Nuvacore's Williams lands $2.5B for a chip that doesn't exist

Nuvacore Raises Millions at $2.5B Valuation for a Chip That Doesn't Exist

There is no chip. No benchmarks, no tape-out date, no process node, no factory line with Nuvacore's name on it. What there is: a six-month-old startup in San Jose, three famous engineers, a PowerPoint-worthy idea about how to design processors backwards — and, according to a Reuters exclusive published Friday, a line of investors ready to hand it hundreds of millions of dollars at a valuation of roughly $2.5 billion.

If that sounds absurd, you're not wrong. But you're also, probably, missing the point. Nobody is buying silicon here. They're buying the people who built the silicon inside your iPhone — and betting that lightning strikes twice.

File photo: a processed silicon wafer covered in individual dies. Nuvacore has not published any images of its WarpCore design — the company has no product yet. Photo: Inductiveload, public domain via Wikimedia Commons.

The number that stopped the room

The details come from two people familiar with the fundraising, speaking to Reuters reporters Max A. Cherney and Krystal Hu. The headline facts: Nuvacore is raising "hundreds of millions of dollars" at a roughly $2.5 billion valuation to develop a new central processor for data centers. The valuation figure had not been previously reported. The Information had earlier reported the company was raising $200 million or more — though the original report is paywalled and we couldn't independently verify its date.

Two caveats matter enormously, and both come straight from Reuters' own sourcing. First: the round has not closed, and the two sources "cautioned that the valuation and size of the capital raise could change." Second: Nuvacore declined to comment. So treat the $2.5 billion as what investors are currently willing to pay, not what the company is definitively worth.

We checked, by the way — there is no press release, no website statement, no funding filing attached to this raise. Beyond a single X post on September 29, the company is operating in near-total stealth. This is a fundraise happening by whisper network, and the whispers are getting louder.

The men behind the money

Nuvacore was founded by Gerard Williams, John Bruno and Ram Srinivasan — Reuters calls them "renowned engineers," which in chip-industry terms is roughly the equivalent of calling Michael Jordan a basketball player.

Williams is the name that opens wallets. He previously worked on Arm's Cortex processors, then became, per Tom's Hardware, "a leading architect of [Apple's] custom silicon" — the custom cores that made Apple Silicon the thing Intel still hasn't caught up with. In 2019 he left to co-found Nuvia, a server-chip startup that raised $240 million to build an Arm data-center chip. Two years later, Qualcomm bought the whole thing for $1.4 billion.

That acquisition is the reason Williams can raise a quarter of a ten-billion-dollar valuation on a company with no product. Nuvia's technology didn't die inside Qualcomm — it became the Oryon CPU cores inside the Snapdragon X Elite, the chips powering the current wave of Windows-on-Arm laptops. As a Qualcomm spokesperson told XDA-Developers back in 2022: "The creation of our custom CPU was started by Nuvia engineers while employed at Nuvia and, after the acquisition of Nuvia by Qualcomm Technologies, the custom CPU was completed by engineers at Qualcomm Technologies."

Put less diplomatically: the chips in this year's flagship Windows laptops began life in Williams' last startup. Kedar Kondap, Qualcomm's senior vice president and general manager of Compute and Gaming, told Tom's Hardware at the X Elite launch: "Because what you will see with the Snapdragon X Elite platform is something you haven't seen before in terms of leadership, in terms of technology, performance per watt. You'll see the Qualcomm Oryon CPU distinguish itself in its capabilities and the efficiency that it will drive." He was, in effect, describing Williams' work.

Bruno and Srinivasan, per Fudzilla's reporting on the Tom's Hardware original, bring further experience from Nuvia, Qualcomm and Apple, and are described as former Apple and Nvidia architects. The team has since added David Williamson — described as a former Apple CPU engineering leader — as senior vice president of hardware engineering, plus Qualcomm veteran Anthony Scarpino. Sequoia Capital, which also led Nuvacore's seed round earlier this year, is evidently happy to double down on the same resumes.

WarpCore: building the chip before picking the rules

The one concrete thing Nuvacore has shown the world is an idea. On September 29, the company revealed its first processor design, codenamed WarpCore, along with an unconventional development strategy it calls "Core First": build the core functionality of the CPU before committing to an instruction set architecture — x86, Arm or RISC-V.

