Subscribe
GPT-5.6 Sol · input$5/MDeepSeek V4 Pro · AA Index v4.3 · open36DeepSeek V4.1 Flash · output$0.60/MDeepSeek V4.1 Flash · input$0.15/MDeepSeek V4.1 Flash · cached input$0.003/MKimi K3 · AA Index v4.3 · open44
Hardware · Reported3 min read · 7 claims · 3 sources · every one quoted

A20 Pro is the first 2nm phone chip. The number that matters is 32 Neural Engine cores after three years at 16

Apple doubled the Neural Engine, widened memory bandwidth 50 percent, moved the DRAM off the thermal path and tripled the vapor chamber. Every one of those decisions is about running models on the phone.

In briefA20 Pro on 2nm; 50% more memory bandwidth; six-core CPU; seven-core GPU up to 40% faster; Dual 16-core Neural Engine, 32 cores, double AI processing1Packaging beside memory, off the thermal path; vapor chamber 3x surface area; up to 40% sustained gain2Siri AI uses on-device processing and Private Cloud Compute3

Apple's A20 Pro, announced Wednesday for the iPhone 18 Pro, Pro Max and the iPhone Duo, is the first smartphone system-on-chip built on a 2-nanometer process. (Apple's first 2nm part is the M6, announced in August for a Mac mini shipping this month. The A20 Pro is the first in a phone.)

The CPU and GPU claims are ordinary for an Apple September: a six-core CPU with two performance "super cores" up to 20 percent faster than A19 Pro, and a seven-core GPU, one more than last year, up to 40 percent faster. Those are the numbers Apple put on the big slide. The ones that tell you what the chip is for were smaller.

A20 Pro against A19 Pro, by Apple's numbers (% change)
Neural Engine cores100Memory bandwidth50Vapor chamber surface area200GPU (up to)40CPU super cores (up to)20

What changed for AI

The Neural Engine doubled. Apple held it at 16 cores for three consecutive generations. The A20 Pro has a "Dual 16-core Neural Engine ... 32 total cores for double the AI processing power of A19 Pro." Tom's Hardware adds that the GPU's new neural accelerators give "twice-as-fast FP8 compute," which is the precision most on-device language models actually run at.

Memory bandwidth is up 50 percent. Apple calls it "the widest memory interface in an iPhone." Why does that matter more than the CPU number? Because for a local model, bandwidth, not compute, is usually the ceiling on tokens per second. Every generated token re-reads the weights.

The packaging is Mac-style. Apple placed the die beside the memory instead of stacking it, "removing the memory from the thermal path of the chip and allowing A20 Pro to attach directly to a next-generation vapor chamber," which has three times the surface area of the 17 Pro's. The claim that follows is "up to 40 percent" better sustained performance than the 17 Pro and, per Tom's Hardware, double the 16 Pro. Sustained, not peak, is what matters for a model that runs for the length of a conversation rather than the length of a benchmark.

There is also a new in-house C2 modem replacing Qualcomm across the line. Apple says it uses 15 percent less energy than C1X and adds mmWave in the United States.

What it is for

Apple's release names the workload: Siri AI, "an entirely new version of Siri powered by Apple Intelligence," running "on-device processing and Private Cloud Compute." The on-device half of that sentence is what the doubled Neural Engine buys. Decrypt reports the heavy half runs on Google's Gemini in the cloud, which Apple's release does not mention.

Our read

We think the A20 Pro is Apple's real AI announcement this year, and that it says more than the Siri demo did. A company that doubles its NPU after three flat generations, re-plumbs its packaging around a vapor chamber and spends a slide on memory bandwidth is building a phone to run a model that does not yet exist in shipping form. Siri AI in beta, in English, outside the EU, is the placeholder. The silicon is the commitment.

The comparison to watch is not the next Snapdragon but Apple's own M-series: the A20 Pro's packaging is "inspired by M-series Apple silicon," and Tom's Hardware reports Apple will skip high-end M6 Macs for an AI-focused M7 in 2027. We'd expect Apple to quote a tokens-per-second figure for an on-device model on stage by the iPhone 19 event, the way it quotes video-playback hours now. Until it does, the 32 cores are a promise, and anyone selling a dedicated AI wearable on the premise that the phone can't do this locally should read the spec sheet twice.

Sources

7 claims · every one quoted
01
A20 Pro on 2nm; 50% more memory bandwidth; six-core CPU; seven-core GPU up to 40% faster; Dual 16-core Neural Engine, 32 cores, double AI processingBuilt using the latest 2-nanometer process technology, A20 Pro sets a new bar for mobile computing. Featuring 50 percent more memory bandwidth than A19 Pro, the new 6-core CPU features integrated Neural Accelerators and is faster than…” — apple.com · primary · Sep 9
02
Packaging beside memory, off the thermal path; vapor chamber 3x surface area; up to 40% sustained gainA20 Pro introduces custom packaging inspired by M-series Apple silicon that places the silicon die side by side with the memory, removing the memory from the thermal path of the chip and allowing A20 Pro to attach directly to a…” — apple.com · primary · Sep 9
03
Siri AI uses on-device processing and Private Cloud ComputeTo protect users’ privacy, Apple Intelligence uses on-device processing and Private Cloud Compute, which extends the privacy and security of iPhone into the cloud.” — apple.com · primary · Sep 9
Show all 7 sources
04
Neural Engine was 16 cores for three generations; first smartphone chip on TSMC N2Apple doubled it to a dual, 32-core design, after keeping the Neural Engine at 16 cores across three straight chip generations. ... The A20 Pro is Apple's first chip built on TSMC's 2-nanometer (N2) process—the first time any smartphone…” — decrypt.co · reported · Sep 9
05
Siri AI's cloud side runs on Gemini per reportingApple introduced Siri AI at its "Surprise and Shine" event on September 9, running on Google's Gemini technology under a deal reportedly worth about $1 billion a year” — decrypt.co · reported · Sep 9
06
M6 was Apple's first 2nm chip; super cores 20% faster; twice-as-fast FP8; 2x sustained over 16 Pro; C2 modem 15% less energy, mmWaveThe A20 Pro is Apple's first 2-nanometer chip in an iPhone. (Its first-ever 2 nm chip is the M6, which the company announced in August and which will debut in the Mac Mini later this month). ... new neural accelerators that the company…” — tomshardware.com · reported · Sep 9
07
Report that Apple will skip high-end M6 Macs for an AI-focused M7 in 2027Apple will skip its high-end M6 Mac chips and fast-track an AI-focused M7 generation for 2027, report claims” — tomshardware.com · reported · Sep 9
Up next · Keep readingAI · Reported · 3 min

Anthropic's alignment lead puts the odds above 10%. His colleague just quit over it

Jacob Coxon resigned saying both OpenAI and Anthropic are gambling with our lives. Evan Hubinger, who runs alignment science at Anthropic, replied that he agrees with the stakes, and that there is no plan yet.

Continue ↓