Capacity
Thresholds
23 facility inventories sum to a month of threshold-scale compute
Theoretical dense FP16 capacity, summed across each admitted inventory under shared utilization assumptions. This does not establish that the chips can train one model together. Article 51 concerns a model’s actual training compute.
Plate I Time to 10²⁵ FLOP 184 sites · none sampled out
Plate I. One point per facility, ranked left to right by the days of 24-hour sustained operation its estimated daily FLOP budget would need to accumulate 10²⁵ FLOP, with the whole-site budget allocated to a single model at the methodology's utilization factors.
The vertical axis is that duration on a log scale, bounded at one day and ten years to keep shorter durations legible; values outside those bounds sit at the edges of the plot.
The first three counts overlap: a site reaching the line in twelve days appears in the month, quarter and year counts. A site’s estimate follows the complete reconciled component inventory of an operational compute facility; unsupported or incomplete inventories are withheld.
The plot contains 184 estimable sites, and those reaching the line within a year span 21 countries. The stored catalogue snapshot holds 4,582 facilities, 514 of which carry a chip count or a power figure; the population filter then removes 11 duplicate site records, 26 earlier phases of operating clusters whose later phase already counts the same chips, and 1 chip-manufacturing record, whose power is not training capacity.
Planned and non-operational facilities are excluded. Recorded inventories may still overstate usable capacity; operational status does not establish training availability. 292 further sites had an input the estimator could not use.
Article 51 presumes high-impact capabilities for general-purpose models trained above 10²⁵ FLOP; a provider can rebut that presumption, and the Commission can designate a model on other capability criteria. Of 184 estimable sites, 23 could accumulate that total within a month under the assumptions below, the fastest in 1 day.
Facility and model compute estimates compared with regulatory thresholds and policy proposals. The EU systemic-risk register is a dated reconstruction.
5 corrections · latest Sep 26, 2026
- Sep 26, 2026 · Compute estimates: communication losses discounted twice
- Sep 23, 2026 · Capacity totals: planned capacity counted as present capacity
- Sep 21, 2026 · Gadi: GPU partitions in the count and the compute estimate
- Sep 19, 2026 · Condor Galaxy 1: Cerebras systems no longer given an H100 rate
- Sep 7, 2026 · Stargate UAE chip inventory withheld
Per site, per regime
Facility estimates and evidence
Choosing a regime recalculates durations for admitted facility estimates against its numerical threshold. Other records retain their evidence reasons. The reverse calculator uses hypothetical inputs and the shared dense FP16 specifications and assumptions.
Estimated daily FLOP and days to the selected line, for sites with a usable input.
EU AI Act: Systemic Risk GPAI
European Union: 10²⁵ FLOP
Regulation (EU) 2024/1689, Article 51(2) + Annex XIII · Official Journal of the European Union · Tier 1
Article 51(2): "A general-purpose AI model shall be presumed to have high impact capabilities ... when the cumulative amount of computation used for its training measured in floating point operations is greater than 10^25." Article 52 provides for notification and a provider’s substantiated rebuttal; the Commission may also designate models on other capability or impact grounds. The July 2025 guidelines explain training-compute estimation and GPAI scope. Their indicative 10²³-FLOP criterion is not an exemption for every smaller model. EU market scope matters regardless of the provider’s location. Models placed on the market before August 2, 2025 have until August 2, 2027 to comply with the applicable GPAI obligations.
FACILITIES ≤ 30 DAYS (10²⁵ FLOP)
FACILITIES ≤ 90 DAYS
FACILITIES ≤ 1 YEAR
MODELS ABOVE EU
MODELS ABOVE 10²⁶ FLOP
The two model counts come from the 17-model frontier set behind the Frontier Models tab, and “above EU” is the part of it whose Epoch estimate clears 1025 FLOP. The Article 52(6) register lower on this page draws on a broader extract, including unreleased models and separately listed base-model rows, so it holds more rows.
184 admitted · 292 withheld
Print includes the complete admitted and withheld facility ledger, regardless of the selected evidence population, search or page. Open Facility Analysis before printing this ledger.
184 of 184 facilities
| Facility | Country | Daily FLOP | Days to 10²⁵ FLOP | Estimation Path | Regime Exposure | Expand details |
|---|---|---|---|---|---|---|
| US | ~1 | Hardware: NVIDIA H100 SXM5 80GB (200,000); NVIDIA GB200 (30,000) | No country match | |||
| US | ~3 | Hardware: NVIDIA H100 SXM5 80GB (100,000) | No country match | |||
| US | ~3 | Hardware: NVIDIA H100 SXM5 80GB (100,000) | No country match | |||
| US | ~4 | Hardware: NVIDIA H200 SXM (65,536) | No country match | |||
| US | ~6 | Hardware: NVIDIA H100 SXM5 80GB (50,000) | No country match | |||
| US | ~7 | Hardware: NVIDIA H200 SXM (42,000) | No country match | |||
| US | ~8 | Hardware: NVIDIA H100 SXM5 80GB (35,000) | No country match | |||
| US | ~9 | Hardware: NVIDIA H100 SXM5 80GB (32,000) | No country match | |||
| US | ~12 | Hardware: NVIDIA H100 SXM5 80GB (24,576) | No country match | |||
| US | ~12 | Hardware: NVIDIA H100 SXM5 80GB (24,576) | No country match | |||
| US | ~14 | Hardware: AMD Instinct MI300X (16,384) | No country match | |||
| US | ~15 | Hardware: NVIDIA H100 SXM5 80GB (20,000) | No country match | |||
| US | ~15 | Hardware: NVIDIA H100 SXM5 80GB (20,000) | No country match | |||
| KR | ~17 | Hardware: NVIDIA B200 (7,656) | No country match | |||
| NO | ~18 | Hardware: NVIDIA H100 SXM5 80GB (16,384) | No country match | |||
| US | ~18 | Hardware: NVIDIA H100 SXM5 80GB (16,384) | No country match | |||
| US | ~20 | Hardware: MI250X (37,888) | No country match | |||
| US | ~20 | Hardware: AMD Radeon Instinct MI250X (37,632) | No country match | |||
| US | ~20 | Hardware: NVIDIA H100 SXM5 80GB (14,400) | No country match | |||
| US | ~27 | Hardware: NVIDIA H100 SXM5 80GB (10,752) | No country match | |||
| US | ~28 | Hardware: NVIDIA A100 (32,768) | No country match | |||
| US | ~29 | Hardware: NVIDIA H100 SXM5 80GB (10,000) | No country match | |||
| US | ~29 | Hardware: NVIDIA H100 SXM5 80GB (10,000) | No country match | |||
| ~36 | Hardware: NVIDIA H100 SXM5 80GB (8,192) | HQ MATCH | ||||
| US | ~37 | Hardware: NVIDIA H100 SXM5 80GB (8,000) | No country match | |||
| FI | ~37 | Hardware: NVIDIA H100 SXM5 80GB (8,000) | COUNTRY MATCH | |||
| US | ~37 | Hardware: NVIDIA A100 (25,000) | No country match | |||
| JP | ~48 | Hardware: NVIDIA H200 SXM (6,128) | No country match | |||
| IT | ~54 | Hardware: AMD Radeon Instinct MI250X (13,888) | COUNTRY MATCH | |||
| US | ~58 | Hardware: NVIDIA A100 (16,000) | No country match | |||
| IS | ~59 | Hardware: NVIDIA H200 SXM (4,992) | HQ MATCH | |||
| FI | ~63 | Hardware: AMD Radeon Instinct MI250X (11,912) | COUNTRY MATCH | |||
