Jensen Huang

Nvidia signed the definitive agreement on September 2 and announced it the next morning: about $11.9 billion to Hugging Face stockholders plus up to $1 billion in retention equity, closing expected in the first half of 2027 subject to US and EU regulatory approvals. Huang committed that Nvidia compute would not be required to build on or deploy through Hugging Face and that it would keep supporting other silicon vendors, the openness question regulators are expected to focus on. CEO Clément Delangue said he had approached Huang weeks earlier, after Hugging Face was hacked.

Second-quarter fiscal 2027 results showed revenue of $96.2 billion, up 106% year on year, with data center revenue of $89.0 billion. Jensen Huang said Vera Rubin was in full production, and the release noted racks running at partners.

Speaking to press at GTC 2026, Jensen Huang said Nvidia had been licensed for many customers in China for H200, had received purchase orders from many of them, and was restarting H200 manufacturing.

DLSS 5 was announced as a real-time neural rendering model that infuses pixels with photoreal lighting and materials, which Nvidia billed as its biggest graphics breakthrough since real-time ray tracing in 2018. Jensen Huang framed it as blending handcrafted rendering with generative AI. The company committed only to a autumn launch window.

Nvidia announced that the seven chips of the Vera Rubin platform were in full production and described the platform as five racks and one supercomputer. This kept to the second-half-2026 timing set a year earlier.

The largest investment Nvidia had ever proposed: a non-binding agreement to put up to $100 billion into OpenAI progressively as each gigawatt of Nvidia systems is deployed, starting with the first gigawatt on Vera Rubin in the second half of 2026. Critics called it circular financing; the definitive terms were still being negotiated months later.

Nvidia agreed to buy $5 billion of Intel common stock at $23.28 a share and to co-develop custom x86 CPUs linked to Nvidia GPUs over NVLink for data centers and PCs, a month after the US government took its own 10 percent stake in Intel. The same day CNBC reported the Enfabrica acquihire.

Nvidia agreed to buy US$5 billion of Intel stock at US$23.28 a share. Intel will build Nvidia-custom x86 CPUs for Nvidia's data center platforms and x86 system-on-chips with Nvidia RTX GPU chiplets for PCs, connected with NVLink.

After Jensen Huang met officials in Washington and travelled to Beijing, Nvidia said the U.S. government had assured it that H20 export licenses would be granted and that it hoped to start deliveries soon. It also announced an RTX Pro GPU aimed at the China market.

In the GTC 2025 keynote, Nvidia dated the next two generations. Vera Rubin NVL144 systems were slated for the second half of 2026 and Rubin Ultra systems for the second half of 2027.

Fourth-quarter fiscal 2025 results reported data center revenue of $35.6 billion. Nvidia said Blackwell AI supercomputers achieved billions of dollars in sales in their first quarter, without giving an exact figure in the release.

In its third-quarter fiscal 2025 results, Nvidia described Blackwell as being in full production and reported record data center revenue of $30.8 billion, up 17% from the prior quarter.

On the earnings call, CFO Colette Kress said Nvidia had changed the Blackwell GPU mask to improve production yields and that the production ramp would begin in the fourth quarter. Jensen Huang said Nvidia meant to ship product, not merely start production, in Q4. Nvidia also said Blackwell demand was well above supply.

In his Computex keynote in Taipei, Jensen Huang showed the Rubin platform as Blackwell's successor, together with an Arm-based CPU called Vera and new networking parts. He described a one-year release rhythm for the whole data center stack. No launch year was given in Nvidia's own write-up.

Nvidia announced the Blackwell architecture in San Jose, listing the B200 Tensor Core GPU and the GB200 Grace Blackwell Superchip, which pairs two B200 GPUs with a Grace CPU. The release claims up to 25x lower cost and energy for LLM inference compared with H100.

After Nvidia's quarterly results, Jensen Huang told Reuters the company was sampling two new China-market AI chips with customers and that both complied with the rules without needing a licence. He did not name the chips or the customers. The move was Nvidia's attempt to defend a market position that the October 2023 controls had put at risk.

With first-quarter fiscal 2024 data-center revenue at a record $4.28 billion, Jensen Huang said the computer industry was going through two transitions at once, accelerated computing and generative AI, and that Nvidia was significantly increasing supply to meet surging demand. The guidance that followed became the clearest public marker of the H100 shortage.

Nvidia said it had modified its flagship H100 into a version legal to export to China, the H800, and that the cloud computing units of Alibaba, Baidu and Tencent were already using it. It was the same move as the A800 applied to the newer generation, and Jensen Huang pointed Chinese startups building large language models at those clouds.

Seventeen months after signing, the companies terminated the agreement, citing significant regulatory challenges, and Nvidia wrote off the $1.36 billion it had prepaid. Arm listed on Nasdaq in September 2023 at a $54 billion valuation; Nvidia kept a 20-year Arm architecture licence and went on to build its own Arm-based Grace CPUs.

The largest chip deal ever proposed would have put the architecture behind nearly every smartphone under the company that dominated AI accelerators, with SoftBank keeping under 10 percent of Nvidia. Huang promised to keep Arm's open licensing model and its Cambridge headquarters; customers including Qualcomm, Google and Microsoft objected within months.

The acquisition completed at a transaction value of about $7 billion, adding Mellanox's roughly 3,000 employees and the InfiniBand, Spectrum Ethernet and BlueField DPU lines that became the networking half of Nvidia's data center segment. Within five years networking was a multi-billion-dollar quarterly business inside Nvidia.

Nvidia's first big acquisition, at $125 a share in cash, brought in the Israeli maker of InfiniBand and Ethernet interconnects that already sat inside most of the world's top supercomputers, after reports that Intel and Xilinx had also bid. Jensen Huang framed it as uniting compute and networking for data-center-scale computing; the deal needed approvals in the US, EU, Mexico and China.

At GTC 2017 Nvidia launched the Volta platform and the Tesla V100, with 640 Tensor Cores delivering 120 teraflops of deep-learning performance, which it equated to 100 CPUs. Tensor Cores made matrix math a first-class hardware feature, the design choice every later generation has extended.