Huawei · Ascend · Layer 2
Huawei Ascend 910C
Buyers inside China, where export controls make the NVIDIA question moot.
restricted Export controls or regional restrictions govern access.
Specifications
- Memory
- Not reliably published outside China
- Bandwidth
- Not reliably published
- Compute
- Positioned against H100-class parts
- Power
- Not published
- Interconnect
- HCCS
- Software
- CANN, MindSpore
- Form factor
- accelerator
- Workload
- both
Verified 2026-09-06. Fields reading “not published” are exactly that — we do not estimate a figure a vendor withholds.
The catch
Export controls cut both ways: availability outside China is effectively nil, and independent Western benchmarks are scarce enough that every published figure should be treated as a vendor claim.
The economics
Not a price comparison. It is a jurisdiction decision.
Cost per million tokens is the metric that decides most real purchases, and there is no neutral benchmark for it — every published figure comes from a company selling one side of the comparison. Read all of them, ours included, as claims rather than measurements.
The power question underneath this
A rack of frontier accelerators draws 120–200 kW against a 2026 average of about 27 kW, and the US grid interconnection queue exceeds 2,600 GW with waits approaching five years. Whether you can energise this part is now a harder question than whether you can buy it.
Layer 1 — Energy · The interconnection queue · Rack power density · Tokens per watt
Compared against
- Huawei Ascend 910C vs NVIDIA H100 — A jurisdiction comparison, not a performance one.Compare with NVIDIA H100
Related silicon
- AMD Instinct MI300X — 192 GB HBM3, rent
- AMD Instinct MI350X — 288 GB HBM3E, rent
- AMD Instinct MI355X — 288 GB HBM3E, rent
- AWS Trainium 2 — Not published in directly comparable terms, cloud only
- AWS Trainium 3 — 144 GB HBM3e per chip, cloud only
- Google TPU v6e (Trillium) — Not published in directly comparable terms, cloud only