| US | ~63 | Hardware: NVIDIA H100 SXM5 80GB (4,608) | No country match | |||
| ES | ~65 | Hardware: NVIDIA H100 SXM5 80GB (4,480) | COUNTRY MATCH | |||
| US | ~66 | Hardware: NVIDIA H100 SXM5 80GB (4,424) | No country match | |||
| IT | ~67 | Hardware: NVIDIA A100 (13,824) | COUNTRY MATCH | |||
| US | ~67 | Hardware: NVIDIA A100 (13,760) | No country match | |||
| FR | ~71 | Hardware: NVIDIA H100 SXM5 80GB (4,096) | COUNTRY MATCH | |||
| FR | ~71 | Hardware: NVIDIA H200 SXM (4,096) | COUNTRY MATCH | |||
| IN | ~71 | Hardware: NVIDIA H100 SXM5 80GB (4,096) | No country match | |||
| US | ~73 | Hardware: NVIDIA H100 SXM5 80GB (4,000) | No country match | |||
| US | ~73 | Hardware: NVIDIA H100 SXM5 80GB (4,000) | No country match | |||
| ~73 | Hardware: NVIDIA H100 SXM5 80GB (4,000) | No country match | ||||
| US | ~73 | Hardware: NVIDIA H200 SXM (4,000) | HQ MATCH | |||
| SE | ~73 | Hardware: NVIDIA H100 SXM5 80GB (4,000) | COUNTRY MATCH | |||
| US | ~74 | Hardware: AMD Instinct MI300X (3,000) | No country match | |||
| FI | ~74 | Hardware: MI250X (10,240) | COUNTRY MATCH | |||
| US | ~75 | Hardware: NVIDIA H100 SXM5 80GB (3,632); NVIDIA A100 SXM4 40 GB (768) | No country match | |||
| US | ~105 | Hardware: NVIDIA Tesla V100 SXM2 (22,000) | No country match | |||
| US | ~116 | Hardware: NVIDIA A100 (8,000) | No country match | |||
| KR | ~117 | Hardware: NVIDIA H100 SXM5 80GB (2,496) | No country match | |||
| US | ~126 | Hardware: NVIDIA A100 SXM4 80 GB (7,360) | No country match | |||
| US | ~129 | Hardware: NVIDIA A100 SXM4 40 GB (6,144); NVIDIA A100 SXM4 80 GB (1,024) | No country match | |||
| JP | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | |||
| US | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | |||
| GB | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | |||
| ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | ||||
| ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | ||||
| NO | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | |||
| NO | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | |||
| IL | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,048) | No country match | |||
| JP | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,040) | No country match | |||
| JP | ~143 | Hardware: NVIDIA H100 SXM5 80GB (2,040) | No country match | |||
| JP | ~145 | Hardware: NVIDIA H100 SXM5 80GB (2,016) | No country match | |||
| CA | ~146 | Hardware: NVIDIA H100 SXM5 80GB (2,000) | No country match | |||
| US | ~172 | Hardware: NVIDIA A100 (5,400) | No country match | |||
| DK | ~191 | Hardware: NVIDIA H100 SXM5 80GB (1,528) | COUNTRY MATCH | |||
| FI | ~192 | Hardware: NVIDIA H100 SXM5 80GB (1,520) | COUNTRY MATCH | |||
| US | ~207 | Hardware: NVIDIA A100 (4,480) | No country match | |||
| US | ~221 | Hardware: AMD Instinct MI300X (1,000) | No country match | |||
| AE | ~230 | Hardware: NVIDIA H100 SXM5 80GB (1,272) | No country match | |||
| US | ~230 | Hardware: NVIDIA A100 (4,032) | No country match | |||
| US | ~232 | Hardware: NVIDIA A100 SXM4 40 GB (4,000) | No country match | |||
| ~232 | Hardware: NVIDIA A100 (4,000) | No country match | ||||
| DE | ~248 | Hardware: NVIDIA A100 (3,744) | COUNTRY MATCH | |||
| US | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| TH | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| US | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| US | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| IN | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | ||||
| TH | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| SG | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| TW | ~286 | Hardware: NVIDIA H100 SXM5 80GB (1,024) | No country match | |||
| FR | ~288 | Hardware: NVIDIA H100 SXM5 80GB (1,016) | COUNTRY MATCH | |||
| US | ~292 | Hardware: NVIDIA H100 SXM5 80GB (1,000) | No country match | |||
| KR | ~292 | Hardware: NVIDIA H100 SXM5 80GB (1,000) | No country match | |||
| US | ~292 | Hardware: NVIDIA H100 SXM5 80GB (1,000) | No country match | |||
| SA | ~293 | Hardware: NVIDIA Tesla V100 SXM2 (7,912) | No country match | |||
| JP | ~305 | Hardware: NVIDIA H100 SXM5 80GB (960) | No country match | |||
| US | ~322 | Hardware: NVIDIA A100 SXM4 80 GB (2,880) | No country match | |||
| CA | ~358 | Hardware: NVIDIA H100 SXM5 80GB (816) | No country match | |||
| US | ~393 | Hardware: AMD Radeon Instinct MI250X (1,920) | No country match | |||
| US | ~414 | Hardware: NVIDIA A100 (2,240) | No country match | |||
| KR | ~414 | Hardware: NVIDIA A100 (2,240) | No country match | |||
| US | ~439 | Hardware: NVIDIA A100 (2,112) | No country match | |||
| JP | ~453 | Hardware: NVIDIA A100 (2,048) | No country match | |||
| BR | ~460 | Hardware: NVIDIA A100 (2,016) | No country match | |||
| US | ~464 | Hardware: NVIDIA A100 SXM4 40 GB (2,000) | No country match | |||
| US | ~464 | Hardware: NVIDIA A100 (2,000) | No country match | |||
| KR | ~527 | Hardware: NVIDIA A100 (1,760) | No country match | |||
| US | ~529 | Hardware: NVIDIA A100 (1,752) | No country match | |||
| SE | ~538 | Hardware: NVIDIA H100 SXM5 80GB (544) | COUNTRY MATCH | |||
| FR | ~559 | Hardware: AMD Radeon Instinct MI250X (1,352) | COUNTRY MATCH | |||
| US | ~580 | Hardware: NVIDIA H100 SXM5 80GB (504) | No country match | |||
| US | ~580 | Hardware: NVIDIA H100 SXM5 80GB (504) | No country match | |||
| RU | ~583 | Hardware: NVIDIA A100 SXM4 80 GB (1,592) | No country match | |||
| ~585 | Hardware: NVIDIA H100 SXM5 80GB (500) | No country match | ||||
| FR | ~620 | Hardware: NVIDIA A100 (416); NVIDIA Tesla V100 SXM2 (2,696) | COUNTRY MATCH | |||
| KR | ~630 | Hardware: AMD Radeon Instinct MI250X (1,200) | No country match | |||
| AU | ~653 | Hardware: NVIDIA H100 SXM5 80GB (448) | No country match | |||
| NL | ~660 | Hardware: NVIDIA H100 SXM5 80GB (352); NVIDIA A100 SXM4 80 GB (288) | COUNTRY MATCH | |||
| SA | ~703 | Hardware: NVIDIA H100 SXM5 80GB (416) | No country match | |||
| US | ~707 | Hardware: NVIDIA A100 (1,312) | No country match | |||
| US | ~707 | Hardware: NVIDIA A100 (1,312) | No country match | |||
| NL | ~707 | Hardware: NVIDIA A100 (1,312) | COUNTRY MATCH | |||
| US | ~707 | Hardware: NVIDIA A100 (1,312) | No country match | |||
| US | ~739 | Hardware: NVIDIA H100 SXM5 80GB (296); NVIDIA A100 SXM4 80 GB (316) | No country match | |||
| KR | ~828 | Hardware: NVIDIA A100 (1,120) | No country match | |||
| US | ~828 | Hardware: NVIDIA A100 (1,120) | No country match | |||
| PL | ~840 | Hardware: NVIDIA H100 SXM5 80GB (348) | COUNTRY MATCH | |||
| RU | ~846 | Hardware: NVIDIA A100 SXM4 40 GB (1,096) | No country match | |||
| RU | ~852 | Hardware: NVIDIA A100 SXM4 80 GB (1,088) | No country match | |||
| KR | ~892 | Hardware: NVIDIA A100 (1,040) | No country match | |||
| KR | ~914 | Hardware: NVIDIA H100 SXM5 80GB (320) | No country match | |||
| PL | ~962 | Hardware: NVIDIA H100 SXM5 80GB (304) | COUNTRY MATCH | |||
| AU | ~984 | Hardware: AMD Radeon Instinct MI250X (768) | No country match | |||
| JP | ~999 | Hardware: NVIDIA A100 SXM4 80 GB (928) | No country match | |||
| ~1,142 | Hardware: NVIDIA H100 SXM5 80GB (256) | No country match | ||||
