← Back to stack

Chip pricing

What Chinese chips cost

The pricing layer of the tracker: 63 chips — 51 with credible public prices — 150 dated, source-linked observations, all shown in USD with the as-reported figure underneath. The question this page exists to answer: how far apart are Chinese chips and their global rivals, in dollars?

Dataset updated 2026-08-27 · every observation source-linked

131 TFLOPS/$1kcheapest compute on the tracker — AMD Instinct MI300X at ~$10K
29 TFLOPS/$1kbest Chinese-vendor value — Ascend 910C at ~$27.6K
~$9.7K vs ~$16.5Kmedian accelerator price, China-market (12) vs global (8)
+80%Nvidia A100 80GB SXM, 2023-02-232026-08 (~$10K~$18K)
-61%AMD EPYC 9654 (Genoa), 2022-112026-08 (~$11.8K~$4.6K)
11price observations on Nvidia H20 — launch to halt

The gap

China vs the global frontier

The spread of FP16 compute per dollar on each side, per generation — every qualifying chip, not just the leader. Gap compares each side’s best. Sides split by vendor nationality — China-market variants of US chips (H20, L20) sit on neither side.

FP16 compute per dollar: Chinese vendors sit about 1–6× behind the global frontier, depending on which chips you pair.

Why Nvidia still dominates. Nothing here says AMD builds the best AI chip — the ranking is peak-FP16-per-dollar and nothing else. AMD tops it by quoting large FP16 figures and pricing aggressively (our own MI325X note records “weak customer interest and room for price cuts”); Nvidia wins where it matters — CUDA and the software stack, real-world utilization of peak, NVLink networking, and whole-system delivery. There is also a price-observation bias: Nvidia publishes no list prices, so its figures skew toward scarcity street prices ($25–40K H100), while AMD's best observations skew toward negotiated volume deals ($10K MI300X via Microsoft/Citi). Deployments tell the story: Nvidia holds the overwhelming share of AI compute; AMD's share is real but small.

Why domestic chips still sell. The gap compares quoted prices in markets that export controls keep apart: the MI300X's ~$10K is a US-market contract price a Chinese buyer cannot legally pay. When US silicon does reach China, scarcity repricing closes most of the gap — the H800's post-ban China street price ran $35–70K per card, putting it in the same ~15–25 TFLOPS/$1k band as the best domestic parts. And buyers don't optimize peak FP16 per dollar: the H20 sold over a million units in 2024 at 11–15 TFLOPS/$1k — worse value than any chip above — because inference is memory-bound and supply security, after two H20 bans in 18 months, is worth a premium. State procurement (China Mobile's >RMB 1B Kunlunxin orders) does the rest.

GenerationChinese vendors — rangeGlobal — rangeGap (range)
2025–202629.0 TFLOPS/$1kAscend 910C 29.0100.0 TFLOPS/$1kAMD Instinct MI355X 100.0*33×*
202424.8–26.3 TFLOPS/$1kMLU590 (思元590) 26.3Kunlunxin P800 24.875.0–87.2 TFLOPS/$1kAMD Instinct MI325X 87.2Nvidia B200 75.034×
2022–20236.1–100.0 TFLOPS/$1kAtlas 300I Duo 100.0*MLU370-X8 46.4*Hygon DCU DeepComputing II (深算二号) 24.6Ascend 910B 23.0MTT S4000 6.1*31.9–130.7 TFLOPS/$1kAMD Instinct MI300X 130.7Nvidia H100 SXM 39.6Nvidia H200 SXM 31.916×
2021 and earlier13.0 TFLOPS/$1kIluvatar CoreX TianGai 100 / BI series 13.0*17.3–47.9 TFLOPS/$1kAMD Instinct MI250X 47.9Nvidia A100 80GB SXM 17.314×*

Ratios compare each side's best value (most FP16 TFLOPS per $1,000) within a generation, priced at each chip's latest MARKET observation (street, contract, or list) — analyst models are used only where no market price exists, and marked “estimate” with a confidence dot. A ratio of 3× means the best global chip delivers three times more peak FP16 per dollar; real-world gaps are smaller where software, memory, and interconnect favor the global part — and larger where export controls inflate China-market pricing. Channel price spreads (e.g. Ascend 910C's ¥80k–200k quotes) can move a ratio substantially. Chips and gaps marked * are priced by estimate where no market observation exists — true of the entire 2025–2026 global side: neither MI355X nor B300 has a market per-card price yet, only analyst models.

