ESWIN Computing

There is a long-running argument in the chip industry about whether the future belongs to Arm, whose designs power nearly every phone, or to x86, which still owns the data center and most laptops. Both come with a catch: to use them, you pay. Arm licenses its architecture; x86 is effectively a two-company club you cannot join. ESWIN Computing, a semiconductor firm headquartered in Beijing, looked at that argument and quietly walked through a third door. It builds its chips on RISC-V, an instruction set that is open, royalty-free, and - this is the part that keeps getting more interesting - increasingly capable of real work.

RISC-V is not a company. It is a specification, governed by a nonprofit, that anyone can implement without asking permission or writing a check. For most of its life that was mostly an academic virtue. The bet ESWIN is making is that the virtue is becoming a business: if you are a chip company in 2020s China, where access to the incumbent architectures is complicated by both cost and geopolitics, an open instruction set is not just philosophically nice. It is strategically load-bearing. You cannot be cut off from something nobody owns.

ESWIN Computing is the IC-and-solutions arm of the larger Beijing ESWIN Technology Group, which was founded in 2016 and formally incorporated in 2020. Its chairman, Wang Dongsheng, is not a first-time founder trying to look serious. He is the man who built BOE Technology into one of the world's largest display makers, then stepped down as its chairman in 2019 to start over in semiconductors. That is an unusual second act - the corporate equivalent of a championship coach leaving to run a start-up league team - and it gives ESWIN both capital and credibility that most RISC-V start-ups do not have.

What ESWIN actually ships is silicon. Its flagship is the EIC77 family, unveiled around 2024. The entry point, the EIC7700 and its faster sibling the EIC7700X, is a single-die system-on-chip built for the edge: four 64-bit RISC-V processors using the RV64GC instruction set, clocking up to 1.8 GHz, paired with a neural processing unit that delivers about 13.3 TOPS at int8 precision. In plainer terms, that is enough on-chip AI to run image classification, object detection, segmentation and tracking without phoning home to a server. Above it sit the dual-die EIC7702 and EIC7702X, which stitch two dies together to reach performance ESWIN pitches at RISC-V AI-class PCs.

The detail that makes the EIC7700 genuinely interesting is what is inside it. The CPU cores come from SiFive, the American RISC-V pioneer. The GPU is licensed from Imagination Technologies. The NPU is ESWIN's own. Three companies' intellectual property, from three different corners of the industry, coexisting on one piece of silicon - held together by the fact that RISC-V is a shared standard rather than a walled garden. This is the open-architecture argument made physical: you can mix best-of-breed parts because none of them is trying to lock you in.

That approach showed up most visibly in the HiFive Premier P550, a development board SiFive released and marketed as the first commercially available out-of-order RISC-V machine. The board's brain is the EIC7700X, which ESWIN designed. Then a third party, Canonical, joined to bring Ubuntu 24.04 LTS to it, so the thing boots a mainstream Linux distribution instead of some bespoke image you have to coax into life. Increased production pushed the price down to as low as $399 for the 16GB version - cheap enough that a curious engineer can buy one and find out for themselves whether RISC-V is ready. Bo Wang, vice chairman of ESWIN Computing, said the company was pleased to work with SiFive and eager to see the boards spread into the market.

Selling boards to developers is not incidental; it is the strategy. An instruction set is only as valuable as the software that runs on it, and software follows hardware that people can actually get their hands on. ESWIN understands this well enough to maintain its own public forks of Linux, U-Boot and OpenSBI on GitHub, plus a software platform it calls RISAA. The point is to make sure that when its chip lands on a desk, the operating system already knows what to do with it. Hardware companies that forget the software half tend to ship beautiful silicon that nobody can program.

None of this happens in a vacuum, and it is worth being precise about scale, because ESWIN is really a group of businesses. Alongside the chip-design arm there is a silicon-materials business that makes 12-inch wafers and an advanced packaging-and-testing operation. In October 2025 the wafer arm, Xi'an ESWIN Material Technology, listed on Shanghai's STAR Market, closed its first day up roughly 199%, and briefly carried a market value near $14.6 billion - one of China's largest IPOs of that year, and notably one of the first for a still-unprofitable company under new listing rules. That IPO was for the materials business, not the chip designer, but it tells you something about the gravity ESWIN has accumulated: the group can raise serious money on the promise of being vertically integrated from raw wafer to finished RISC-V chip.

The honest caveats belong here too. RISC-V, for all its momentum, is still climbing toward the software maturity that Arm and x86 spent decades building. ESWIN's own group finances have been described publicly as unprofitable, which is common for capital-hungry chip firms but is not nothing. And "13.3 TOPS" is a real number for edge AI, not a data-center number - ESWIN is aiming at phones, cars, smart homes, robots and PC-class devices, not at competing with the biggest AI accelerators. What it is trying to prove is narrower and, arguably, more useful: that open silicon can be good enough, cheap enough and available enough to become a default rather than a curiosity.

That is the whole wager, and it is a patient one. ESWIN is not promising to win the chip wars. It is betting that the terms of the war are changing - that "who owns the architecture" is quietly becoming the wrong question, and that a company positioned around the answer nobody owns will look, in a few years, like it saw something early. Whether that is right is not yet knowable. But it is a specific, testable bet, made with real chips you can buy today, which is more than most companies with a slogan about the future can say.