TLDR
- Nvidia stock rose 1.6% in premarket trading Wednesday, reaching $177.97
- Arm Holdings launched its first data center CPU, the Arm AGI CPU, targeting $15B in annual revenue by 2031
- Arm’s CPU is not seen as a direct rival to Nvidia’s GPUs — but may compete with Nvidia’s stand-alone Vera CPUs
- AWS agreed to buy 1 million Nvidia GPUs for AI inference, a deal likely worth over $50 billion
- Nvidia is resuming production of H200 chips and plans a China-compliant version of its new Groq 3 chips, worth an estimated $32B annually
Nvidia stock climbed in early Wednesday trading, shrugging off news that Arm Holdings had entered the AI chip market for the first time. The move came alongside two other major revenue developments that had been largely under the radar.
Arm announced its first-ever data center CPU on Tuesday — the Arm AGI CPU — with Meta Platforms and OpenAI named as early customers. Arm set out ambitious targets in a post-market presentation, projecting the CPU business will generate around $15 billion in annual revenue by 2031, as part of a broader $25 billion revenue goal.
Arm’s stock jumped 12% in premarket trading following the announcement.
But analysts were quick to note that the new chip doesn’t go head-to-head with Nvidia’s core GPU business.
Benchmark Research analyst Cody Acree wrote that Arm’s move is “less about catching up to the accelerator wave and more about inserting itself deeper into the architecture that governs how AI infrastructure actually runs.”
Nvidia CEO Jensen Huang was even quoted in Arm’s launch materials, calling the two companies’ partnership of nearly two decades a foundation for “one seamless platform, from cloud to edge to AI factories.”
Where it does get interesting is around Nvidia’s stand-alone Vera CPUs, launched just last week at its developer conference. J.P. Morgan analyst Harlan Sur flagged that Arm’s chip could overlap with that product line. He also pointed out that Meta has already signed a deal with Nvidia for its Arm-based CPUs — making the competitive picture a little muddy.
AWS Goes All-In on Nvidia
Meanwhile, Amazon Web Services announced it would purchase 1 million Nvidia GPUs to power its AI inference operations. This came as a bit of a surprise — AWS had previously been positioned as home to “the largest cluster of non-Nvidia chips in the world” after its Indiana data center went live in October 2025.
The deal also includes a “broad mix” of six additional Nvidia chip types, including the new Groq 3 inference chips, plus Nvidia networking hardware. The full package is expected to cost well north of $50 billion and is set to close by end of 2027.
That single deal represents roughly 25% of Nvidia’s entire 2025 annual revenue.
China Revenue Back on the Table
CEO Jensen Huang last week confirmed that Nvidia is restarting production of its H200 chip — designed to comply with U.S. export restrictions — for the Chinese market. Reports also indicate that a China-compatible version of the Groq 3 chip is in the works.
Nvidia had included zero Chinese data center revenue in its Q1 guidance. In 2025, those sales were running at roughly $8 billion per quarter — around $32 billion annually, or about 15% of total 2025 revenue.
Combined, the AWS deal and China restart represent more than $82 billion in revenue that wasn’t baked into Nvidia’s existing projections.
Nvidia stock was up 1.6% at $177.97 in premarket trading Wednesday, recovering from a 0.3% dip the prior session.