The generations

Every chip, generation by generation

One dot per market-priced accelerator; full tables on the category pages.

$2k$5k$10k$20k2025–20262 pricedAscend 910C — ~$27.6K (per card) as of 2026, CNY-reportedAscend 950PR — ~$9.7K (per card) as of 2026-04, CNY-reported20245 pricedAMD Instinct MI325X — ~$15K (per card) as of 2025-03Kunlunxin P800 — ~$13.9K (per card) as of 2025-07, CNY-reportedMLU590 (思元590) — ~$9.7K (per chip) as of 2025-08, CNY-reportedNvidia B200 — ~$30K (per B200 SXM module) as of 2025-07-14Nvidia H20 — ~$7.2K (per complete 8-card machine-month (H20-141G, Jinan carrier equipment room, 1-month deposit + 3 months prepaid)) as of 2026-08-27, CNY-reported2022–20238 pricedAMD Instinct MI300X — ~$10K (per card) as of 2024-01Ascend 910B — ~$13.9K (per card) as of 2024-05-24, CNY-reportedHygon DCU DeepComputing II (深算二号) — ~$7.8K (per card) as of 2025, CNY-reportedMetaX C500 (曦云C500) — ~$6.6K (per card (ASP)) as of 2025-06-30, CNY-reportedNvidia A800 80GB — ~$10K (per card) as of 2023Nvidia H100 SXM — ~$25K (per card) as of 2026-08Nvidia H200 SXM — ~$31K (per card) as of 2026-08Nvidia H800 — ~$35K (per card) as of 2023-122021 and earlier3 pricedAMD Instinct MI250X — ~$8K (per card) as of 2026-08Enflame Suiyuan/CloudBlazer DTU (current: S60, L600) — ~$1.9K (per accelerator card/module (ASP)) as of 2024-12-31, CNY-reportedNvidia A100 80GB SXM — ~$18K (per card (80GB SXM)) as of 2026-08

Every market-priced AI-server accelerator, one dot per chip: columns are generations (newest first), height is the latest market price in USD at observation-year FX (log scale), red = Chinese vendor, gray = global. Hover any dot for its as-reported price; click through to the chip page. 15 tracked accelerators have no market-priced observation and are omitted. Full tables — with every observation, source, and unpriced chip — live on the category pages.

Full tables: AI accelerator (36) · Server CPU (15) · Memory (7) · Edge AI (5) · Cloud rental rates

The value map

What a dollar of compute buys

Ranked by peak FP16 per dollar — one axis, not a verdict: AMD tops paper FLOPs-per-dollar; Nvidia wins deployments on software, utilization, and systems. Leaderboard first, then every priced accelerator on one map.

