MICHAEL ZHANG/gettyimages.com
Xiaomi on Monday unveiled three in-house chips at its Xring Chip Technology Communication Conference in Beijing — and the most consequential story isn’t the record benchmark score, it’s the architectural philosophy behind the chip that set it.
The Xring O3 posted an AnTuTu v11 score of 5,228,014, becoming the first mobile SoC past five million points — but the number is a by-product of a design decision that separates the O3 from its predecessor in a more meaningful way: Xiaomi abandoned the mixed performance-and-efficiency cluster design that every other Android chip maker uses and built a ten-core processor made entirely of high-performance cores. That choice, combined with the chip’s adoption of LPDDR6 — the newest mobile memory standard, finalized by JEDEC in July 2025 — makes the O3 the first generation of Xring silicon to argue for architectural leadership rather than benchmark parity.
What All-Big-Core Actually Means
Conventional mobile chips divide their CPU cores into three tiers: one or two ultra-high-performance prime cores for peak tasks, a cluster of mid-tier performance cores for everyday workloads, and a cluster of small efficiency cores that handle idle processes and light tasks to preserve battery life. Qualcomm’s Snapdragon 8 Elite, MediaTek’s Dimensity 9400, and Xiaomi’s own Xring O1 all follow this blueprint in varying forms.
The Xring O3 discards the efficiency tier entirely. Its all-big-core CPU design details break into three groups — two C1-Ultra cores running at up to 4.35GHz, four C1-Premium cores at 3.68GHz, and four C1-Pro cores at 3.15GHz — and every core is, by mobile standards, a big core. Xiaomi reports a 60% improvement in multi-core performance over its baseline, with a Geekbench 15221 multi-core result in Geekbench v6.5 — a 61% jump over the Xring O1 and the first time any Xiaomi chip has cleared that threshold.
The trade-off is real and worth naming: all-big-core designs optimize for peak and sustained workload performance at the potential cost of battery efficiency under light and idle loads. Whether the O3 manages that trade-off better than Qualcomm’s approach is a question that only retail hardware testing under sustained real-world conditions will answer. The chip was built on TSMC N3P node and transistors numbering 24 billion into a 133mm² (approximately 0.21 in²) die.
LPDDR6: The Memory Standard That Changes the Equation
The O3 is Xiaomi’s first mobile chip with LPDDR6 support, and multiple independent industry sources confirm it is the LPDDR6 first mobile SoC to implement the new standard in a shipping product. JEDEC, the global semiconductor standards body, published the LPDDR6 specification (JESD209-6) in July 2025; Samsung and SK Hynix have been ramping production since early 2026, with SK Hynix 1c LPDDR6 validated at 14.4 Gbps in March 2026.
What LPDDR6 changes structurally is channel architecture. Where LPDDR5 used two 32-bit sub-channels, LPDDR6 uses LPDDR6 four 24-bit sub-channels — a design that delivers higher concurrency and lower latency rather than just raw speed. The O3 runs LPDDR6 at 10,667 Mbps, the standard’s base data rate, delivering LPDDR6 at 113.8 GB/s bandwidth — a 48% uplift over the Xring O1’s LPDDR5X implementation.
The JEDEC LPDDR6 press release describes LPDDR6 as “an ideal solution” for edge AI inference, and the O3’s memory design reflects that positioning: with more bandwidth per watt and higher concurrency than its predecessor, the chip is built to sustain the simultaneous data movement that on-device AI models require.
Distributed AI Acceleration: Why 200 TOPS Is Different This Time
The Xring O3’s NPU delivers 200 TOPS AI performance of tensor computing power and 3.13 TFLOPS of vector performance — a 45% improvement over the Xring O1. What distinguishes the O3’s approach from the dedicated-NPU model used by most competitors is distribution: rather than routing all AI workloads through a single neural processing block, Xiaomi spreads acceleration across the CPU, GPU, ISP, DPU, and ADSP. The company reports A19 Pro static memory latency of 82 nanoseconds, which it claims improves on Apple’s A19 Pro in that specific measurement — a claim that, like the AnTuTu headline, requires independent validation on retail hardware before it can be taken as settled.
