Xiaomi Didn't Catch Apple. ARM Did.

Hacker News does not usually lose its mind over a CPU score, but Xiaomi’s Xring O3 sitting next to Apple on a single-thread chart collected 917 upvotes. The score is the easy part. The harder question is whether Xiaomi drew that core, ARM drew it, or someone who used to draw Apple’s did.

Single-thread is still the status line

Chip people treat single-thread as a character test. Staple enough cores together and multicore climbs. How fast one core goes is a design problem, not a shopping problem.

Apple made that line its own. Wide decoders that swallow a lot of instructions per clock. Out-of-order pipelines that keep execution units busy even when program order is a mess. Branch predictors that bet aggressively and usually win. Everyone else was on the same ARM ISA. Apple was faster anyway. The gap lived in IPC.

That is why an Xring O3 score next to Apple produced the obvious headline first: a Chinese phone maker caught Cupertino. HN comments swallowed that framing. A few replies later the question flipped. Is that even a Xiaomi core?

A custom chip is not a custom core

Xiaomi opened its in-house SoC era in 2025 with Xring O1. Putting interconnect, an NPU, power domains, and phone software onto one die is not a press-release stunt. It is a company that used to rent its brain from Qualcomm or MediaTek climbing a real rung.

CPU microarchitecture is a different rung. Xring O3’s big core reads as ARM’s C1-Ultra, the new top-end performance core after ARM retired the old Cortex names. Licensees get the blueprint. They pick a process, size the caches, set the clocks. They do not invent the pipeline.

Apple licenses the instruction set and draws the rest in Cupertino. Xiaomi did something closer to the opposite. It took ARM’s fastest big core and sat it on Xiaomi silicon. Custom chip and custom core get mashed into one word in marketing. They are not the same word.

C1-Ultra just moved the catalog ceiling

ARM’s business is old and simple. It does not sell phones. It sells core IP. Qualcomm, MediaTek, and Samsung shop that catalog. Apple was the exception that made the model look optional: ISA license, in-house cores, everyone else on Cortex.

If C1-Ultra is near Apple on single-thread, Xiaomi is not the company that caught up. ARM is. Xiaomi is the customer that put that core under its own logo first, and loudest. The 900-plus upvote pile makes more sense in that light. People were not stunned by a Chinese handset brand. They were stunned that a licensed core could threaten the Apple P-core’s signature metric.

If that holds, the shock does not stop at Xiaomi. Anyone who can open the same catalog can, in theory, buy a similar big core. You do not need an Apple-sized CPU team. You need money, a process node, and enough integration skill not to waste what you bought.

So who actually drew it

The community got itchy one sentence after the score. What changed inside ARM’s big-core team that an Apple-only gap shrank this far.

There is no confirmed designer roster. The rumor has been circulating for years anyway. People who used to lay out Apple Silicon pipelines moved to ARM and its licensees, and pulled issue width and cache policy closer to Cupertino. The cruder version says someone read patents and copied a philosophy. Neither version is proven.

The question still stands. If the score looks like Apple, does the design notebook look like Apple too. A benchmark line gets consumed by Monday. Who set decoder width and who retuned the branch predictor is what remains. Call Xring O3 a Xiaomi win and that layer gets erased.

A tied score is not a tied product

Apple’s moat was exclusivity. We are the only ones who draw this core. A lot of why an iPhone felt a class faster than an Android flagship lived there. If that exclusivity wobbles in the license market, Apple’s one-line brag gets thinner.

What remains is not one core. It is whether the memory subsystem keeps the pipeline fed. How clocks fold when power is plentiful and when it is not. How the compiler lays instructions onto that pipeline. Years of software work do not copy with a bench run. Equal single-thread numbers do not mean the fan kicks in at the same moment or the battery falls on the same curve.

The claim that Xiaomi caught Apple is usually only half a claim. Matching one axis is not matching the silicon as a product. Xring O3 still has to prove the next part. Not the lab number. Whether that number shows up in a phone in your hand, and what that speed costs in watts.

The next iPhone and the next Xiaomi flagship can meet on the same benchmark line. Sharing a line does not put them in the same chip business. Whether you read that line as Xiaomi’s design talent or as a new ceiling in ARM’s catalog is the part that actually lingers.