Every AI Chip · Tracked
Every AI chip from every manufacturer and every foundry. Specs, power, cloud pricing, daily GPU rental prices, known buyers, and supply tightness · cross-linked to foundries, memory suppliers, racks, datacenters, and the models trained on them.
- Weighted tightness68
- Lead time multiplier2.09×
- Capex accelerator1.85×
- Frontier holdout premium+60
New since Apr 14, 2026
verified on the source page · specs added after review- Zhenwu M890Alibaba T-Headannounced2026-05-20source
T-Head formally debuted the Zhenwu M890 as its latest AI training and inference processor for datacenter use.
- TPU 8tGoogleannounced2026-04-22source
Google announced TPU 8t as its training-focused eighth-generation TPU for AI workloads.
- TPU 8iGoogleannounced2026-04-22source
Google announced TPU 8i as its inference-focused eighth-generation TPU for AI workloads.
Tier Ladder
Frontier · Mainstream · Specialized · Legacy · set in the curated dataset
Current-generation flagship silicon shipping at scale to hyperscalers. The chips training the biggest models right now.
Current-gen non-flagship or last-gen flagship still deployed widely. The workhorses of the AI economy.
Wafer-scale, custom hyperscaler silicon, or experimental architectures serving niche workloads.
Older chips being phased out or already past EOL. Still used in research clusters and price-sensitive workloads.
Performance per Watt
FP16 TFLOPs vs TDP · log scale · bubble size = HBM capacity
Up-and-to-the-left wins. Chips above the 4 T/W line are the efficiency frontier · that's where the power-budget game is won at hyperscale. Wafer-scale and multi-die superchips sit top-right because they trade watts for raw throughput.
Chip Demand Ladder
How tight supply is · hand-curated from order books and earnings calls, dated with the dataset
Supply notes
- GB200Sold out through 2027 · Oracle, Microsoft, Meta competing
- GB300xAI Memphis + Stargate phase 2
- B200CoWoS-L packaging capacity is the bottleneck
- Ascend 910CChina export-ban alternative · SMIC N7 constrained
- WSE-3G42 Condor Galaxy absorbing most wafers
- B300Ultra variant ramping Q4 2025
This is the per-chip preview of the AI Compute Demand Index landing on /compute · unbounded above 100, goes to 250+ when the market is on fire. Updated weekly from hyperscaler order books and earnings disclosures.
Best for
Six curated shortlists computed from the spec data
Most raw FP16 throughput
Top 5 by FP16/BF16 TFLOPs · ignore price and power · pure compute
Most efficient (TFLOPs/W)
Best FP16 TFLOPs per Watt · the chips that make datacenter power budgets work
Most HBM capacity
Biggest memory for fitting 400B+ parameter models in a single node
Cheapest cloud rental
Lowest $/h in public clouds · best for quick experiments
Current flagship chips
The chips training the biggest frontier models right now
Full leaderboard
Sorted by Gecko score · composite of tier, recency, adoption, perf density
| # | Chip | Maker | Score |
|---|---|---|---|
| 1 | NVIDIA | 94 | |
| 2 | NVIDIA | 86 | |
| 3 | 83 | ||
| 4 | AWS | 78 | |
| 5 | NVIDIA | 75 | |
| 6 | AMD | 75 | |
| 7 | Huawei | 75 | |
| 8 | 71 | ||
| 9 | NVIDIA | 70 | |
| 10 | AMD | 68 | |
| 11 | AMD | 65 | |
| 12 | Cerebras | 60 | |
| 13 | Intel | 57 | |
| 14 | NVIDIA | 55 | |
| 15 | NVIDIA | 55 | |
| 16 | Meta | 52 | |
| 17 | 49 | ||
| 18 | Microsoft | 42 | |
| 19 | Groq | 34 | |
| 20 | NVIDIA | 31 |
By manufacturer
10 companies · 3 foundries · 2 countries
Frequently asked
Computed from the dataset on this page · see its as-of date
Which AI chip has the most FP16 TFLOPs?
GB300 leads dense FP16 throughput among tracked GPUs and accelerators at 5,000 TFLOPs, followed by TPU v7 (4,614) and GB200 (4,500). Wafer-scale chips are left out because one wafer replaces many accelerators. Figures come from manufacturer datasheets, as of Apr 14, 2026.
Which AI chip is most efficient per Watt?
Among tracked chips with a published TDP, TPU v7 leads at 7.69 FP16 TFLOPs per Watt, followed by MTIA v2 (3.93) and TPU v6e (2.62). The median across tracked chips is 1.64 TFLOPs per Watt.
Who manufactures AI chips besides NVIDIA?
10 companies make the AI accelerators tracked on BenchGecko: NVIDIA, Google, AWS, AMD, Huawei, Cerebras, Intel, Meta, Microsoft and Groq. Hyperscalers also design their own silicon (TPU v7, Trainium2, TPU v6e, MTIA v2, TPU v5p and Maia 100) alongside NVIDIA and AMD GPUs.
Which fabs make AI chips?
90% of tracked chips are fabbed at TSMC. Other foundries in the dataset: SMIC (Ascend 910C) · GlobalFoundries (LPU).
How do you decide which chip is "frontier" tier?
Frontier means current-generation flagship silicon shipping at scale to hyperscalers. Mainstream means current-generation non-flagship parts or last-generation flagships still widely deployed. Specialized means wafer-scale, LPU and other custom designs. Announced means not yet shipping in volume, and legacy means end of life or phasing out. Tiers are set in the curated dataset and dated with it.
Where does BenchGecko get chip data from?
Manufacturer datasheets for specs, SEC filings and earnings transcripts for buyer disclosures, cloud provider list prices for $/hour, and press releases for release dates. GPU rental prices on chip pages refresh daily from the Vast.ai marketplace. Every chip page links its sources.
See also
Keep exploring the compute graph