| TW | ~1,148 | Hardware: NVIDIA Tesla V100 SXM2 (2,016) | No country match | |||
| LU | ~1,159 | Hardware: NVIDIA A100 (800) | COUNTRY MATCH | |||
| KR | ~1,171 | Hardware: NVIDIA A100 (792) | No country match | |||
| RU | ~1,171 | Hardware: NVIDIA A100 SXM4 80 GB (792) | No country match | |||
| IT | ~1,179 | Hardware: NVIDIA H100 SXM5 80GB (248) | COUNTRY MATCH | |||
| IS | ~1,179 | Hardware: NVIDIA H100 SXM5 80GB (248) | No country match | |||
| DE | ~1,208 | Hardware: NVIDIA A100 SXM4 40 GB (768) | COUNTRY MATCH | |||
| DE | ~1,208 | Hardware: NVIDIA A100 (768) | COUNTRY MATCH | |||
| SE | ~1,233 | Hardware: NVIDIA A100 SXM4 80 GB (752) | COUNTRY MATCH | |||
| DE | ~1,253 | Hardware: NVIDIA A100 (740) | COUNTRY MATCH | |||
| TH | ~1,317 | Hardware: NVIDIA A100 (704) | No country match | |||
| BR | ~1,317 | Hardware: NVIDIA A100 (704) | No country match | |||
| IN | ~1,414 | Hardware: NVIDIA A100 (656) | No country match | |||
| GB | ~1,449 | Hardware: NVIDIA A100 (640) | No country match | |||
| CA | ~1,458 | Hardware: NVIDIA A100 (636) | No country match | |||
| DE | ~1,467 | Hardware: NVIDIA A100 (632) | COUNTRY MATCH | |||
| CZ | ~1,610 | Hardware: NVIDIA A100 (576) | COUNTRY MATCH | |||
| DE | ~1,811 | Hardware: NVIDIA A100 (512) | COUNTRY MATCH | |||
| US | ~1,811 | Hardware: NVIDIA A100 (512) | No country match | |||
| US | ~1,811 | Hardware: NVIDIA A100 (512) | No country match | |||
| GB | ~2,070 | Hardware: NVIDIA A100 (448) | No country match | |||
| ~2,108 | Hardware: NVIDIA A100 (440) | No country match | ||||
| JP | ~2,128 | Hardware: NVIDIA Tesla V100 SXM2 (1,088) | No country match | |||
| US | ~2,319 | Hardware: NVIDIA A100 (400) | No country match | |||
| PL | ~2,415 | Hardware: NVIDIA A100 SXM4 40 GB (384) | COUNTRY MATCH | |||
| US | ~2,576 | Hardware: NVIDIA A100 SXM4 40 GB (360) | No country match | |||
| JP | ~2,576 | Hardware: NVIDIA A100 (360) | No country match | |||
| JP | ~2,619 | Hardware: NVIDIA Tesla V100 SXM2 (884) | No country match | |||
| SG | ~2,635 | Hardware: NVIDIA A100 (352) | No country match | |||
| JP | ~2,760 | Hardware: NVIDIA A100 SXM4 80 GB (336) | No country match | |||
| IT | ~2,898 | Hardware: NVIDIA A100 (320) | COUNTRY MATCH | |||
| US | ~2,898 | Hardware: NVIDIA A100 (320) | No country match | |||
| GB | ~2,898 | Hardware: NVIDIA A100 (320) | No country match | |||
| JP | ~3,014 | Hardware: NVIDIA Tesla V100 SXM2 (768) | No country match | |||
| US | ~3,312 | Hardware: NVIDIA A100 (280) | No country match | |||
| DE | ~3,410 | Hardware: NVIDIA A100 SXM4 40 GB (272) | COUNTRY MATCH | |||
| US | ~3,513 | Hardware: NVIDIA A100 SXM4 40 GB (264) | No country match | |||
| US | ~3,623 | Hardware: NVIDIA A100 (256) | No country match | |||
| US | ~3,680 | Hardware: NVIDIA A100 PCIe (252) | No country match | |||
| SI | ~3,864 | Hardware: NVIDIA A100 PCIe (240) | COUNTRY MATCH | |||
| US | ~4,019 | Hardware: NVIDIA Tesla V100 SXM2 (576) | No country match | |||
| GB | ~4,593 | Hardware: NVIDIA Tesla V100 SXM2 (504) | No country match | |||
| BR | ~4,637 | Hardware: NVIDIA A100 (200) | No country match | |||
| US | ~4,637 | Hardware: NVIDIA A100 (200) | No country match | |||
| US | ~4,830 | Hardware: NVIDIA A100 (192) | No country match | |||
| FR | ~4,830 | Hardware: NVIDIA A100 (192) | COUNTRY MATCH | |||
| FR | ~4,830 | Hardware: NVIDIA A100 (192) | COUNTRY MATCH | |||
| US | ~4,830 | Hardware: NVIDIA A100 (192) | No country match | |||
| DE | ~4,830 | Hardware: NVIDIA A100 (192) | COUNTRY MATCH | |||
| RU | ~5,796 | Hardware: NVIDIA A100 (160) | No country match | |||
| US | ~5,796 | Hardware: NVIDIA A100 (160) | No country match | |||
| VN | ~5,796 | Hardware: NVIDIA A100 (160) | No country match | |||
| BR | ~6,290 | Hardware: NVIDIA Tesla V100 SXM2 (368) | No country match | |||
| KR | ~6,440 | Hardware: NVIDIA A100 (144) | No country match | |||
| BR | ~7,419 | Hardware: NVIDIA Tesla V100 SXM2 (312) | No country match |
Withheld facility evidence (292)
Daily FLOP and threshold duration are unknown for these records; withheld values do not assert zero compute.
- AWS Project Rainier (New Carlisle, IN)United States
No positive whole-number GPU count is recorded.
- Bridge Data Centres MY06 (Sedenak Tech Park)Malaysia
No positive whole-number GPU count is recorded.
- China Mobile Intelligent Computing Center — HarbinChina
No GPU inventory is recorded.
- China Mobile Smart Computing Center — Hohhot (Horinger Campus)China
No GPU inventory is recorded.
- China Unicom Zhongwei Data CenterChina
No GPU inventory is recorded.
- Condor Galaxy IndiaIndia
Facility status is announced, not operational.
- CoreWeave Kenilworth NJ (NEST Campus)United States
Facility status is under construction, not operational.
- CoreWeave Lancaster PAUnited States
Facility status is announced, not operational.
- Helios (Cyfronet AGH)Poland
No GPU inventory is recorded.
- DayOne Nusajaya Tech ParkMalaysia
No positive whole-number GPU count is recorded.
- DayOne SG1 SingaporeSingapore
Facility status is under construction, not operational.
- Eclairion Bruyères-le-Châtel (Mistral Compute)France
No GPU inventory is recorded.
- Alibaba ZhangbeiChina
No positive whole-number GPU count is recorded.
- Amazon Madison Mega SiteUnited States
No positive whole-number GPU count is recorded.
- Amazon RidgelandUnited States
No positive whole-number GPU count is recorded.
- CoreWeave Chester VAUnited States
No positive whole-number GPU count is recorded.
- CoreWeave Dalton 1 & 2United States
No positive whole-number GPU count is recorded.
- CoreWeave Denton TXUnited States
No positive whole-number GPU count is recorded.
- CoreWeave Ellendale NDUnited States
No positive whole-number GPU count is recorded.
- Coreweave HeliosUnited States
No positive whole-number GPU count is recorded.
- CoreWeave Marble NCUnited States
No positive whole-number GPU count is recorded.
- TeraWulf Lake Mariner CampusUnited States
No positive whole-number GPU count is recorded.
- AGH Cyfronet HeliosPoland
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
440 × NVIDIA GH200Held: No reviewed dense FP16 rate
24 × NVIDIA H100 SXM5 80GB989.5 TFLOP/s dense FP16 per accelerator
- Alps Supercomputer Phase 2Switzerland
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
10,752 × NVIDIA GH200Held: No reviewed dense FP16 rate
- Applied Digital CoreWeave Ellendale Phase 1United States
Facility status is announced, not operational.
45,000 × NVIDIA GB200Held: Not present capacity
- Applied Digital CoreWeave Ellendale Phase 2United States
Facility status is announced, not operational.
110,000 × NVIDIA GB200Held: Not present capacity
- Applied Digital Ellendale Possible Phase 3United States
Facility status is announced, not operational.
180,000 × NVIDIA GB200Held: Not present capacity
- Aramco Groq Inference ClusterSaudi Arabia
GroqChip LPU v1 includes unsupported hardware.
19,725 × GroqChip LPU v1Held: No supported per-chip rate for this architecture
- Argonne NL AuroraUnited States
Intel Data Center GPU Max 1550 is unsupported.
63,744 × Intel Data Center GPU Max 1550Held: No supported per-chip rate for this architecture
- AWS EC2 Trn1United States
Amazon Trainium1 includes unsupported hardware.
