Founded for just 6 months, it has a valuation of 16.7 billion.

In mid-October, Reuters, citing two people familiar with the matter, reported that Nuvacore, a chip startup established only six months ago, is raising around 200 million U.S. dollars at a valuation of about 2.5 billion U.S. dollars (equivalent to about 16.7 billion yuan) to develop a new central processing unit for AI data centers. This round of financing has not been closed yet, and the valuation and amount may still change.

It is worth noting that Nuvacore has no products, no revenue, and no performance test reports at all, and it has previously completed a seed round led by HSG. Now it is raising financing at a unicorn valuation, which inevitably makes people focus their attention on the "people" behind the company.

Nuvacore has three core figures: Gerard Williams, John Bruno, and Ram Srinivasan. The three form a team that has been developing CPU cores at Arm, Apple, and Qualcomm for more than 20 years, among whom Williams has the most impressive resume: during his tenure as chief CPU architect at Apple, he worked on multiple generations of self-developed chips from A7 to M1.

Those who know Apple may understand the significance of these two generations of chips to Apple. The A7 is Apple's first 64-bit mobile chip, and it is also the first large-scale shift to the 64-bit architecture in the smartphone industry. With this chip, Apple pulled ahead of its competitors by a full generation in terms of performance; the M1 is the first chip for Apple's Mac product line to switch from Intel to self-developed chips. In the following four years, the shipment structure, gross profit margin, and chip discourse power of Mac have all been rewritten. These two chips respectively defined Apple's two architectural transitions in the mobile and PC sectors, and Williams was involved in both.

The Third Venture of Three Seasoned Chip Veterans

The story of Nuvacore's founding team dates back to Arm.

Williams worked at Arm for 12 years and was the technical lead of Cortex-A8. This is Arm's first superscalar core, which was later used on the iPhone 3GS. He also participated in key architectures such as Cortex-A15. Around 2010, he joined Apple, and served as Senior Director of Platform Architecture for nearly ten years, responsible for CPU and SoC development, and is the inventor of more than 60 Apple patents.

In February 2019, Williams left Apple and co-founded Nuvia with John Bruno and others, aiming to develop high-performance Arm CPU cores for servers.

This startup only lasted for two years, but the result was not bad: the company raised a total of 293 million U.S. dollars in financing, and was acquired by Qualcomm for 1.4 billion U.S. dollars in cash in early 2021. Its design later became the basis of Qualcomm's Oryon architecture. Williams joined Qualcomm with the transaction and served as Senior Vice President of Engineering.

It was also during the Nuvia period that the team first encountered troubles in architecture selection. In 2022, Arm sued Qualcomm over how the Nuvia license would be used after the acquisition. After several court hearings with wins and losses on both sides, there is still no unanimous ruling on the Nuvia license issue, and the follow-up still needs to be handled by the court.

The turning point came in early 2026, when Williams announced his departure from Qualcomm. Soon, he reunited with two old partners — John Bruno (whose career started at ATI, joined AMD after ATI was acquired by AMD, then worked in Apple's platform architecture team for five years, and later served as a system architect at Google) and Ram Srinivasan (who worked as a performance architect at Intel for seven years, then was responsible for system performance architecture at Apple, joined Nuvia, and entered Qualcomm through the acquisition) to establish Nuvacore.

This time, their slogan is more ambitious, claiming to "rewrite the rules of silicon chips". The team expanded rapidly thereafter. From April to September this year, the company successively appointed senior vice presidents of software, operations and marketing, among whom David Williamson, former head of Apple's CPU engineering, served as Senior Vice President of Hardware Engineering.

On September 29, the company announced its technical roadmap on social platforms: the first core design is named WarpCore, which adopts a "Core First" strategy opposite to the traditional chip development process — first design core IPs such as execution units, branch predictors, caches and memory subsystems, and finally decide whether to adopt the x86, Arm or RISC-V instruction set. The company posted on the same day: "We said we would rewrite the rules of silicon chips. We mean it."

This strategy is quite subversive to industry conventions. But it has its technical basis: modern high-performance CPUs translate software instructions into internal micro-operations before execution, and the instruction decoder is the only module in the entire chip that is bound to the instruction set. Jim Keller, a well-known figure in the chip design industry, mentioned in a speech in 2022: "All modern computers are essentially RISC machines internally, and the only thing you need to change is the decoder." Nuvacore's statement is that more than 95% of the core basic IP can be developed before the instruction set is selected.

However, this idea has had unsuccessful precedents in the industry. In the years around 2014, AMD once advanced both x86 and Arm lines, namely the SkyBridge platform and K12 core, but eventually stopped all of them due to scattered resources and turned to Zen. But Nuvacore's approach is the opposite of AMD's: instead of working on two instruction sets at the same time, it does not choose any instruction set first, and makes the choice after the back-end matures.

There is no answer yet as to which of the two routes is correct. But the debate over the route ultimately depends on who executes it. For a team that has already delivered the Nuvia results, the market is at least willing to give it a chance first.

Six Months, Valuation Nearly Doubled

Nuvacore's first financing took place in April this year, led by HSG, with the amount undisclosed; the second round is the current one: seeking more than 200 million U.S. dollars in financing at a valuation of about 2.5 billion U.S. dollars.

As a comparison, Nuvia, the previous project of the same team, took about 19 months to reach the Series B round, and the final exit price was: Qualcomm acquired it for 1.4 billion U.S. dollars in cash. Only six months after Nuvacore's seed round, the valuation discussion has reached 2.5 billion U.S. dollars, which is equivalent to 1.8 times the exit price of the entire previous startup cycle. Moreover, Nuvia's price was based on the existing product "Phoenix" that had entered the internal benchmark test, while Nuvacore has no such product.