Compute per dollar 15 priced accelerators

  1. 1AMD Instinct MI300X131 TFLOPS/$1k2022–2023 · $10K · 2024-01
  2. 2AMD Instinct MI325X87 TFLOPS/$1k2024 · $15K · 2025-03
  3. 3Nvidia B20075 TFLOPS/$1k2024 · $30K · 2025-07-14
  4. 4AMD Instinct MI250X48 TFLOPS/$1k2021 and earlier · $8K · 2026-08
  5. 5Nvidia H100 SXM40 TFLOPS/$1k2022–2023 · $25K · 2026-08
  6. 6Nvidia H200 SXM32 TFLOPS/$1k2022–2023 · $31K · 2026-08
  7. 7Nvidia A800 80GB31 TFLOPS/$1k2022–2023 · $10K · 2023
  8. 8Ascend 910C29 TFLOPS/$1k2025–2026 · $27.6K · from ¥200,000 at 2026 avg 7.25 CNY/USD · 2026
  9. 9Nvidia H80028 TFLOPS/$1k2022–2023 · $35K · 2023-12
  10. 10MLU590 (思元590)26 TFLOPS/$1k2024 · $9.7K · from ¥70,000 at 2025 avg 7.19 CNY/USD · 2025-08
  11. 11Kunlunxin P80025 TFLOPS/$1k2024 · $13.9K · from ¥100,000 at 2025 avg 7.19 CNY/USD · 2025-07
  12. 12Hygon DCU DeepComputing II (深算二号)25 TFLOPS/$1k2022–2023 · $7.8K · from ¥56,000 at 2025 avg 7.19 CNY/USD · 2025
  13. 13Ascend 910B23 TFLOPS/$1k2022–2023 · $13.9K · from ¥100,000 at 2024 avg 7.20 CNY/USD · 2024-05-24
  14. 14Nvidia H2021 TFLOPS/$1k2024 · $7.2K · from ¥52,000 at 2026 avg 7.25 CNY/USD · 2026-08-27
  15. 15Nvidia A100 80GB SXM17 TFLOPS/$1k2021 and earlier · $18K · 2026-08

Longer bar = more peak FP16 compute per dollar. This measures one narrow thing — quoted dense-FP16 throughput divided by the latest per-unit price (USD at observation-year average FX) — and deliberately ignores memory bandwidth, interconnect, software ecosystem, and real-world utilization, which is why AMD parts top raw FLOPs-per-dollar while Nvidia dominates actual deployments. Bars mix generations by default: an older chip at a discounted 2026 street price can outrank a new flagship at launch price — check each bar's generation tag, or filter to one generation for a like-for-like view. Datacenter class only; consumer gaming cards never rank here. Bar color: red = China-vendor, gray = global.

Metric:
1002005001k2k$2k$5k$10k$20kAI compute (TFLOPS) → strongerprice ↑ higher↖ expensive & weakercheapest per unit of compute ↘AMD Instinct MI250X — ~$8K as of 2026-08 (per card), 383 fp16/bf16 dense tflopsAMD Instinct MI300X — ~$10K as of 2026-08 (per card), 1,307.4 fp16/bf16 dense tflopsAMD Instinct MI325X — ~$10K as of 2026-08 (per card), 1,307.4 fp16/bf16 dense tflopsAMD Instinct MI355X — ~$25K as of 2026-08 (per card), 2,500 fp16/bf16 dense tflopsAscend 910B — ~$13.9K (from ¥100,000) as of 2024-05-24 (per card), 320 fp16/bf16 dense tflopsAscend 910C — ~$27.6K (from ¥200,000) as of 2026 (per card), 800 fp16/bf16 dense tflopsAtlas 300I Duo — ~$1.4K as of 2025-10-20 (per card), 140 fp16/bf16 dense tflopsHygon DCU DeepComputing II (深算二号) — ~$7.8K (from ¥56,000) as of 2025 (per card), 192 fp16/bf16 dense tflopsIluvatar CoreX TianGai 100 / BI series — ~$11.3K (from ¥79,999) as of 2023 (per card), 147 fp16/bf16 dense tflopsKunlunxin P800 — ~$13.2K (from ¥96,000) as of 2026-08-27 (per card), 345 fp16/bf16 dense tflopsMLU370-X8 — ~$2.1K (from ¥15,000) as of 2026-08 (per card), 96 fp16/bf16 dense tflopsMLU590 (思元590) — ~$9K as of 2026-08 (per chip), 256 fp16/bf16 dense tflopsMTT S4000 — ~$10.9K (from ¥78,851) as of 2026-08-25 (per card), 100 fp16/bf16 dense tflopsNvidia A100 80GB SXM — ~$18K as of 2026-08 (per card (80GB SXM)), 312 fp16/bf16 dense tflopsNvidia A800 80GB — ~$10K as of 2023 (per card), 312 fp16/bf16 dense tflopsNvidia B200 — ~$25K as of 2026-08 (per GPU), 2,250 fp16/bf16 dense tflopsNvidia H100 SXM — ~$25K as of 2026-08 (per card), 989.5 fp16/bf16 dense tflopsNvidia H20 — ~$7.2K (from ¥52,000) as of 2026-08-27 (per complete 8-card machine-month (H20-141G, Jinan carrier equipment room, 1-month deposit + 3 months prepaid)), 148 fp16/bf16 dense tflopsNvidia H200 SXM — ~$31K as of 2026-08 (per card), 989.5 fp16/bf16 dense tflopsNvidia H800 — ~$35K as of 2023-12 (per card), 989.5 fp16/bf16 dense tflopsNvidia L20 — ~$3.4K as of 2026-08 (per card), 119.5 fp16/bf16 dense tflops

