Nuvacore, a chip startup in San Jose, California, is designing a new processor for AI data centres. It is raising hundreds of millions of dollars at a valuation of about $2.5B, according to two people familiar with the fundraising who spoke to the news agency Reuters. It has no product on sale.
The venture capital firm Sequoia Capital led Nuvacore’s seed round, announced in April without a disclosed amount. The new round has not closed, and its size and price could still change. The technology news site The Information earlier reported that the company was looking to raise $200M or more. Nuvacore declined to comment.
Much of the interest centres on the founding team behind the company. Gerard Williams spent nearly a decade as Apple’s chief central processing unit (CPU) architect, designing the processors inside iPhones and Macs.
A Team That Has Done It Before
Williams co-founded Nuvia in 2019 with John Bruno and Ram Srinivasan, two other Apple chip engineers. The company raised close to $300M and was bought by the chipmaker Qualcomm for $1.4B in 2021. Nuvia’s designs became the basis of the Oryon processors that Qualcomm now uses in laptops and phones. Williams stayed on as a senior vice president of engineering at Qualcomm before announcing his exit.
The same three founders are now behind Nuvacore. Srinivasan, who specialises in designing complete chips that combine many parts on a single piece of silicon, has worked at Apple, Intel, Nuvia and Qualcomm. In April the company also hired David Williamson, another former Apple CPU engineering leader, as senior vice president of hardware engineering.
The founders’ earlier work at Nuvia also brought a legal fight. Arm, the British company whose chip designs most phone processors are built on, sued Qualcomm in 2022 over how Nuvia’s Arm-based designs were licensed after the takeover. That history is relevant to Nuvacore, because one of the biggest decisions a processor maker faces is which family of instructions its chip will understand.
What A CPU Does In An AI Data Centre
A CPU is the general-purpose brain of a computer, running the operating system and directing the work of everything else in the machine. In AI data centres, most of the heavy calculation happens on graphics processing units (GPUs), the chips Nvidia made famous. CPUs sit alongside them, moving data to and from the GPUs and running the software that coordinates each job.
That role has grown as AI agents spread, because the programs that plan tasks, call other tools and juggle many requests at once run mainly on CPUs. Demand has climbed with it. Intel has said it could not make enough chips to meet demand, and AMD has reported rising CPU sales. Nvidia launched its own Vera CPU this year and expects to sell $20B worth in the financial year ending in January.
Demand comes from the companies building ever-larger AI facilities. Morning Tick has followed Anthropic’s $518B build-out and Tencent’s move to rent 100,000 AI chips from Oracle, two examples of the scale of computing that these processors now serve.
Designing The Core Before The Language
Every processor is built around an instruction set architecture (ISA), the basic list of commands software uses to tell the chip what to do. Data centres today run mostly on x86, the design family used by Intel and AMD, with Arm-based chips made by Amazon, Microsoft and Nvidia gaining ground. RISC-V, an open design that anyone can use without paying a licence fee, is a third option.
Nuvacore has not picked one. In late September it described what it calls a “Core First” approach, in which it designs most of the processor core, the part that actually carries out instructions, before choosing which ISA the chip will speak. Its first design, called WarpCore, is being developed this way. According to the technology site Xenospectrum, the company has said more than 95% of the core’s basic building blocks can be designed before that choice is made.
In principle, the approach could let Nuvacore match the final chip to whatever its customers or partners want. An Arm version would require a licence from Arm, while a RISC-V version would need none. Nuvacore’s job adverts mention all three families, and the company has promised to “rewrite the rules of silicon” with its core. It has not yet published any measured speed, power or size results for WarpCore.
A Crowded Field For Chip Money
Nuvacore is raising money at a time when investors are pouring cash into chip startups. Semiconductor startups raised about $10.7B in the first five months of 2026, close to the $12.2B raised in all of 2025, according to the data company Crunchbase. Shares in both Intel and AMD have risen more than 180% this year.
Several of the largest recent deals have been for AI accelerators, chips built to run AI models, rather than CPUs. Etched, which makes such chips, raised $700M at a $21B valuation in August, and Nvidia agreed in December 2025 to buy assets from the AI chipmaker Groq in a deal reported at $20B. Closer to Nuvacore’s field, Efficient Computer, a Carnegie Mellon University spin-out that builds general-purpose processors, raised more than $97M at a $650M valuation on 29 September.
Paying For A Track Record
Nuvacore has not disclosed its ISA, its manufacturing process or its timeline for a first chip. Nor has it said whether it will sell finished chips or license its designs to others. Investors at a $2.5B valuation would be backing the team behind Nuvia and Apple’s processors before any results from Nuvacore’s own design are available.
Published performance figures for WarpCore would give the first outside measure of the design. Until then, the reported valuation suggests how highly investors now rate experienced CPU designers as AI data centres spread.
