Reuters: Nvidia Pitches Vera CPU to Chinese Customers, Shipments Could Begin as Early as August
Miles Bennett
Nvidia has told Chinese clients its first standalone CPU, Vera, could ship as early as August — a move to rebuild China revenue through CPUs after GPU exports hit a wall.
What is Vera, and why is Nvidia pushing it now?
Vera is Nvidia's first standalone CPU for agentic AI — AI systems that plan and execute multi-step tasks on their own. Built on Arm architecture, Nvidia says it runs roughly 1.8× faster than comparable rivals.
This means → Vera is not a GPU (the parallel-computing chip Nvidia is famous for) but a CPU (the general-purpose chip that orchestrates workloads) — a deliberate new product line.
CEO Jensen Huang expects Vera to become the company's next multi-billion-dollar business. It launched in March; Chinese clients can now place orders.
Why target China right now?
Nvidia's GPU business in China has effectively collapsed. Huang himself said in October that U.S. export controls and China's push for self-reliance had driven Nvidia's China share to virtually zero.
Its second-strongest AI chip, the H200, has been stuck for months: the U.S. government issued purchase licenses to about ten Chinese firms, but none has completed delivery — Chinese regulators have not cleared the shipments.
In plain terms = the GPU road is blocked, so Nvidia is probing with a CPU product that sits outside the restricted list, testing whether it can reopen a China revenue stream.
Are Chinese buyers interested?
Reuters reports one major Chinese cloud company plans to order over 300 servers, each with two Vera CPUs, initially for testing — with a decision on bulk procurement to follow.
Alibaba and ByteDance were previously named by Nvidia as Vera deployment partners, but neither responded to requests for comment.
This means → the market is at the "trial" stage — big clients show interest but no commitment.
What stands in the way of large-scale adoption?
Sources cite three uncertainties: whether the software ecosystem is mature enough, compatibility with existing systems, and the difficulty of migrating workloads built around domestic Chinese AI chips.
In plain terms = Chinese clients have spent years adapting to homegrown chips; switching back to a new Nvidia platform carries real migration costs.
This reflects a deeper shift: export controls didn't just block Nvidia's GPUs — they gave Chinese buyers their own technological inertia, and reversing course is not straightforward.
How much does Vera cost?
According to research firm SemiAnalysis, a single Vera processor is priced well above $20,000 before volume discounts. A full 256-chip rack runs about $10 million, varying by memory configuration.
Most chips are initially allocated to the full-rack format preferred by hyperscale cloud providers. The two-processor server version is expected to ramp later.
This means → Nvidia's first target customers are the largest cloud companies; smaller buyers will have to wait.
Content is for reference only, not financial advice.