Each dot is one chip: farther right = more compute, higher = more expensive (both scales stretch logarithmically so a cheap edge chip and a $30,000 flagship fit on one picture — equal steps multiply, they don't add). Bottom-right dots deliver the most compute per dollar; see the ranked bars above for the same story as a plain leaderboard. Point size = source confidence; red = China-market, gray = global. All prices in USD at the observation year's average exchange rate (the native figure shows on hover).

Compute per rental dollar 12 rates

  1. 1Ascend 910B1129 TFLOPS/$1hr$0.28/card-hr (derived) · per 8-card server-month (SMM Yangtze River Delta 91…
  2. 2Ascend 910B985 TFLOPS/$1hr$0.32/card-hr (derived) · per 8-card server-month (SMM West China 910B-64G av…
  3. 3Nvidia B200594 TFLOPS/$1hr$4/card-hr · per GPU-hour (cloud on-demand, 2025-07)
  4. 4AMD Instinct MI325X581 TFLOPS/$1hr$2/card-hr · per GPU-hour (cloud on-demand)
  5. 5Ascend 910C546 TFLOPS/$1hr$1/card-hr (derived) · per 8-card server-month (SMM industry-average lease…
  6. 6MTT S4000390 TFLOPS/$1hr$0.26/card-hr · per card-hour (AutoDL market, MTT S4000 48GB, pay-a…
  7. 7AMD Instinct MI355X291 TFLOPS/$1hr$9/card-hr · per GPU-hour (cloud on-demand)
  8. 8Ascend 910B178 TFLOPS/$1hr$2/card-hr · per card-hour (Huawei Cloud Ascend 910B 64GB, on-de…
  9. 9Nvidia H20157 TFLOPS/$1hr$0.94/card-hr (derived) · per 8-card server-month (apetops.com public listing…
  10. 10Nvidia H20138 TFLOPS/$1hr$1/card-hr (derived) · per 8-card server-month (SMM H20-141G Chengdu-Chong…
  11. 11Nvidia H20121 TFLOPS/$1hr$1/card-hr (derived) · per complete 8-card machine-month (H20-141G, Jinan …
  12. 12Ascend 910B60.3 TFLOPS/$1hr$5/card-hr · per card-hour (China Telecom Tianyi Cloud pak3.4xla…

Longer bar = more FP16 compute per $1/hour of rental — the rental twin of the leaderboard above. Monthly 8-card machine leases are normalized to per-card-hour (÷8 cards ÷730 hours) and flagged “derived”; wholesale leases may bundle hosting and power, so treat them as a floor on rental value. CPUs are excluded (no FP16 figure). The rental market’s striking feature: wholesale 910B leasing delivers the most compute per rental dollar anywhere — while the same chip at official retail rates delivers the least.