Traditionally, chip designers pick an ISA first and let its requirements shape everything else. Nuvacore's argument is that modern CPU designs already translate instructions into internal micro-operations, so most of the execution machinery can operate independently of the original instruction format. Build the fast part first, pick the language later — and meanwhile, the company says, its engineers can "avoid the limitations and requirements of specific architectures and optimize a design for the type of computations that are most common in data centers and AI infrastructure."

It's a genuinely clever pitch. It's also, as Fudzilla drily noted, "considerably more complicated than swapping one decoder for another." Instruction decoding, register handling, memory ordering and software compatibility all still impose architectural requirements, and postponing the decision doesn't make those requirements vanish — it just moves them later in the schedule. Nuvacore's hiring material references RISC-V, Arm64 and x86, the only public hint about eventual direction. The company has said nothing about which it will pick.

And critically: Fudzilla reports that "for now, Nuvacore has provided no benchmark results, manufacturing process details, cache specifications or confirmed tape-out schedule." Notebookcheck's Anil Ganti, who covered the September reveal in depth, adds that the process node, cache hierarchy, pipeline configuration, vector capabilities and tape-out schedule all remain undisclosed. Even the business model is unconfirmed — whether Nuvacore intends to sell complete processors, license reusable CPU IP, build semi-custom silicon, or some combination, nobody outside the company knows.

The hardest part isn't the chip

Here's the unglamorous truth about building a CPU in 2026: the transistors are the easy part. The licenses are the minefield.

Implementing x86 requires rights that are, in Fudzilla's words, "tightly controlled" — Intel and AMD are not in the business of handing them out. An Arm implementation would require architectural rights, which gets awkward given Arm's past legal dispute with Qualcomm over the Nuvia designs. (Arm sued Qualcomm over whether Nuvia's core designs could transfer to Qualcomm after the acquisition — the kind of precedent that makes lawyers nervous.) RISC-V is comparatively straightforward, being open — but an open ISA doesn't come with an existing software ecosystem, which is the whole reason the x86 and Arm duopoly has survived this long.

Nuvacore's Core First strategy can be read as an elegant dodge of exactly this problem: defer the licensing fight until the silicon is worth fighting over. Whether investors see it that way, or whether they just see a team that has already navigated these waters once, is an open question. Probably both.

The money is real even if the silicon isn't

Zoom out, and Nuvacore's raise is less an anomaly than a data point. Crunchbase data, via Reuters, shows investors put about $10.7 billion into semiconductor startups in the first five months of 2026 — already outpacing the $12.2 billion raised globally in all of 2025. Over the past decade, venture firms treated hardware as a bad habit: long development cycles, huge capital needs, cyclical demand. The AI boom broke that discipline.

The comparables are stacking up fast. Etched, the AI inference-chip startup, raised $300 million at a $10.3 billion valuation in July — led by Sequoia, the same firm backing Nuvacore — then followed with $700 million at a $21 billion valuation in September, led by Jane Street. Earlier this month, TechCrunch reported Etched is now fielding funding offers at $40–50 billion valuations. Those are offers, not a signed deal — offers can evaporate — but the direction of travel is unmistakable.

Then there's Positron AI, which raised $875 million at a $5 billion valuation in September, more than quadrupling its February valuation. The Reno-based company's CEO, Mitesh Agrawal, is funding the final design stage of a chip called Asimov, targeting production in the second half of 2027, with some 50 first-generation Atlas racks already shipping and more deploying at Oracle Cloud Infrastructure.

As Kleiner Perkins managing partner Mamoon Hamid put it in commentary carried by globalpolitics.info on the Etched round: "Inference is becoming one of the most important infrastructure markets in AI, and winners will be judged by how many tokens they can process per dollar and per watt." Investors aren't funding chip companies anymore. They're funding token factories.

Why CPUs are suddenly the bottleneck

The strangest part of the Nuvacore story is what it's building: not an AI accelerator, but a CPU — the unglamorous traffic cop of the data center. Reuters' reporting frames the demand bluntly: CPUs "perform vital functions as traffic controllers for Nvidia's AI chips" and run the autonomous AI software known as agents, with data-crunching demand swelling to power chatbots such as Anthropic's Claude.

