Nuvacore with Sequoia backing develops new CPU for AI workloads and data centers

The battle for AI infrastructure goes beyond GPU: startup Nuvacore wants to prove that the next bottleneck in data centers is the CPU. The company of former Apple architect Gerard Williams is betting on a processor originally designed for constant and heavy AI workloads, not adapted to them after the fact.

According to Bloomberg, Nuvacore has already attracted support from Sequoia Capital and is completing its seed round, after which it expects to launch Series A in the coming months. The startup is also opening offices in Silicon Valley, Austin, and Canadian Markham—major hubs of semiconductor expertise.

The company has just emerged from obscurity, but states its ambition very clearly: to create a more powerful central processor for the AI era. At Nuvacore, they believe that GPU dominance alone is no longer sufficient.

Graphics accelerators remain the primary computational engine for model training and inference, but it is the CPU that manages data flow, services system software, and coordinates memory, networking, and accelerator operations. If this layer becomes a bottleneck, the overall efficiency of the AI cluster drops.

Nuvacore itself describes the product as a universal CPU core created from scratch for data centers and AI infrastructure. The main goals are maximum performance and high efficiency in chip area utilization.

The company separately emphasizes that it is not targeting ordinary server tasks, but rather sustained, continuous, and computationally intensive workloads, including advanced AI systems and agentic scenarios.

This is an important emphasis: the market increasingly looks not at peak benchmarks, but at how hardware performs under continuous load, where not only speed matters, but also energy efficiency, packaging density, and total cost of ownership.

Behind the project stands a team with serious credentials. Gerard Williams is one of the most renowned CPU architects of his generation: at Apple, his name is associated with developing high-performance ARM cores from the A7 era through the line that ultimately led to the M1 family of chips.

Alongside him, Nuvacore was founded by John Bruno and Ram Srinivasan—veterans of Apple, Nuvia, and Qualcomm. Previously, this team had already launched Nuvia, a startup for server processors, which Qualcomm acquired in 2021 for approximately $1.4 billion.

The new launch shows that the team wants to play once again at the level of fundamental computing architecture, rather than simply integrating into others' roadmaps.

Against this backdrop, betting on the CPU no longer looks like a niche idea, but part of a broader shift. The largest cloud providers and hyperscalers have long been building their own silicon for specific workloads, because generic solutions do not always provide the best balance between performance, power consumption, and cost.

Nevertheless, the focus is usually on GPUs, while CPUs are perceived as a mature and nearly exhausted category. Nuvacore disputes precisely this thesis: if AI services are becoming permanent, autonomous, and ubiquitous, then the central processor must also be designed for such a mode from day one.

The company has not yet disclosed key technical details, including the chosen instruction set architecture, nor has it definitively decided whether it will sell finished chips or license the design.

But it is already clear that the initial focus is data centers, where the cost of error is high, and even small gains in efficiency quickly translate into big money.

The main takeaway is simple: Nuvacore is not just another broad-based AI startup, but an attempt to redefine the role of the CPU in the AI stack. A finished product is still far away, and the risks for such a project are enormous, but the very fact of its emergence shows that the next race in AI infrastructure will not only revolve around GPUs and accelerators, but also around which processor controls the entire system.