The 16-core G2-Ultra NX GPU is Xiaomi’s own design rather than a standard ARM Mali reference, with claimed G2-Ultra NX GPU 85 percent improvement in graphics performance and 64% in power efficiency over the prior generation.
Debut Devices: Xiaomi 18 Fold and Pad 9 Pro Max
The O3 will make its commercial debut in two products launching in China in September 2026: the Xiaomi 18 Fold foldable smartphone and the Xiaomi Pad 9 Pro Max tablet.
The Xiaomi 18 Fold enters a segment where Huawei holds a dominant position — one market research firm placed Huawei’s share of China’s foldable market at approximately 68% in the second quarter of 2026, with 1.6 million units shipped over that period. Reuters 18 Fold shipment targets, citing people familiar with Xiaomi’s plans, reported shipment targets of 200,000 to 300,000 units for the device — a fraction of Huawei’s volume, but a meaningful signal of intent from a company whose first Xring-powered devices shipped more than one million units combined.
The Pad 9 Pro Max places the O3 into a large-screen context where its memory bandwidth and AI acceleration could differentiate it in productivity and creative workloads.
Xring O100: On-Device AI at Terabit-Scale Bandwidth
The most architecturally significant chip unveiled on Monday may not be the one with the headline benchmark score. The Xring O100, built on a 6nm process, is purpose-built for on-device AI inference, and its defining feature is a near-memory design that physically collapses the distance between compute and data.
The core problem near-memory architectures address is the memory wall: in conventional processor designs, data must travel repeatedly between the chip’s compute units and off-chip memory, and that movement is expensive in both energy and time. Research published in IEEE Computer (May 2026) by Google DeepMind scientists David Patterson and Xiaoyu Ma identified data movement costs — not arithmetic throughput — as the primary bottleneck in LLM inference. Fudan University research has quantified the problem: data movement energy is approximately two orders of magnitude higher than computation energy in conventional architectures.
The Xring O100 addresses this by stacking compute and memory vertically, with compute connected to memory via millions of high-speed vertical channels. Xiaomi says this yields claimed memory bandwidth of 1.22 terabits per second — an order of magnitude beyond the O3’s 113.8 GB/s. The architecture is designed to run large language models on-device without a cloud connection, supporting Xiaomi’s in-house MiMo LLM across smartphones, vehicles, and robotics. Qualcomm HBC near-memory computing announced its own High-Bandwidth Compute near-memory architecture for datacenter AI at its June 2026 investor day, validating the design direction independently.
The O100 has completed development and is targeted for deployment in 2027. It is not scheduled to appear in any September launch device.
Xring D100: Intelligent Driving at 3nm
The Xring D100 extends Xiaomi’s silicon portfolio into automotive territory. Built on the same 3nm node as the O3, it is described as China’s first homegrown 3nm AI processor designed for intelligent driving applications.
Its specifications include a 20-core CPU, a 16-core NPU, and support for up to 160GB of unified memory. Xiaomi claims the chip can run AI models with up to 200 billion parameters locally — a figure that, if validated in production, would represent frontier-scale reasoning capability deployed inside a vehicle without cloud connectivity. The D100 has completed validation and is planned for commercial deployment in 2027, aligned with Xiaomi’s electric vehicle expansion.
Competitive Reality: A Record With an Expiration Date
One number that does not appear in Xiaomi’s announcement is the process node of its primary competitors’ next chips: 2nm. Apple Qualcomm MediaTek 2nm chips are all preparing flagship processors on TSMC’s 2nm process for release in late 2026, while the Xring O3 ships on 3nm.
That gap matters more than benchmark scores suggest. A chip built on a more advanced node can achieve equal or superior performance at lower power, or better power at equal performance — and that combination typically produces a chip that wins in the real-world test that benchmarks are proxies for: sustained performance under thermal constraints in a thin device chassis. The Xring O3’s five-million AnTuTu score was recorded in Xiaomi’s own lab under unspecified conditions; independent validation still needed on retail hardware running sustained workloads in a production foldable will tell a different story, as it always does.
