NVIDIA’s next big AI play is not just another monster GPU rack. According to Wccftech, the company’s upcoming Vera CPU racks have already attracted early adoption interest from some very serious names: CoreWeave, Meta, Oracle and Alibaba.
That matters because the AI race is no longer only about who can buy the most GPUs. As agentic AI workflows get heavier — think AI systems that plan, execute tasks, call tools and handle multi-step jobs — data centres also need CPUs that can feed, manage and move all that data efficiently. Basically, GPUs are still the headline act, but CPUs are becoming the backbone that keeps the whole concert from collapsing.
The report cites GF Holdings Hong Kong, which claims those four companies have secured Vera CPU racks as early adopters. It also suggests this may only be the first wave, with more AI firms expected to place orders as CPU demand ramps up.
Alibaba being on the list is especially interesting. China faces restrictions on access to NVIDIA’s latest AI chips, including Rubin-class hardware, so seeing a major Chinese AI player linked to Vera CPU adoption hints at how tight the global AI supply chain has become. If companies cannot get every accelerator they want, the surrounding compute stack — CPUs, memory, networking and rack-level systems — becomes even more important.
For Malaysia and SEA, this is not just Silicon Valley server-room gossip. We are seeing more data centre activity across the region, especially around Johor, Singapore-linked infrastructure and cloud expansion. When hyperscalers like Meta, Oracle and AI cloud players such as CoreWeave start locking in new platforms, it eventually affects what developers, startups and enterprises here can access through cloud services. Faster AI infrastructure can mean better tools, lower latency and more capable AI products — but if supply gets squeezed, pricing may stay painful for smaller teams.
NVIDIA is positioning Vera as a serious data centre CPU, not just a sidekick. The chip is built on NVIDIA’s custom Arm architecture, codenamed Olympus, and is listed with 88 cores and 176 threads using NVIDIA Spatial Multi-Threading. It also supports a 1.8 TB/s NVLink-C2C coherent memory interconnect, up to 1.5 TB of system memory, 1.2 TB/s memory bandwidth through SOCAMM LPDDR5X, and rack-scale confidential compute.
NVIDIA claims Vera delivers very strong single-thread performance, high data output and excellent energy efficiency. The company is also pushing the fact that Vera uses LPDDR5 memory in a data centre CPU, aiming for stronger performance-per-watt. Compared with Grace, Vera is said to offer 2x performance in areas like data processing, compression and CI/CD workloads.
The bigger play is that Vera will not only appear inside the Vera Rubin platform. NVIDIA also expects to ship Vera CPUs as standalone products, opening what it sees as another multi-billion-dollar business beyond its GPU-heavy rack systems.
There is a memory angle too. Since Vera can support up to 1.5 TB of LPDDR5X-based system memory, large-scale adoption could increase pressure on DRAM supply. For PC gamers, that does not instantly mean RAM prices naik tomorrow, but when AI data centres start eating huge amounts of advanced memory, the wider hardware market usually feels the ripple eventually.
The report says Vera Rubin is expected to launch in the second half of the week, with mass production of initial racks starting soon. For now, the takeaway is simple: NVIDIA is trying to own more of the AI data centre stack, and the early customer list suggests the market is already listening.
Source: Wccftech Gaming