30,000 × Amazon Trainium1Held: No supported per-chip rate for this architecture
- AWS Fast BERT TrainingUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
2,048 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Azure OpenAI GPT-3 ClusterUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
10,000 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Brazil Scala AI City Phase 1Brazil
Facility status is announced, not operational.
- Calcul Québec BélugaCanada
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
688 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- CEA EXA1-HE Phase 2France
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
1,908 × NVIDIA GH200Held: No reviewed dense FP16 rate
- CEA EXA1-HE Phase 3France
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
2,688 × NVIDIA GH200Held: No reviewed dense FP16 rate
- Center for Advanced Intelligence Project, RIKEN, RAIDENJapan
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
416 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Cineca Marconi-100Italy
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
3,952 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Condor Galaxy 1 (CG-1)United States
Cerebras CS-2 includes unsupported hardware.
64 × Cerebras CS-2Held: No supported per-chip rate for this architecture
- Condor Galaxy 3 (CG-3)United States
Facility status is announced, not operational.
64 × Cerebras CS-3Held: Not present capacity
- Condor Galaxy 4 (CG-4)Country not recorded
Facility status is announced, not operational.
64 × Cerebras CS-2Held: Not present capacity
- Condor Galaxy 5 (CG-5)Country not recorded
Facility status is announced, not operational.
64 × Cerebras CS-2Held: Not present capacity
- Condor Galaxy 6 (CG-6)Country not recorded
Facility status is announced, not operational.
64 × Cerebras CS-2Held: Not present capacity
- Condor Galaxy 7 (CG-7)Country not recorded
Facility status is announced, not operational.
64 × Cerebras CS-2Held: Not present capacity
- Condor Galaxy 8 (CG-8)Country not recorded
Facility status is announced, not operational.
64 × Cerebras CS-2Held: Not present capacity
- Condor Galaxy 9 (CG-9)Country not recorded
Facility status is announced, not operational.
64 × Cerebras CS-2Held: Not present capacity
- Continental DGX SupercomputerGermany
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
448 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Core42 AI-03United Arab Emirates
AMD Instinct MI210 includes unsupported hardware.
4,320 × AMD Instinct MI210Held: No supported per-chip rate for this architecture
- CoreWeave MuskogeeUnited States
Facility status is announced, not operational.
40,000 × NVIDIA GB200Held: Not present capacity
- Corvex B200sUnited States
Facility status is announced, not operational.
256 × NVIDIA B200Held: Not present capacity
- Crusoe Cheyenne WyomingUnited States
Facility status is announced, not operational.
- Crusoe Goodnight in Claude, TexasUnited States
Facility status is announced, not operational.
- CSC PuhtiFinland
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
320 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- CSCS Piz Daint Phase 2Switzerland
NVIDIA Tesla P100 PCIe 16GB includes unsupported hardware.
5,704 × NVIDIA Tesla P100 PCIe 16GBHeld: No supported per-chip rate for this architecture
- DataVolt Neom 1.5 GW Phase 2Saudi Arabia
Facility status is announced, not operational.
- Eni HPC2Italy
NVIDIA Tesla K20X includes unsupported hardware.
3,000 × NVIDIA Tesla K20XHeld: No supported per-chip rate for this architecture
- Eni HPC4 Phase 1Italy
NVIDIA P100 includes unsupported hardware.
3,170 × NVIDIA P100Held: No supported per-chip rate for this architecture
- Eni HPC4 Phase 2Italy
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
250 × NVIDIA A100312 TFLOP/s dense FP16 per accelerator
2,750 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Eni HPC5Italy
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
7,280 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- ExxonMobil Discovery 6United States
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
4,032 × NVIDIA GH200Held: No reviewed dense FP16 rate
- Fluidstack France Gigawatt CampusFrance
Facility status is announced, not operational.
500,000 × NVIDIA GB200Held: Not present capacity
- Foxconn Big Innovation Cloud AI factoryTaiwan
Facility status is announced, not operational.
5,000 × NVIDIA GB300 (Blackwell Ultra)Held: Not present capacity
5,000 × NVIDIA GB200Held: Not present capacity
- Foxconn Hon Hai Kaohsiung SupercomputerTaiwan
Facility status is announced, not operational.
4,608 × NVIDIA GB200Held: Not present capacity
- FPT AI Factory JapanJapan
Facility status is announced, not operational.
3,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- FPT AI Factory VietnamVietnam
Facility status is announced, not operational.
3,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- G42 ArtemisUnited Arab Emirates
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
1,296 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- G42 Microsoft 100 MW UAE ClusterUnited Arab Emirates
Facility status is announced, not operational.
55,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- G42 Microsoft 30 MW UAE Cluster AUnited Arab Emirates
Facility status is announced, not operational.
15,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- G42 Microsoft 30 MW UAE Cluster BUnited Arab Emirates
Facility status is announced, not operational.
15,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Gcore data center Phase 2South Korea
Facility status is announced, not operational.
500 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
500 × NVIDIA GB200Held: Not present capacity
- Gemini 1.0 Ultra training cluster AUnited States
No dense FP16 rate for Google TPU v4 has passed evidence review.
28,672 × Google TPU v4Held: No reviewed dense FP16 rate
- Google A3 VMsUnited States
Facility status is announced, not operational.
26,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Google Hypercomputer TPU v5p podCountry not recorded
No dense FP16 rate for Google TPU v5p has passed evidence review.
8,960 × Google TPU v5pHeld: No reviewed dense FP16 rate
- Google MLPerf 0.7 SubmissionUnited States
Google TPU v3 includes unsupported hardware.
4,096 × Google TPU v3Held: No supported per-chip rate for this architecture
- Google New Albany OhioUnited States
Facility status is announced, not operational.
- Google Oklahoma TPU v4 PodsUnited States
No dense FP16 rate for Google TPU v4 has passed evidence review.
32,768 × Google TPU v4Held: No reviewed dense FP16 rate
- Google TensorFlow Research CloudUnited States
Google TPU v2 includes unsupported hardware.
4,096 × Google TPU v2Held: No supported per-chip rate for this architecture
- Google TPU v4 PodUnited States
No dense FP16 rate for Google TPU v4 has passed evidence review.
4,096 × Google TPU v4Held: No reviewed dense FP16 rate
- Google TPUv3 POD GenericCountry not recorded
Google TPU v3 includes unsupported hardware.
1,024 × Google TPU v3Held: No supported per-chip rate for this architecture
- Google TPUv5eCountry not recorded
No dense FP16 rate for Google TPU v5e has passed evidence review.
50,944 × Google TPU v5eHeld: No reviewed dense FP16 rate
- GSIC TSUBAME 2.5Japan
NVIDIA Tesla K20X includes unsupported hardware.
4,224 × NVIDIA Tesla K20XHeld: No supported per-chip rate for this architecture
- GSIC TSUBAME 3.0Japan
NVIDIA Tesla P100 SXM2 includes unsupported hardware.
2,156 × NVIDIA Tesla P100 SXM2Held: No supported per-chip rate for this architecture
- HPE GreenLake for LLMs Q01Canada
No GPU inventory is recorded.
- Huawei Pangu Ultra MoE 910BsCountry not recorded
No dense FP16 rate for Huawei Ascend 910B has passed evidence review.
6,000 × Huawei Ascend 910BHeld: No reviewed dense FP16 rate
- HUMAIN Saudi Arabia/NVIDIA Phase 1Saudi Arabia
Facility status is announced, not operational.
18,000 × NVIDIA GB300 (Blackwell Ultra)Held: Not present capacity
- HUMAIN Saudi Arabia Phase 2Saudi Arabia
Facility status is announced, not operational.
- IBM DYEUSUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
296 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- iGenius ColosseumItaly
Facility status is announced, not operational.
5,760 × NVIDIA GB200Held: Not present capacity
- Immunity Bio ClusterUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
320 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- IndiaAI Mission SupercomputerIndia
Facility status is announced, not operational.
- Inflection AI ClusterUnited States
Facility status is announced, not operational.
22,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Intel Stability Gaudi 2United States
Intel Habana Gaudi2 includes unsupported hardware.
4,000 × Intel Habana Gaudi2Held: No supported per-chip rate for this architecture
- IREN ChildressUnited States
Facility status is announced, not operational.
- IREN Sweetwater 1United States
Facility status is announced, not operational.
- IREN Sweetwater 2United States
Facility status is announced, not operational.
- JAMSTEC ZettaScaler-2.2 GyoukouJapan
PEZY-SC2 is unsupported.
