GPUs for Colossus 2 and leases them to xAI for 5 years; Nvidia puts up to $2 billion into the equity portion.

Nvidia puts up to $2 billion of equity into the $20 billion vehicle that buys its GPUs and leases them to xAI: the money flows back to Nvidia as chip sales.

The British neocloud raises $1.1 billion with Aker, Nvidia, Nokia and Dell.

Letter of intent: Nvidia would have invested up to $100 billion as OpenAI deployed 10 GW of Nvidia systems. Never formalized; in February 2026 replaced by a $30 billion equity investment.

Hardware part of the letter of intent: at least 10 GW of Nvidia systems (Vera Rubin platform, first GW in 2H 2026). Scaled back in February 2026 to commitments of 3 GW + 2 GW.

Nvidia buys about 4% of Intel and starts joint development of x86 CPUs for data centers and PCs with NVLink. Closed on 26/12/2025 after FTC approval; in June 2026 the stake was worth about $30 billion.

Nvidia announced £2B to catalyse the UK AI startup ecosystem alongside Accel, Air Street, Balderton and others; reported recipients include Synthesia, Revolut, PolyAI and Latent Labs.

Nvidia is among investors in the round of over $1 billion that values the humanoid robot maker at $39 billion post-money.

Nvidia commits to buy all cloud capacity that CoreWeave's customers do not use through 2032: a safety net for the neocloud in which it is a shareholder.

Nvidia takes part in the ASML-led round valuing Mistral at about $14 billion.

Nvidia leases about 18,000 GPU servers (with Nvidia chips) from Lambda for $1.5 billion; it is also an investor in Lambda and becomes its first customer.

In exchange for export licences, Nvidia pays the US government 15% of revenue from H20 chip sales in China.

At VivaTech Mistral and Nvidia launched Mistral Compute, a European AI cloud with 18,000 Grace Blackwell systems in phase one (Essonne) and multi-site expansion planned for 2026.

According to the Financial Times, Oracle is buying about 400,000 GB200 chips for the Abilene (Texas) data center destined for OpenAI.

At Choose France, MGX, Bpifrance, Mistral AI and Nvidia formed a JV to build Europe's largest AI campus (1.4 GW) near Paris, with construction from H2 2026.

Preliminary US-UAE deal: up to 500,000 advanced Nvidia chips a year, 20% to G42 and the rest to Microsoft, Oracle and other US groups.

The AI vehicle of Saudi fund PIF announces AI factories of up to 500 MW with 'several hundred thousand' Nvidia GPUs; no amount disclosed.

Nvidia will produce up to $500bn of AI infrastructure in the US over four years: Blackwell wafers at TSMC Arizona, packaging by Amkor and SPIL, servers by Foxconn in Houston and Wistron in Dallas.

After DeepSeek, ByteDance, Alibaba and Tencent ordered at least $16bn of Nvidia H20 chips in Q1 2025, before the April US ban.

Nvidia takes part in Nebius's $700 million private round, taking about a 0.5% stake in the company.