By the end of 2025, Phytium had shipped more than 13 million processors, and you would struggle to find a single buyer who picked one because it was the best chip for the job. That gap, between enormous volume and nonexistent voluntary demand, is the whole truth about Phytium. The company moves Arm-based CPUs into Chinese government offices, banks, power grids, and rail networks by the million, and almost every one of those sockets was filled by a procurement mandate rather than a purchasing decision. Phytium is what it looks like when a state decides it needs a domestic Arm processor and simply wills the demand into existence.
The origin explains the customer base. Phytium, Feiteng in Chinese, traces its lineage to the National University of Defense Technology, the institution behind China’s Tianhe supercomputers, and it now sits under the state-owned China Electronics Corporation with Dou Qiang as president, headquartered in Tianjin. Its chips are built on Arm, licensed Armv8 and v8.2 architectures with Phytium’s own FTC cores layered on top. That military pedigree is exactly why the US added Phytium to the Entity List in April 2021, with reporting tying the move to the company’s role in supplying processors for military supercomputing work, including hypersonics modeling. Of everyone in this series, Phytium has the most explicitly defense-rooted history, and it wears the consequences: the deepest procurement guarantees at home, and some of the heaviest scrutiny abroad.
Four product lines, one customer
Phytium organizes its silicon into four families, and the naming is its own small lesson in Chinese branding ambition: Tengyun for servers, Tengrui for desktops, Tenglong for embedded, plus a chipset line. The server flagship that put Phytium on the map is the Tengyun S2500, a 64-core part built on FTC663 cores with an out-of-order design, a packaged system-on-chip architecture, and a 2D-mesh interconnect that supports eight-way parallelism for configurations reaching 512 cores across a single system. The newer server generation, the S5000C and its variants, moves to FTC862 and FTC870 cores, with Phytium claiming the FTC870 approaches Arm’s Neoverse N2 in performance, a claim I would treat with real caution until someone independent benchmarks it, because Phytium’s marketing has every incentive to flatter and almost no independent reviewers ever get their hands on these chips. The S5000C has been designed into China Mobile’s Lingyun platform for 5G and 6G infrastructure and has won awards for liquid-cooled deployments, with the telecom state-owned enterprise doing exactly what the government desktop does: buying domestic because it is told to.
| Chip | Cores | Architecture | Use | Notes |
|---|---|---|---|---|
| Tengyun S2500 | 64C (to 512 in 8-way) | FTC663, Armv8 | Server | 2D-mesh, OoO |
| Tengyun S5000C | server-class | FTC862 / FTC870 | Server, telecom | FTC870 claimed ~Neoverse N2 |
| Tengrui D3000 | 8C | FTC663 | Desktop | Government workstations |
| Tengrui D3000M | multi-core | FTC, low-power | Laptop | First Phytium laptop CPU (2025) |
On the client side, the milestone worth noting is the Tengrui D3000M, Phytium’s first laptop processor, launched in September 2025. It clocks above 2.9 GHz, pairs with LPDDR5x memory, draws between 18 and 25 watts, claims 8 to 10 hours of battery life, carries the Level-2 domestic security certification, and reportedly moved more than 500,000 units within months of launch. A Chinese-designed Arm laptop chip with a real mobile power envelope is a genuine step, and the half-million figure is not trivial. It is also, almost certainly, another story of mandated demand, government and state-enterprise laptop refreshes flowing to the domestic option, rather than consumers walking into a store and choosing a Phytium notebook over a Snapdragon or an Apple machine. The desktop D3000 plays the same role in office workstations.
The ecosystem play
Where Phytium has spent real effort is on the software and partner ecosystem, because mandated hardware volume is worthless if nothing runs on it. The company touts more than 6,500 ecosystem partners, hardware security through its PSPA 2.0 specification, support for China’s SM cryptographic standards, and forward-looking platform support for DDR5 and PCIe 5.0. Two integration milestones matter more than the spec sheet. Phytium chips now support the full DeepSeek model series, which plugs them into the most credible Chinese AI software story going, and the Kylin operating system, China’s flagship domestic Linux, formally adapted its V11 release to Phytium in August 2025. Those two pairings are how a mandated chip becomes a usable platform: get the dominant domestic OS and the hottest domestic AI models running natively, and the captive buyers at least get a machine that does something useful, rather than an expensive compliance checkbox. It is the same ecosystem grind every Arm vendor faces, run on a captive audience.
The foundry question nobody answers
Here is the deepest unknown in this entire series, and it belongs to Phytium. Where are these chips actually made? Before the 2021 sanctions, Phytium, like everyone, used TSMC, with older parts on roughly 16nm. After being cut off, it faces a problem sharper than the others: its military lineage makes SMIC nervous about serving it openly, since any foundry caught knowingly producing for a flagged military-adjacent designer risks losing access to the tools and chemicals it depends on. Phytium discloses almost nothing about its current manufacturing, far less than Huawei volunteers about SMIC or Loongson about its node transitions, and that silence is itself the data point. A company shipping 13 million chips has to be building them somewhere, on something, and the fact that it will not say where is the loudest thing about it. Until that question has a credible answer, every performance and volume claim Phytium makes sits on a foundation nobody can actually inspect.
The market position follows directly from all of this. Phytium’s volume goes almost entirely into government administration, finance, telecommunications, the power grid, and rail, the load-bearing infrastructure of the Chinese state, the sectors where a procurement directive can simply specify domestic Arm silicon and 13 million sockets get filled. There is no meaningful consumer presence, no export story worth the name, and no scenario in which someone outside a mandate environment weighs a Phytium chip against an alternative and chooses it on merit. The cumulative shipment figure climbed past 10 million by the end of 2024 and 13 million a year later, which is real momentum, but it is momentum measured in mandates fulfilled, not markets won.
So Phytium leaves me with the question this whole series keeps circling in its starkest form. Guaranteed demand is the best thing that ever happened to Phytium and quite possibly the worst. It bankrolls the engineering, fills the fabs Phytium will not name, and funds an ecosystem of 6,500 partners that a chip winning no open-market fights could never afford. It also means Phytium has never once had to be the right answer to a question a customer asked freely, which is the exact pressure that turns a competent state supplier into a genuinely good chip company. Loongson at least has the excuse of an unproven ISA; Zhaoxin has the discipline forced on it by trying to stay sellable. Phytium has neither, just a firehose of mandated orders and a foundry it will not discuss, and I cannot shake the sense that a company that never has to win is a company that slowly forgets how. Thirteen million chips is an impressive number right up until you realize not one of them had to earn its socket.
This is one of five company deep dives in a series on China’s fabless CPU makers. Start with the overview.