10,000 × PEZY-SC2Held: No supported per-chip rate for this architecture
- JCAHPC MiyabiJapan
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
1,120 × NVIDIA GH200Held: No reviewed dense FP16 rate
- Jean Zay Supercomputer Phase 4France
GPU inventory total 1,872 does not equal total GPU count 3,704.
1,456 × NVIDIA H100 SXM5 80GB989.5 TFLOP/s dense FP16 per accelerator
416 × NVIDIA A100 SXM4 80 GB312 TFLOP/s dense FP16 per accelerator
- Jupiter, JülichGermany
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
23,536 × NVIDIA GH200Held: No reviewed dense FP16 rate
- KAUST Shaheen-IIISaudi Arabia
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
2,800 × NVIDIA GH200Held: No reviewed dense FP16 rateKAUST press release 2025-11-17 and KSL Shaheen III documentation
- KDDI Sharp SakaiJapan
Facility status is announced, not operational.
2,000 × NVIDIA GB200Held: Not present capacity
- Lawrence Livermore NL El Capitan Phase 2United States
AMD Instinct MI300A includes unsupported hardware.
44,544 × AMD Instinct MI300AHeld: No supported per-chip rate for this architecture
- Lawrence Livermore NL Lassen Phase 2United States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports. 2 of 2 partitions are held.
3,168 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
1 × Cerebras CS-1Held: No supported per-chip rate for this architecture
- Lawrence Livermore NL RZAdamsUnited States
AMD Instinct MI300A includes unsupported hardware.
512 × AMD Instinct MI300AHeld: No supported per-chip rate for this architecture
- Lawrence Livermore NL SierraUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
17,280 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Lawrence Livermore NL TuolumneUnited States
AMD Instinct MI300A includes unsupported hardware.
4,608 × AMD Instinct MI300AHeld: No supported per-chip rate for this architecture
- Los Alamos NL VenadoUnited States
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
2,560 × NVIDIA GH200Held: No reviewed dense FP16 rate
- Meta 2017 P100 ClusterUnited States
NVIDIA P100 includes unsupported hardware.
992 × NVIDIA P100Held: No supported per-chip rate for this architecture
- Meta Louisiana DatacenterUnited States
Facility status is announced, not operational.
- Microsoft Ares/MaiaUnited States
Maia 100 (M100) is unsupported.
2,048 × Maia 100 (M100)Held: No supported per-chip rate for this architecture
- Microsoft Azure Immunity BioUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
1,250 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Microsoft Azure ND v2 Largest StatedUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
800 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Microsoft Research HyperclusterUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
576 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Microsoft Sweden GävleSweden
Facility status is announced, not operational.
- Microsoft Sweden SandvikenSweden
Facility status is announced, not operational.
- Microsoft Sweden StaffanstorpSweden
Facility status is announced, not operational.
- MIT LLSC TX-GAIAUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
896 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- MIT SupercloudUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
488 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- MITRE Federal AI SandboxUnited States
Facility status is announced, not operational.
256 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Moscow State University Lomonosov 2Russia
NVIDIA Tesla K40m includes unsupported hardware.
1,472 × NVIDIA Tesla K40mHeld: No supported per-chip rate for this architecture
- MSU-270Russia
No GPU inventory is recorded.
- NCI Australia GadiAustralia
NVIDIA H200 141GB does not identify an H200 form factor.
640 × NVIDIA Tesla V100 SXM2125 TFLOP/s dense FP16 per accelerator
120 × NVIDIA H200 141GBHeld: Variant not stated; rate depends on itDocumented variants: H200 SXM 989.5; H200 NVL 835.5 TFLOP/s; none applied.NCI HPC systems specification and NCI Annual Report 2025 director's message
16 × NVIDIA A100 80GB312 TFLOP/s dense FP16 per acceleratorNCI HPC systems specification and NCMAS 2024 applicant information (2023-11-09)
- Nebius Finland Phase 2Finland
Facility status is announced, not operational.
60,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Nebius Kansas City Phase 2United States
Facility status is announced, not operational.
8,000 × NVIDIA B200Held: Not present capacity
18,000 × NVIDIA H200 SXMHeld: Not present capacity
- Nebius New JerseyUnited States
Facility status is announced, not operational.
150,000 × NVIDIA GB200Held: Not present capacity
- Neevcloud planned GPUsIndia
Facility status is announced, not operational.
- Nscale LoughtonUnited Kingdom
Facility status is announced, not operational.
45,000 × NVIDIA GB200Held: Not present capacity
- NVIDIA CoreWeave Eos-DFW Rumored Phase 2United States
Facility status is announced, not operational.
32,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- NVIDIA DGX SuperPOD 2019Country not recorded
NVIDIA Tesla V100 DGXS 32 GB is not identified as the V100 SXM2, the only V100 the estimator supports.
1,536 × NVIDIA Tesla V100 DGXS 32 GBHeld: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- NVIDIA HeliosCountry not recorded
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
1,024 × NVIDIA GH200Held: No reviewed dense FP16 rate
- NVIDIA Israel Blackwell SupercomputerIsrael
Facility status is announced, not operational.
- NVIDIA SATURN V Phase 3United States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
280 × NVIDIA A100312 TFLOP/s dense FP16 per accelerator
5,280 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- NVIDIA SATURN V VoltaUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
264 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- NVIDIA Taipei-1Taiwan
NVIDIA L40 includes unsupported hardware.
512 × NVIDIA H100 SXM5 80GB989.5 TFLOP/s dense FP16 per accelerator
256 × NVIDIA L40Held: No supported per-chip rate for this architecture
- Oak Ridge NL SummitUnited States
Facility status is decommissioned, not operational.
27,648 × NVIDIA Tesla V100 SXM2Held: Not present capacity
- Oak Ridge NL TitanUnited States
Facility status is decommissioned, not operational.
18,688 × NVIDIA Tesla K20XHeld: Not present capacity
- OneAsia OBON ClustersThailand
Facility status is announced, not operational.
4,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- OpenAI/Microsoft AtlantaUnited States
Facility status is announced, not operational.
300,000 × NVIDIA B200Held: Not present capacity
- OpenAI/Microsoft Mt Pleasant, Wisconsin Phase 1United States
Facility status is announced, not operational.
150,000 × NVIDIA GB200Held: Not present capacity
- OpenAI/Microsoft Mt Pleasant, Wisconsin Phase 2United States
Facility status is announced, not operational.
700,000 × NVIDIA GB200Held: Not present capacity
- Oracle OCI Supercluster B200sUnited States
Facility status is announced, not operational.
131,072 × NVIDIA B200Held: Not present capacity
- Oracle Vantage Data Centers FrontierUnited States
Facility status is announced, not operational.
- PanaAI AUS AISFAustralia
Facility status is announced, not operational.
4,088 × NVIDIA H200 SXMHeld: Not present capacity
- Paper on AlphaZeroUnited States
Google TPU v1 includes unsupported hardware.
5,000 × Google TPU v1Held: No supported per-chip rate for this architecture
- Paper on PaLMUnited States
No dense FP16 rate for Google TPU v4 has passed evidence review.
6,144 × Google TPU v4Held: No reviewed dense FP16 rate
- Petrobras AtlasBrazil
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
1,088 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Petrobras DragãoBrazil
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
2,176 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Petrobras Fênix Phase 2Brazil
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
720 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Poolside 10k ClusterCountry not recorded
Facility status is announced, not operational.
10,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Preferred Networks MN-1bJapan
Facility status is decommissioned, not operational.
512 × NVIDIA V100Held: Not present capacityDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
1,024 × NVIDIA P100Held: Not present capacity
- Preferred Networks MN-2Japan
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
1,024 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Project Ceiba Phase 2United States
Facility status is announced, not operational.
20,736 × NVIDIA GB200Held: Not present capacity
- Reliance Industries SupercomputerIndia
Facility status is announced, not operational.
450,000 × NVIDIA GB200Held: Not present capacity
- RPI AiMOSUnited States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
1,512 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- RPI Supercomputer 2United States
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
608 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- S. Korea 6th national supercomputerSouth Korea
Facility status is announced, not operational.
8,800 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Sakura's B200s Phase 2Japan
Facility status is announced, not operational.
4,000 × NVIDIA B200Held: Not present capacity
6,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Sandia NL El DoradoUnited States
AMD Instinct MI300A includes unsupported hardware.
1,520 × AMD Instinct MI300AHeld: No supported per-chip rate for this architecture
- Saudi Data & AI Authority Sovereign AI factorySaudi Arabia
Facility status is announced, not operational.