The incumbents, meanwhile, are visibly straining. Intel said earlier this year it could not manufacture enough chips to meet demand — and, in a line that should be read twice, was able to sell products it had "previously determined were too faulty for sale." AMD described swelling CPU sales on a recent earnings call. Shares in both companies are up more than 180% this year. And Nvidia itself launched a CPU called Vera this year, with CEO Jensen Huang saying the company expects to sell $20 billion of it in the fiscal year ending January.

Read that again: the world's most valuable chip company is now selling CPUs, the CPU incumbents can't make enough of them, and a six-month-old startup with no product is worth $2.5 billion for promising to build one. That's not a market. That's a stampede. It's also the clearest signal yet that the AI infrastructure buildout has moved past the GPU land-grab and into the unsexy plumbing underneath it — which is where the ex-Intel veterans come in: the ex-Intel team that raised $100M to fix AI's network bottleneck is chasing the same thesis from the networking side.

Our take: this is a people bet, and people bets are the riskiest kind

Let's be honest about what a $2.5 billion valuation for a pre-product company means. It means investors have looked at every alternative — Intel's shortages, AMD's constraints, the licensing minefields, the years-long gap before any Nuvacore silicon reaches a data center — and decided the scarcest resource in the AI buildout isn't chips. It's the dozen-odd humans on Earth who have designed a world-class CPU core from scratch and shipped it.

Williams has done it twice: Apple's custom silicon, then Nuvia's cores inside Qualcomm's Oryon. That track record is genuinely rare, and it explains why Sequoia wrote the seed cheque and why the current round is reportedly oversubscribed in spirit if not yet in fact. The bull case writes itself: give the best CPU architects alive a blank sheet, no legacy ISA baggage, and a data-center market desperate for alternatives, and get out of the way.

The bear case writes itself too. The graveyard of chip startups is full of brilliant architects — brilliant architects are, in fact, the one thing chip startups never lack. What kills them is everything else: the five-year development cycles, the billion-dollar tape-outs, the licensing fights, the moment a hyperscaler decides to build the thing in-house. Nuvacore's Core First strategy is clever, but cleverness is not a moat. And a valuation set by "two people familiar with the fundraising," for a round that hasn't closed, for a company that declined to comment, is the thinnest possible foundation for the thickest possible number.

Whether this is visionary or a colossally expensive lottery ticket will be decided by one thing: tape-out. There is no tape-out date. Until there is, $2.5 billion is not a valuation. It's a compliment — the most expensive kind.

Sources

• Reuters exclusive (Oct 9, 2026), reporting by Max A. Cherney and Krystal Hu — via Channel NewsAsia and Economic Times CIO rewrites (the original is paywalled)
• The Information (via Reuters' reference) — earlier report that Nuvacore was raising $200 million or more
• Fudzilla (Oct 1, 2026), relaying Tom's Hardware — WarpCore "Core First" reveal
• Notebookcheck, Anil Ganti (~Sept 29–30, 2026) — WarpCore reveal details and hiring-material analysis
• Tom's Hardware (Oct 2022–23) — Snapdragon X Elite / Oryon lineage; Kedar Kondap interview
• XDA-Developers (2022) — Qualcomm spokesperson on the Nuvia→Oryon lineage
• TechCrunch, Marina Temkin (Oct 5, 2026) — Etched $40–50B funding offers; July/September Etched rounds via BigGo Finance aggregation
• Reuters (Sept 10, 2026), reporting by Prathik Jayaprakash — Positron AI $875M raise
• globalpolitics.info (~Aug 2026) — Mamoon Hamid (Kleiner Perkins) remarks on the inference market
• Related TechPulse USA coverage: Etched fielding funding offers at $40B; TypeSafe's $870M raise at a 25% haircut

We cross-checked the Reuters exclusive's key claims across the Channel NewsAsia and Economic Times CIO rewrites this morning — they match verbatim on the figures that matter ($2.5B valuation, hundreds of millions, round not closed, declined to comment). The original Reuters piece is paywalled; The Information's earlier "$200 million or more" report could not be independently dated. Deliberately excluded: any precise founding date for Nuvacore (only "six-month-old" per Reuters is verified), the seed round's size, exact founder job titles at Apple, and any performance claims about WarpCore — none exist.

By Daniel Rao — Daniel Rao covers AI funding, frontier labs and the economics of machine intelligence for TechPulse USA. He has previously reported on the Isomorphic Labs raise, the Firmus IPO withdrawal, and the decision-model benchmark wars.