BEIJING, 26 AUG 2026 — Xiaomi has unveiled the XRING O3, a mobile processor with 24 billion transistors, a 10-core CPU running to 4.35GHz and a 16-core GPU. It is reported as the first smartphone chip to support LPDDR6 memory, and it debuts in the Xiaomi 18 Fold in September.
It is also being described everywhere as homegrown, which is true of the design and not of the manufacture. The O3 is fabricated by TSMC on its N3P process, in Taiwan.
Homegrown means designed, not made
The distinction is not pedantry in this context; it is the whole strategic question.
Designing a competitive mobile processor is difficult and expensive, and Xiaomi has now done it repeatedly. The capability reduces its dependence on Qualcomm and MediaTek for the part of the device where margin concentrates.
Fabricating one at 3nm is a different problem entirely, and nobody in mainland China can currently do it at this node. The O3 depends on a Taiwanese foundry, which depends on lithography equipment subject to export controls, which is precisely the dependency the phrase homegrown is usually invoked to suggest has been overcome.
Xiaomi has shed a commercial dependency rather than a geopolitical one. The company could survive losing access to Qualcomm; it could not survive losing access to TSMC.
That is not a hypothetical distinction in the current climate. The same week this chip was announced, Taiwanese prosecutors indicted nine people over restricted AI servers diverted to China. The supply line that produces Xiaomi's independence runs through the jurisdiction actively policing what reaches the mainland.
The memory is the interesting part, not the clock
The 4.35GHz clock and the 16-core graphics unit get the headlines. The LPDDR6 support is what will matter in practice.
Running a model on a phone is a memory bandwidth problem rather than an arithmetic one. The processor spends most of its time waiting for model weights to be read out of memory. Adding cores to a workload that is already waiting on memory buys very little.
The reported figures are up to 10,667Mbps and 113.8GB/s, an increase of about 48 per cent over the XRING O1. For on-device inference that is the number that moves the experience, and being first to a memory standard is a more durable advantage than a clock speed that competitors match within a product cycle.
The timing is awkward. Xiaomi's own results show handset margins under pressure from memory prices, and adopting the newest and most expensive memory standard first is a decision to spend margin on capability. That is a defensible bet on differentiation, and it is a bet.
Ten cores and no small ones
The CPU is an all-big-core design, a departure from the mixed-core arrangement used in mobile processors for more than a decade.
The standard arrangement pairs a few large, fast cores with several small, efficient ones, and moves work between them: background tasks and idle housekeeping run on the small cores at a fraction of the power, while the large cores wake for anything demanding. It is the main reason modern phones last a day.
Dropping the small cores trades that away for consistency. Every thread gets a fast core, so there is no scheduling decision to get wrong and no penalty when work lands on the weaker half — a real problem in practice, since a misplaced foreground task on a small core is felt immediately as stutter.
The trade-off hinges on how well those large cores idle. If a big core can drop to a very low power state when lightly loaded, the small ones are redundant. If it cannot, this design choice will show up as a battery life complaint. That is a question for reviewers with a shipping device, and it is the thing worth reading about in September rather than the benchmark score.
Why almost nobody else can do this
Few handset makers design their own silicon, not because the engineering is unattainable but because the fixed cost only makes sense spread across enormous volume.
Bringing a processor to production at a leading node runs into the hundreds of millions of dollars before a single chip is sold, once design tools, licensed processor and graphics intellectual property, mask sets, verification and the engineering organisation are counted. That cost is the same whether the part ships in one million phones or one hundred million.
Xiaomi ships at a scale where the arithmetic works, and that is the actual barrier separating it from most of its competitors. It is also why the strategic question is volume rather than capability. A first-party chip confined to a halo folding phone does not recover its own development cost; the same chip across a mainstream range does.
From O1 to O3, with an O2 that did not appear
The naming deserves a note, because the comparison figures depend on it.
An XRING O2 was widely reported as planned for 2026 on a 3nm process. What has been announced is an O3, and the performance comparisons being quoted are against the O1 — the part that actually shipped.
Whether the O2 was skipped, folded into this design or simply renamed has not been explained publicly, and it does not much matter except for one thing: a 48 per cent gain measured against the O1 spans whatever happened in between. Read as a generation-on-generation improvement it is flattering; read as the distance from the last chip Xiaomi actually shipped, it is accurate.
Vendor benchmarks on a device nobody has
The circulating scores are an AnTuTu result above 5.2 million and a Geekbench multi-core figure above 15,000, both described as class-leading.
Take these with the usual caution. Vendor benchmarks run on engineering hardware, often outside the thermal limits of a finished product, and this chip is debuting in a folding phone — about the most thermally constrained design there is. A folding device has less internal volume, thinner sections and two panels of heat-generating display.
Sustained performance in a fold is typically well below peak, and peak is what a benchmark run captures. None of this suggests the numbers are fabricated. It suggests they describe a ceiling that the shipping device will touch briefly.
What it means from here
For buyers in the region the practical consequence arrives next year rather than next month. A first-party chip in a flagship fold is a halo product; the question is whether XRING reaches the volume tiers where Xiaomi actually competes across ASEAN.
If it does, the effect is on price rather than on specification. A manufacturer that designs its own processor keeps the margin it was paying to a chip vendor, and in markets where handsets compete hard on price that margin tends to reach the shelf.
The strategic takeaway is simpler. Chinese manufacturers are closing the chip design gap quickly, while the advanced manufacturing gap at this node is not closing at all. Every announcement in this category is evidence of the first and, read carefully, evidence of the second.
View author profile → · Editorial policy