5,000 × NVIDIA GB200Held: Not present capacity
- SberCloud ChristofariRussia
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
1,200 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Sesterce Grand Est France AFrance
Facility status is announced, not operational.
150,000 × NVIDIA GB200Held: Not present capacity
- Sesterce Grand Est France BFrance
Facility status is announced, not operational.
150,000 × NVIDIA GB200Held: Not present capacity
- Sesterce PegasusCountry not recorded
Facility status is announced, not operational.
10,496 × NVIDIA GB200Held: Not present capacity
- Sesterce Southern France 250MWFrance
Facility status is announced, not operational.
120,000 × NVIDIA GB200Held: Not present capacity
- Sesterce ValenceFrance
Facility status is announced, not operational.
40,000 × NVIDIA GB200Held: Not present capacity
- Simon Fraser University/Compute Canada CedarCanada
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
768 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Singtel DC TuasSingapore
Facility status is announced, not operational.
- SK Group AWS Uslan Phase 2South Korea
Facility status is announced, not operational.
60,000 × NVIDIA GB200Held: Not present capacity
- SoftBank Planned B200 SuperpodJapan
Facility status is announced, not operational.
4,000 × NVIDIA B200Held: Not present capacity
- Sustainable Metal Cloud Singapore Phase 2Singapore
Facility status is announced, not operational.
5,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- TACC FronteraUnited States
NVIDIA Quadro RTX 5000 is unsupported. 2 of 2 partitions are held.
360 × NVIDIA Quadro RTX 5000Held: No supported per-chip rate for this architecture
448 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- TACC LonghornUnited States
Facility status is decommissioned, not operational.
448 × NVIDIA Tesla V100 SXM2Held: Not present capacity
- Telangana Yotta H1 Hyderbad AI City Cluster Phase 1India
Facility status is announced, not operational.
4,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Telangana Yotta Hyderbad AI City Cluster Phase 2India
Facility status is announced, not operational.
- TensorWave MI300X Cluster 1 Phase 2United States
Facility status is announced, not operational.
10,000 × AMD Instinct MI300XHeld: Not present capacity
- TensorWave MI300X Cluster 2United States
Facility status is announced, not operational.
10,000 × AMD Instinct MI300XHeld: Not present capacity
- Tesla Cortex Phase 2United States
Facility status is announced, not operational.
50,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Tesla Cortex Phase 3United States
Facility status is announced, not operational.
100,000 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Tesla Dojo 1 Phase 1United States
Facility status is announced, not operational.
3,000 × Tesla D1 DojoHeld: Not present capacity
- together.ai 36k GB200sUnited States
Facility status is announced, not operational.
36,000 × NVIDIA GB200Held: Not present capacity
- TotalEnergies Pangea IIIFrance
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
3,384 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Universitaet Frankfurt Goethe-NHRGermany
AMD Instinct MI210 includes unsupported hardware.
880 × AMD Instinct MI210Held: No supported per-chip rate for this architecture
- University of Bristol Isambard-AIUnited Kingdom
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
5,448 × NVIDIA GH200Held: No reviewed dense FP16 rate
- University of Illinois NCSA DeltaUnited States
NVIDIA A40 PCIe includes unsupported hardware.
480 × NVIDIA A100312 TFLOP/s dense FP16 per accelerator
400 × NVIDIA A40 PCIeHeld: No supported per-chip rate for this architecture
- University of Tsukuba CygnusJapan
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
320 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- US Government Supercomputer 1United States
NVIDIA Tesla K40c includes unsupported hardware.
4,160 × NVIDIA Tesla K40cHeld: No supported per-chip rate for this architecture
- US Government Supercomputer 2United States
NVIDIA Tesla K40c includes unsupported hardware.
4,160 × NVIDIA Tesla K40cHeld: No supported per-chip rate for this architecture
- Voltage Park Location 5United States
Facility status is announced, not operational.
4,088 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Voltage Park Location 6United States
Facility status is announced, not operational.
4,088 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Voltage Park Texas Phase 2United States
Facility status is announced, not operational.
4,088 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Voltage Park UtahUnited States
Facility status is announced, not operational.
4,088 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Voltage Park VirginiaUnited States
Facility status is announced, not operational.
4,088 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Voltage Park WashingtonUnited States
Facility status is announced, not operational.
4,088 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- xAI Fulton GeorgiaUnited States
Facility status is announced, not operational.
12,112 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
336 × NVIDIA A100Held: Not present capacity
- XTX Markets ClusterCountry not recorded
NVIDIA V100 is not identified as the V100 SXM2, the only V100 the estimator supports.
10,000 × NVIDIA A100312 TFLOP/s dense FP16 per accelerator
10,000 × NVIDIA V100Held: Variant not stated; rate depends on itDocumented variants: V100 SXM2 125; V100 PCIe 112; V100S PCIe 130 TFLOP/s; none applied.
- Yotta Shakti Cloud D1India
Facility status is announced, not operational.
16,384 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- Yotta Shakti Cloud NM1 Phase 2India
Facility status is announced, not operational.
16,384 × NVIDIA H100 SXM5 80GBHeld: Not present capacity
- YTL AI JohorMalaysia
Facility status is announced, not operational.
15,428 × NVIDIA GB200Held: Not present capacity
- Google ArcolaUnited States
No positive whole-number GPU count is recorded.
- Google ColumbusUnited States
No positive whole-number GPU count is recorded.
- Google Council Bluffs (East)United States
No positive whole-number GPU count is recorded.
- Google LancasterUnited States
No positive whole-number GPU count is recorded.
- Google LincolnUnited States
No positive whole-number GPU count is recorded.
- Google MesaUnited States
No positive whole-number GPU count is recorded.
- Google MidlothianUnited States
No positive whole-number GPU count is recorded.
- Google New AlbanyUnited States
No positive whole-number GPU count is recorded.
- Google OmahaUnited States
No positive whole-number GPU count is recorded.
- Google PapillionUnited States
No positive whole-number GPU count is recorded.
- Google Pryor (North)United States
No positive whole-number GPU count is recorded.
- Google Red OakUnited States
No positive whole-number GPU count is recorded.
- Google Storey CountyUnited States
No positive whole-number GPU count is recorded.
- Google The DallesUnited States
No positive whole-number GPU count is recorded.
- Google Waltham CrossUnited Kingdom
No positive whole-number GPU count is recorded.
- Huawei HoringerChina
No positive whole-number GPU count is recorded.
- Meta GallatinUnited States
No positive whole-number GPU count is recorded.
- Meta JeffersonvilleUnited States
No positive whole-number GPU count is recorded.
- Meta KunaUnited States
No positive whole-number GPU count is recorded.
- Meta Los LunasUnited States
No positive whole-number GPU count is recorded.
- Meta MontgomeryUnited States
No positive whole-number GPU count is recorded.
- Meta PrometheusUnited States
No positive whole-number GPU count is recorded.
- Meta RosemountUnited States
No positive whole-number GPU count is recorded.
- Meta TempleUnited States
No positive whole-number GPU count is recorded.
- Microsoft Fairwater AtlantaUnited States
No positive whole-number GPU count is recorded.
- Microsoft Fairwater WisconsinUnited States
No positive whole-number GPU count is recorded.
- Microsoft GoodyearUnited States
No GPU inventory is recorded.
- Microsoft Project OsmiumUnited States
No positive whole-number GPU count is recorded.
- Microsoft SAT14United States
No positive whole-number GPU count is recorded.
- Microsoft SAT40United States
No positive whole-number GPU count is recorded.
- QTS Richmond 1United States
No positive whole-number GPU count is recorded.
- QTS Richmond 2United States
No positive whole-number GPU count is recorded.
- QTS Richmond 3United States
No positive whole-number GPU count is recorded.
- STACK Infrastructure NVA02United States
No positive whole-number GPU count is recorded.
- Start Campus Sines Data CampusPortugal
No positive whole-number GPU count is recorded.
- Stream PhoenixUnited States
No positive whole-number GPU count is recorded.
- Vantage TX1United States
No positive whole-number GPU count is recorded.
- VNET Bayin UlanqabChina
No positive whole-number GPU count is recorded.
- Colossus 2United States
No GPU inventory is recorded.
- Equinix LD5 (London)United Kingdom
No positive whole-number GPU count is recorded.
- Equinix SG3Singapore
No positive whole-number GPU count is recorded.
- JUPITER (EuroHPC)Germany
No GPU inventory is recorded.
