A chip that can be rewired after it leaves the factory sounds like a curiosity, until a data center needs to adapt to a new AI model without respining silicon. That flexibility is why field-programmable gate arrays — FPGAs (现场可编程门阵列) — sit inside servers, base stations, and cars. Almost all of them come from three foreign giants.
Anlogic (安路科技) is one of the Chinese designers trying to close that gap. In 2025 it lost money and its revenue fell — yet it launched a multi-billion-yuan fundraising push and pointed to AI server sales growing by multiples. The story is less about this year's numbers and more about a long wager on FPGA (现场可编程门阵列) self-reliance.
Key takeaways
- Anlogic's 2025 revenue was RMB 520 million (about US$73 million / HK$565 million), down 20.22%, with a net loss of RMB 272 million (US$38 million / HK$296 million).
- FPGA chips remain 91% of sales at RMB 473 million, and gross margin rose to 43%, up 8.62 percentage points, even as revenue fell.
- From the second quarter of 2025 the business recovered quarter-on-quarter for four straight periods into early 2026, with data-center server FPGA sales growing several-fold.
- It plans to raise up to RMB 1.262 billion (US$176 million / HK$1.38 billion) via a private placement to fund advanced-process and planar-platform FPGA and FPSoC development.
- The global FPGA market is projected near US$19 billion by 2030 (about 10.5% CAGR); AMD, Intel (Altera), and Lattice hold over 85% of it today.
What an FPGA is, and why it is hard
An FPGA (现场可编程门阵列) is silicon you can reprogram in the field — arrays of logic blocks and interconnect that a customer configures for a specific task. That makes it ideal when volumes are too low for a custom chip, or when the standard changes. The catch is difficulty: you must design both the chip and the software toolchain that programs it, and the leaders have decades of both.
That is why the field is concentrated. AMD (via Xilinx), Intel (via Altera), and Lattice together command more than 85% of global FPGA revenue. Breaking in means matching not just the silicon but the ecosystem of designers and IP that surrounds it — a moat far wider than a single product.
Anlogic's actual business
Anlogic, founded in 2011 and listed on Shanghai's STAR Market in 2021, sells FPGA and FPSoC chips plus its own EDA software. Its 2025 result was sobering: revenue down a fifth and a RMB 272 million loss, deeper than 2024. The FPGA line alone was RMB 473 million (US$66 million / HK$515 million), 91% of the total.
But two signals matter more than the annual print. First, gross margin expanded to 43%, suggesting a richer product mix and less price-cutting pain. Second, after a weak start the business grew sequentially every quarter from mid-2025, and the company says data-center server sales — serving major Chinese internet firms — rose several-fold with shipments into the millions of units. That is the AI-driven demand showing up in an unlikely place.
The bet on domestic substitution (国产替代)
Anlogic's fundraise is the real tell. Up to RMB 1.262 billion would go into two projects: very-large-scale FPGAs on an advanced FinFET process (with 2.5D packaging and full-flow EDA), and upgraded planar-platform FPGA/FPSoC for smart-compute servers, autonomous cars, smart grids, and edge computing. The thesis is that as Washington restricts advanced chips, Chinese systems vendors want a home-grown programmable option even if it trails the frontier.
It is a long game. FPGA R&D is rigid and slow; the company itself and analysts note new products may take one to two years or more to pay off. Anlogic is spending through a downturn to be ready when the next up-cycle arrives.
The toolchain is the real moat
The hardest part of an FPGA business is not the silicon — it is the software that maps a customer's logic onto it. Engineers will not switch suppliers if it means rewriting years of verified design flow, IP, and timing closure work. That lock-in is why the incumbents' dominance persists decade after decade, and why Anlogic's own EDA software and its plan for full-flow tooling on advanced processes matter as much as the chip specs. A domestic FPGA that trails on raw logic capacity but offers a usable, supported toolchain can still win captive demand from Chinese systems vendors who simply cannot buy the leading parts. The moat, in other words, is as much about the compiler as the transistor.
Honest limitations
Figures come from Anlogic's 2025 annual report and subsequent exchange filings and media briefings, reported as stated; we have not audited them. The "over 85% share" for the three foreign vendors and the "US$19 billion by 2030" market size are third-party analyst estimates (MarketsandMarkets cited in Chinese media), not filings. The "several-fold" data-center growth and "millions of units" are the company's qualitative claims without a per-customer breakdown, and major-internet-firm names were not disclosed. The fundraise is a proposal subject to approval, not completed financing.
What readers can do now
- If you design hardware, evaluate a domestic FPGA for non-cutting-edge roles — the margin and mix gains suggest the mid-range parts are maturing.
- If you track China's chip independence, watch FPGA as a lagging but strategic front: small today, structurally important if export curbs persist.
- If you invest, treat Anlogic as a wager on a 2027–28 up-cycle, not a 2025 earnings story; losses and dilution risk are real near term.