Xiaomi’s strategic calculus is also clear: the Xring O3 is launching primarily in China, in a premium foldable priced for a market where the 2nm competition isn’t the primary comparison set. Global smartphone shipments are projected to fall IDC global smartphone decline roughly 13% in 2026, according to IDC, compressing the market Xiaomi needs to justify its chip investments. The question isn’t whether the O3 beats a Snapdragon 8 Elite Gen 6 in a benchmark — it’s whether Xiaomi can build enough design iteration experience to be genuinely competitive on a global scale by the time the Xring O5 or O6 arrives.
The O1 shipped in three devices and cleared one million units. The O3 is targeting 200,000 to 300,000 units in a premium foldable. The cadence is real; the scale is still being built.
What Chinese Law Means for Xiaomi Chip Buyers
Xiaomi is headquartered in Beijing and subject to China’s national security legal framework. China National Intelligence Law (2017), Article 7, requires that all organizations and citizens must support, assist, and cooperate with state intelligence work. The China Cybersecurity Law (2017), Article 28, additionally requires network operators to provide technical support and assistance to public security organs. The Data Security Law (2021) imposes data handling obligations on companies operating in China.
These obligations apply to Xiaomi as a Chinese company regardless of where its servers are physically located and regardless of its stated privacy policies. No backdoor has been confirmed in any Xring chip by a named independent security auditor, and no specific government data-handover incident involving Xring-powered devices has been publicly documented. The legal obligation is structural: it exists whether or not it has been exercised, and it is a fixed condition for any enterprise or government buyer evaluating these chips for supply-chain risk.
Xiaomi’s own Xiaomi transparency report states that it does not provide direct government access to customer data or maintain product backdoors. The structural legal obligation and the company’s stated policy are not contradictory — they describe different conditions. Buyers should understand both.
Frequently Asked Questions
What makes the Xring O3’s all-big-core design different from other Android chips?
Most mobile processors mix high-performance cores with small efficiency cores to balance peak power against battery life at idle. The Xring O3 eliminates the small efficiency tier and uses ten high-performance cores across three sub-tiers, optimizing the chip for sustained workloads at the expense of that efficiency fallback. Whether the O3’s thermal and power management can compensate at idle is a question that retail testing will answer; benchmark scores from the manufacturer’s lab cannot.
Why does LPDDR6 matter for AI performance on a smartphone?
LPDDR6 changes the memory architecture from two 32-bit sub-channels (LPDDR5) to four 24-bit sub-channels, increasing concurrency and reducing latency rather than just adding raw bandwidth. For on-device AI inference — where the bottleneck is moving large model weights repeatedly between chip and memory, not executing arithmetic — that concurrency improvement translates to more sustained throughput. The Xring O3 runs LPDDR6 at 10,667 Mbps, delivering 113.8 GB/s of bandwidth: 48% more than the Xring O1.
What is the Xring O100, and how does it differ from the O3?
The Xring O100 is a dedicated AI inference chip, not a smartphone SoC. It uses a near-memory design that stacks compute and memory vertically, eliminating the energy cost of moving data between physically separate chip and memory — the principal bottleneck in running large language models on-device. Xiaomi claims 1.22 Tbps of memory bandwidth, roughly ten times the O3’s figure. The O100 is targeted at smartphones, vehicles, and robotics running Xiaomi’s own MiMo language model, and is scheduled for 2027 deployment, not the September devices.
Does China’s National Intelligence Law apply to Xring chips or Xiaomi products?
As a Chinese company, Xiaomi is subject to China’s National Intelligence Law (2017), Article 7, which requires all organizations to support, assist, and cooperate with state intelligence work; and to the Cybersecurity Law (2017), which requires network operators to provide technical support to security organs. No backdoor or government data-handover incident has been documented for any Xring-powered device. The legal obligation is structural and applies regardless of where Xiaomi’s servers are located or what its privacy policies state — a condition that enterprise and government buyers should weigh alongside the chips’ performance claims.