- Karolina (EuroHPC)Czech Republic
No GPU inventory is recorded.
- Leonardo (EuroHPC)Italy
No GPU inventory is recorded.
- LUMI (EuroHPC)Finland
No GPU inventory is recorded.
- MareNostrum 5 (EuroHPC)Spain
No GPU inventory is recorded.
- MeluXina (EuroHPC)Luxembourg
No GPU inventory is recorded.
- Vega (EuroHPC)Slovenia
No GPU inventory is recorded.
- HammerHAIGermany
Facility status is under construction, not operational.
- MetaUnited States
No facility class is recorded.
- MicrosoftUnited States
No facility class is recorded.
- Khazna–Eni Ferrera Erbognone AI CampusItaly
Facility status is announced, not operational.
- Meta Richland Parish (Hyperion)United States
Facility status is under construction, not operational.
- Mimer AI Factory supercomputerSweden
Facility status is announced, not operational.
- Hut 8Canada
No facility class is recorded.
- PIAST-AI FactoryPoland
Facility status is under construction, not operational.
- Reliance–Meta AI Data Center (Jamnagar)India
Facility status is announced, not operational.
- Scaleway DC2 (PAR1)France
No positive whole-number GPU count is recorded.
- Scaleway DC3 (PAR1)France
No positive whole-number GPU count is recorded.
- Scaleway DC5 (PAR2)France
No positive whole-number GPU count is recorded.
- Expanse (SDSC)United States
NVIDIA H100 does not identify an H100 form factor.
136 × NVIDIA H100Held: Variant not stated; rate depends on itDocumented variants: H100 SXM5 989.5; H100 PCIe 756 TFLOP/s; none applied.SDSC Expanse user guide (undated; read 2026-09-11) and SDSC Annual Report 2024–2025
208 × NVIDIA V100 SXM2 32GB125 TFLOP/s dense FP16 per acceleratorSDSC Expanse user guide (undated; read 2026-09-11)
- Olivia (Sigma2)Norway
No dense FP16 rate for NVIDIA GH200 has passed evidence review.
448 × NVIDIA GH200Held: No reviewed dense FP16 rateSigma2 news 'Olivia expands: Massive GPU capacity upgrade' (2026-02-27)
- Stargate Abilene (OpenAI/Oracle/SoftBank)United States
No positive whole-number GPU count is recorded.
- Stargate UAE (OpenAI/G42/Oracle)United Arab Emirates
Facility status is under construction, not operational.
- Snellius (SURF National Supercomputer)Netherlands
No GPU inventory is recorded.
- Yotta Chennai Data Center Park (Hiranandani Parks)India
Facility status is under construction, not operational.
- Yotta D1India
No positive whole-number GPU count is recorded.
- Yotta D2India
Facility status is announced, not operational.
- Yotta G1 (GIFT City)India
No positive whole-number GPU count is recorded.
- Yotta NM1India
No positive whole-number GPU count is recorded.
- Giga Texas Data CenterUnited States
Facility status is announced, not operational.
- Port Jeff -Mt. Sinai Energy StorageUnited States
Facility status is decommissioned, not operational.
- Massena LoadUnited States
Facility status is decommissioned, not operational.
- North Country Data CenterUnited States
Facility status is announced, not operational.
- St Lawrence Data and Agricultural CenterUnited States
Facility status is announced, not operational.
- Lake Mariner Data IIUnited States
Facility status is announced, not operational.
- Niagara Digital CampusUnited States
Facility status is announced, not operational.
- Hudson Valley Data CenterUnited States
Facility status is decommissioned, not operational.
- Cayuga ComputeUnited States
Facility status is announced, not operational.
- Hudson Valley Data CenterUnited States
Facility status is announced, not operational.
- Proposed Datacenters at 450 Broadway, Buchanan, NY, 10511United States
Facility status is announced, not operational.
- Greenidge 200 MW Data Center ProjectUnited States
Facility status is announced, not operational.
- Data & Technology CampusUnited States
Facility status is announced, not operational.
- Arsenal Data Site 250United States
Facility status is announced, not operational.
- Arsenal Data Site 500United States
Facility status is announced, not operational.
- Arsenal Data Site 1000United States
Facility status is announced, not operational.
- New York State Artificial Intelligence Data CenterUnited States
Facility status is announced, not operational.
- Wulf Compute Data Center IIUnited States
Facility status is announced, not operational.
- Cayuga DataUnited States
Facility status is announced, not operational.
- 1 Gig Data Center East Fishkill, NYUnited States
Facility status is announced, not operational.
- North East Data LLC Data CenterUnited States
Facility status is announced, not operational.
- Pontoon Bridge Road Data CenterUnited States
Facility status is announced, not operational.
- Globe Digital Holdings - 1United States
Facility status is announced, not operational.
- Broome County Tech ParkUnited States
Facility status is announced, not operational.
Sites within a year of 10²⁵ FLOP, by country
⚠ marks a country listed in this catalogue for EU AI Act: Systemic Risk GPAI.
Estimates admit complete, reconciled operational inventories at 24-hour dense FP16 throughput, adjusted by the methodology's MFU and interconnect factors. A named power estimate separately requires a reviewed exact MW, scope, source, and as-of receipt before applying a PUE and GPU share of IT load; none is admitted today. Open a row for its inputs and parameter sensitivity. Methodology version 1.4.0.
Models above the EU compute line in the July snapshot
Article 52(6)
Scrutica's search log records no Article 52(6) list in the sources checked on 25 September 2026. The table sets that dated search beside every model in the pinned Epoch extract whose estimated training compute exceeds 10²⁵ FLOP.
| Model | Provider | Training compute (Epoch est.) | Epoch publication date | Publication cohort | Code of Practice | Public notification record |
|---|---|---|---|---|---|---|
| Grok 4 | xAI | 5 × 10²⁶Speculative | 9 Jul 2025 | Published before 2 Aug 2025 | Signatory (Safety and Security chapter only) | Not found in the enumerated sources as of 25 Sep 2026† |
| GPT-4.5 | OpenAI | 3.8 × 10²⁶Likely | 27 Feb 2025 | Published before 2 Aug 2025 | Signatory (full Code) | Not found in the enumerated sources as of 25 Sep 2026† |
| Grok 3 | xAI | 3.5 × 10²⁶Likely | 17 Feb 2025 | Published before 2 Aug 2025 | Signatory (Safety and Security chapter only) | Not found in the enumerated sources as of 25 Sep 2026† |
| GPT-5 | OpenAI | 6.6 × 10²⁵Speculative | 7 Aug 2025 | Published from 2 Aug 2025 | Signatory (full Code) | Not found in the enumerated sources as of 25 Sep 2026 |
| Llama 4 Behemoth (preview) | Meta AI | 5.184 × 10²⁵Likely | 5 Apr 2025 | Unreleased per Epoch‡ | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026 |
| Gemini 1.0 Ultra | Google DeepMind | 5 × 10²⁵Speculative | 6 Dec 2023 | Published before 2 Aug 2025 | Signatory, full Code (listed as Google) | Not found in the enumerated sources as of 25 Sep 2026† |
| Llama Nemotron Ultra 253B | NVIDIA | 3.911 × 10²⁵Likely | 18 Mar 2025 | Published before 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026† |
| Composer 2.5 | Cursor | 3.87 × 10²⁵Likely | 18 May 2026 | Published from 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026 |
| Llama 3.1-405B | Meta AI | 3.8 × 10²⁵Confident | 23 Jul 2024 | Published before 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026† |
| Claude 3.7 Sonnet | Anthropic | 3.35 × 10²⁵Likely | 24 Feb 2025 | Published before 2 Aug 2025 | Signatory (full Code) | Not found in the enumerated sources as of 25 Sep 2026† |
| Grok-2 | xAI | 2.96 × 10²⁵Confident | 13 Aug 2024 | Published before 2 Aug 2025 | Signatory (Safety and Security chapter only) | Not found in the enumerated sources as of 25 Sep 2026† |
| Claude 3.5 Sonnet | Anthropic | 2.7 × 10²⁵Speculative | 20 Jun 2024 | Published before 2 Aug 2025 | Signatory (full Code) | Not found in the enumerated sources as of 25 Sep 2026† |
| Doubao-pro | ByteDance | 2.505 × 10²⁵Likely | 28 Oct 2024 | Published before 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026† |
| Composer 2 | Cursor | 2.32 × 10²⁵Confident | 19 Mar 2026 | Published from 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026 |
| GPT-4 (Jun 2023) | OpenAI | 2.1 × 10²⁵Likely | 13 Jun 2023 | Published before 2 Aug 2025 | Signatory (full Code) | Not found in the enumerated sources as of 25 Sep 2026† |
| GPT-4 (Mar 2023) | OpenAI | 2.1 × 10²⁵Likely | 15 Mar 2023 | Published before 2 Aug 2025 | Signatory (full Code) | Not found in the enumerated sources as of 25 Sep 2026† |
| Nemotron-4 340B | NVIDIA | 1.8 × 10²⁵Confident | 14 Jun 2024 | Published before 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026† |
| Qwen3-Max | Alibaba | 1.512 × 10²⁵Speculative | 5 Sep 2025 | Published from 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026 |
| Pangu Ultra | Huawei | 1.069 × 10²⁵Confident | 10 Apr 2025 | Published before 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026† |
| Aramco Metabrain AI | Saudi Aramco | 1.05 × 10²⁵Likely | 4 Mar 2024 | Unreleased per Epoch‡ | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026 |
| Inflection-2 | Inflection AI | 1.001 × 10²⁵Confident | 22 Nov 2023 | Published before 2 Aug 2025 | Not on the signatory list | Not found in the enumerated sources as of 25 Sep 2026† |
† Article 111(3) gives providers of GPAI models placed on the market before 2 August 2025 until 2 August 2027 to comply. The cohort column uses Epoch's publication date as a proxy for that placement, which the date alone leaves open, as it does each provider's own notification deadline.