Trends

Where prices are moving

Chips with three or more dated per-unit observations, ordered by total change — the memory supercycle, export-control discounts, and generational price cuts all show up here. Indexed to each chip's first observation; every figure lives on the chip's page with sources.

3675113152191price index (first observation = 100)Enflame Suiyuan/CloudBlazer DTU (current: S60, L600) — ¥8,777 · 2023-12-31 · index 100Enflame Suiyuan/CloudBlazer DTU (current: S60, L600) — ¥14,000 · 2024-12-31 · index 157Nvidia A100 80GB SXM — $10,000 · 2023-02-23 · index 100Nvidia A100 80GB SXM — $10,000 · 2023-12 · index 100Nvidia A100 80GB SXM — $18,000 · 2026-08 · index 180Nvidia A800 80GB — ¥100,000 · 2022-11 · index 100Nvidia A800 80GB — $10,000 · 2023 · index 67Nvidia H100 SXM — $39,995 · 2023-04-14 · index 100Nvidia H100 SXM — $30,000 · 2024-01 · index 75Nvidia H100 SXM — $25,000 · 2024-03 · index 63Nvidia H100 SXM — $25,000 · 2025-10 · index 63Nvidia H100 SXM — $25,000 · 2026-08 · index 63Nvidia H20 — $12,000 · 2024-02-01 · index 100Nvidia H20 — ¥110,000 · 2024-03 · index 127Nvidia H20 — $12,000 · 2024-07 · index 100Nvidia H20 — $10,000 · 2025-10-25 · index 83Nvidia H20 — ¥100,000 · 2025-11 · index 116Nvidia H20 — ¥52,000 · 2026-08-27 · index 60AMD EPYC 9654 (Genoa) — $11,805 · 2022-11 · index 100AMD EPYC 9654 (Genoa) — $4,599 · 2026-08 · index 39AMD EPYC 9755 (Turin) — $12,984 · 2024-10 · index 100AMD EPYC 9755 (Turin) — $7,995 · 2026-08 · index 62Intel Xeon Platinum 8480+ — $10,710 · 2023-01 · index 100Intel Xeon Platinum 8480+ — $7,999 · 2026-08 · index 75Loongson 3A6000 — ¥1,499 · 2024-01 · index 100Loongson 3A6000 — ¥1,900 · 2024-01-23 · index 127Loongson 3A6000 — $372.91 · 2024-07 · index 179Nvidia Jetson AGX Orin 64GB module — $1,999 · 2022-03-22 · index 100Nvidia Jetson AGX Orin 64GB module — $3,499 · 2026-07-22 · index 175Nvidia A100 80GB SXMLoongson 3A6000Nvidia Jetson AGX Ori…Enflame Suiyuan/Cloud…Intel Xeon Platinum 8…Nvidia A800 80GBNvidia H100 SXMAMD EPYC 9755 (Turin)Nvidia H20AMD EPYC 9654 (Genoa)

Each line is one chip's per-unit price observations over time, converted to USD at each observation year's average FX and indexed to its first observation (= 100). Hover a line or legend chip to highlight it; click legend chips to add or remove chips; the dashed line marks unchanged price. Analyst estimates and reseller listings are excluded — every point is a dated, sourced market observation (hover any dot). Chip pages carry the full history.

Methodology

How prices are sourced and graded

Every spec and price carries a source link and a date. Datacenter chips rarely have official list prices, so figures are what the market reported — launch MSRP, street, contract — and every observation is kept, because the trajectory is the story.

All prices display in USDat the observation year's average exchange rate (RMB per USD; HKD at 7.8); the original figure and rate print under each price. Confidence: green dot — vendor-official or multiple independent reports; amber — single major report; gray — estimate or lone secondary source. Raw research batches are archived under docs/chip-research/.

Prices from vendor listings, Reuters, Bloomberg, Tom's Hardware, TrendForce, SemiAnalysis, and Chinese trade media, as dated per observation. Spot prices move; treat every figure as of its date, not as a quote.