Why is the second startup of the same three people nearly twice as expensive as the first one? This has to be viewed from the results of the last startup.

In fact, Nuvia's products were not officially launched either. At the time of the acquisition, its Phoenix core had only entered the GeekBench internal test phase, and the server SoC "Orion" was still under development. But the 1.4 billion U.S. dollar price tag turned out to be a good deal afterwards, because this design was commercialized within Qualcomm, which is today's Oryon CPU.

Looking through Qualcomm's official financial reports, the achievements of Oryon basically support its three product lines of PC, automotive, and flagship mobile phone chips: in the first quarter of fiscal 2026, Qualcomm recorded a record single-quarter revenue of 12.3 billion U.S. dollars, of which automotive business revenue was 1.1 billion U.S. dollars for the quarter, a 15% year-on-year increase, continuously refreshing the record; IoT business revenue was 1.69 billion U.S. dollars, a 9% year-on-year increase; in the same quarter, Qualcomm announced that it and its partners would launch about 150 new models equipped with Snapdragon X series PC processors within this year, and the new generation product X2 Elite Extreme has become the fastest Snapdragon laptop platform to date.

By the third fiscal quarter, Qualcomm's automotive business has maintained double-digit year-on-year growth for 23 consecutive quarters, and the combined revenue of automotive and IoT increased by 28% year-on-year; the CEO gave a target in the financial report, expecting that non-mobile business revenue will reach 40 billion U.S. dollars by fiscal 2029, and the growth rate of non-mobile business including data centers will accelerate from 24% in fiscal 2026 to more than 60% in fiscal 2027.

It can be said that the achievements of Oryon have basically changed the CPU architecture of Qualcomm's product structure. Therefore, Nuvacore's 2.5 billion U.S. dollar valuation reflects the possibility that this team can replicate the success of Nuvia in server CPUs once again.

The Market Is Full of "$2.5 Billion Valuation" Stories

In terms of chip financing, the market rhythms at home and abroad are completely different.

The domestic primary market has cooled down for high-profile large-scale chip projects. Data from China Venture Research Institute shows that the total semiconductor financing in the first half of 2026 fell by about 41% year-on-year, and the number of transactions dropped by 35% year-on-year. The difficulty of raising funds for large-scale later-stage rounds has increased significantly, the project financing cycle has lengthened, and the bargaining power has shifted from sellers to buyers.

Another set of data from China Venture Jiachen CVSource shows that as of the end of July in the first half of 2026, there were 912 financing transactions in the semiconductor track, covering 823 enterprises, with a total amount of 1.627 trillion yuan. The absolute scale is not low, but the structural feature of "capital concentrating on leading players and specific segments" is obvious.

The situation overseas is just the opposite: global semiconductor startup financing is still at a high level, and the density of large rounds and the valuation jump of leading projects are much more aggressive than in China.

Specifically for leading projects in the AI chip track, the pricing has entered another order of magnitude. The latest reference comes from Tenstorrent. This RISC-V chip company led by Jim Keller announced on its official website on October 19, 2026 that it has completed a 1.5 billion U.S. dollars Series E financing at a valuation of 6.2 billion U.S. dollars.

Etched is even more outstanding: it previously completed financing at a valuation of 21 billion U.S. dollars, and according to TechCrunch reports, the company's recent investment offers have reached a valuation of 40 billion U.S. dollars or more.

Another more comparable company is Rivos: it was also founded by former Apple chip team members, also targets data centers, and was also involved in a lawsuit with Apple: Apple sued Rivos for stealing trade secrets in 2022, and the two parties settled in February 2024. Rivos completed a 250 million U.S. dollars Series A round in April 2024 (led by Matrix Capital, with participation from Intel Capital, MediaTek, Dell Technologies Capital, BlackRock, etc.). In October 2025, Meta announced the acquisition of Rivos at a valuation of about 2 billion U.S. dollars.

Putting these transactions together with Nuvacore, the 2.5 billion U.S. dollars valuation is not abrupt in the pricing sequence of overseas chip startups.

However, the high-level operation overseas is largely driven by strategic investments from semiconductor giants. According to Crunchbase data, in the first five months of 2026, global semiconductor startups raised a total of about 10.7 billion U.S. dollars, which is close to the 12.2 billion U.S. dollars for the whole year of 2025. More intuitively, the density of large transactions: semiconductor giants participated in more than 60 startup financing rounds of over 100 million U.S. dollars this year, of which 16 rounds reached a valuation of 1 billion U.S. dollars or more. The total scale of startup financing participated by semiconductor companies exceeds 2.5 trillion U.S. dollars, which is several times the previous peak.

Therefore, to be more precise, the overseas financing narrative is not that "financial VCs are in love with high-profile large-scale chip projects", but that industrial capital replaces self-development with investment and locks the supply chain with capital in the AI computing power competition. The two companies Rivos and Tenstorrent mentioned above both have the shadow of these industrial capitals or strategic buyers behind their financing.

Ten years ago, Silicon Valley VCs were far less interested in hardware than software, on the grounds of long development cycles, large capital requirements, and cyclical demand. Today, AI computing power has pushed hardware back to the table, and the tolerance for "giving valuation before having products" has also been re-priced.

This article is from the WeChat official account "ChinaVenture", written by Li Man, edited by Wang Qingwu, and published with authorization from 36Kr.