‡ Article 52(1) can run before release (“...or it becomes known that it will be met”); unreleased models above the line are retained for that reason, with release status shown as Epoch records it.
Article 113 sets 2 August 2026 for the Commission's GPAI enforcement provisions.
Method note: what Article 52 requires
Article 52(1) requires notification without delay and within two weeks after the Article 51(1)(a) condition is met or it becomes known that it will be met. Providers submit through EU SEND, and Article 78 protects specified confidential information, so a filing can exist without leaving any of the public traces this table is built from.
Article 52(6) requires the Commission to publish and update a list of systemic-risk models, with protections for intellectual property and confidential business information. The search log below contains eleven entries checked on 25 September 2026; it is a dated record of those checks.
Rows retain the compute estimates and per-row vintages in the Epoch extract accessed on 21 July 2026. Display removes floating-point noise and caps values at four significant figures; the export preserves the raw strings. Epoch's confidence label describes the most uncertain populated field among compute, parameters and dataset size. Its documentation expresses the categories as 90% confidence intervals.
These estimates have not been reconciled model by model to the Commission's training-compute accounting guidance. The guidelines' 30% allowance governs provider reporting and leaves the Epoch estimates without a stated interval.
Article 52 permits rebuttal of the compute presumption and Commission designation on other capability criteria, so membership of the official list turns on more than an estimate above the line.
Absence verification, 25 Sep 2026: no published Article 52(6) list located in the eleven enumerated sources. Per-source log.
The log records the sources checked on 25 September 2026 and has not been re-run since. Open-web searches on 2026-09-25 (31 queries, re-running the 2026-07-21 pass) for a Commission publication of the list, and for provider self-disclosures of Article 52(1) notifications, surfaced neither. The Commission's enforcement-start press release (IP/26/1714, 31 July 2026) publishes a list of signatories of the Code of Practice on AI-generated content and no list of models. Regulation (EU) 2026/1744 amends neither Article 51 nor Article 52. Implementing Regulation (EU) 2026/1755 could not be fetched, so what it says about the list is unverified.
| Source | Result | Page vintage | Method |
|---|---|---|---|
| Commission AI regulatory framework policy page | No register, no link to one | Last update 3 August 2026 | direct fetch |
| Guidelines for providers of general-purpose AI models | No register; describes notification duty only | Last update 28 April 2026 | direct fetch |
| Commission FAQ on GPAI provider obligations | No register; states the 2 August 2027 rule for pre-2 August 2025 models | Last update 11 November 2025 | direct fetch |
| AI Act Service Desk, Article 52 (Commission-operated) | Quotes the 52(6) duty verbatim; links no list | No page date; footer "Official version of 13 June 2024" (statute text vintage) | direct fetch |
| Commission factpage on GPAI obligations | No register; confirms the private notification mailbox (EU-AIOFFICEGPAI-SR-PROVIDERS@ec.europa.eu) | Last update 1 August 2025 | direct fetch |
| Commission news item on the GPAI Guidelines | No register | Published 18 July 2025, updated 31 July 2025 | direct fetch |
| GPAI Code of Practice policy page | No register, no link to one | Last update 25 June 2026 | direct fetch |
| Code of Practice contents + signatory list | Signatory list present; no model register | Last update 31 July 2026 | direct fetch |
| Commission Q&A on GPAI models in the AI Act | No register; references Article 111(3) special rules | Last update 9 September 2025 | direct fetch |
| European AI Office policy page | No register; classification work described, no published list | Last update 8 September 2026 | direct fetch |
| Site-scoped searches across Commission AI Office surfaces | Zero register hits; only obligation descriptions and templates | n/a | recorded search |
Sources
Regulation (EU) 2024/1689, Articles 51, 52, 78, 111(3), 113 (tier 1, a primary source) · eur-lex.europa.eu (CELEX 32024R1689)
AI Act Service Desk, Articles 51 and 52 pages (Commission-operated; Article 52(1) and 52(6) wording verified 25 September 2026) · ai-act-service-desk.ec.europa.eu/en/ai-act/article-52
European Commission, Guidelines for providers of general-purpose AI models (EU SEND platform and notification mechanics) · digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers
European Commission, FAQ on obligations for GPAI providers (the 2 August 2027 sentence; page updated 11 November 2025) · digital-strategy.ec.europa.eu/en/faqs/guidelines-obligations-general-purpose-ai-providers
European Commission, GPAI Code of Practice contents and signatory list (page updated 31 July 2026; checked 25 September 2026) · digital-strategy.ec.europa.eu/en/policies/contents-code-gpai
Epoch AI, “Data on AI Models” (tier 2, a research database; CC-BY; accessed 21 July 2026; per-row vintage in the export) · epoch.ai/data/ai-models
Evaluation conditions change the capability observed
What a FLOP count misses
In its May 2026 cyber study, AISI capped tasks at 2.5 million tokens and used a simple agent scaffold to keep comparisons stable over time; the institute says those choices suppress success rates relative to larger budgets and stronger scaffolds. Its July study varied inference budgets across several benchmarks and found newer models gaining more from the additional compute.
For a governance assessment the test conditions belong beside the score: task coverage, inference budget, available tools and scaffold. A failure under one configuration bounds that configuration alone.
Sources
AISI · “How fast is autonomous AI cyber capability advancing?” (13 May 2026): Study and methodology
AISI · “More compute, more capability: why AI agent evals need to account for test-time compute” (July 2026): Study and results
European Parliament & Council · Regulation (EU) 2024/1689, Article 51 (classification of general-purpose AI models with systemic risk): Consolidated regulation
Related research
Default parameters for facility estimates
Parameters
Admitted inventory estimates use the component hardware path at these default settings. Named power comparisons await separate documentary review of the power value and scope. The cards also carry the cost-path parameter, which only the Estimator uses; assumptions changed there leave these defaults in place.
Throughput multiplier for interconnect losses beyond those in the utilization rate. The default applies none, because the utilization default is an end-to-end rate reported from training runs. Lower it only for a cluster whose interconnect is weaker than that of the runs behind that rate, such as one built from the export-restricted H800 or A800.
Utilization multiplier applied to dense peak throughput. PaLM 540B reports 46.2% model FLOP utilization and 57.8% hardware FLOP utilization; the latter includes rematerialization, and both figures describe that run.
Total facility power divided by IT-equipment power. The default converts a reported facility power figure to an assumed IT load; a site's own disclosure, for the same facility scope and operating period, replaces it where one exists.
Assumed share of IT power allocated to accelerator chips. The remainder covers other IT equipment; cooling outside the IT boundary is handled through PUE. A facility’s rack composition and operating GPU count require separate evidence.
Assumed share of a project’s capital expenditure allocated to accelerator purchases. A project budget may also cover later phases, land, buildings and infrastructure, which this share does not separate.
The full derivation chain, with every formula, is in the methodology chapter.
The Estimator shows how one of these durations is produced: a daily FLOP budget back-solved from a disclosed chip count, a nameplate power figure or a construction budget, and how far the first two answers agree for a site disclosing both.