Cognizant Digital Studio Amsterdam · 19 November 2026

Sovereign AI who holds the keys

A journey from the AI frontier to the infrastructure behind it. Europe notices a dependency, then builds. Explore the numbers and their sources.

Explore the story
ARAGORN MEULENDIJKS · FUTURE HISTORIANSOURCES CHECKED · 18 NOV 2026

Europe notices a dependency, then builds.

Aragorn Meulendijks · ItsMrMetaverse

01 Frontier update

The frontier, this week

A snapshot of the models, the benchmarks and the pace of progress.

174
ECI · joint record, two labs, 9 and 17 Nov 2026

Epoch on Fable 5.2: “above what the pre-Astra trend for reasoning models would have predicted” Epoch AI on X

Epoch Capabilities Index (ECI), top models, as of 18 Nov 2026 (current estimates, can be revised)
GPT-6.2 Astra · OpenAI · released 17 Nov174 · TIED #1
Claude Fable 5.2 · Anthropic · released 9 Nov174 · TIED #1
Gemini 4 Carbon169.8
GPT-6 Astra169.2
Claude Opus 5.5168.7
GPT-6.1 Sol168.6
GPT-6.1 Terra167.7
Claude Fable 5.1167.1
Gemini 4 Argon166.6
Muse Wonder · Meta161.7
Kimi K3.1 Pro · Moonshot · open weights161.4
DeepSeek V4.1 Flash 1017 · open weights159.4
Grok 4.6157.8
Qwen 3.8 Max157.6
GLM-5.3 · open weights157.1
Grok 4.7156.4
GPT-6 Luna156.4

Axis starts at 150. Both new models are tied for #1 of 241 tracked models (as of 18 Nov 2026). Epoch ECI

From about 122 to about 174 since March 2023
GPT-4: 121.6 ECI · released 2023-03GPT-4Claude 3.5 Sonnet: 130.0 ECI · released 2024-06o1: 141.2 ECI · released 2024-12o1o3: 147.1 ECI · released 2025-04GPT-5: 150.0 ECI · released 2025-08GPT-5Gemini 3 Pro: 153.0 ECI · released 2025-11GPT-5.2 Pro: 157.3 ECI · released 2025-12GPT-5.4 Pro: 160.3 ECI · released 2026-03GPT-5.5 Pro: 162.8 ECI · released 2026-04Claude Fable 5: 164.2 ECI · released 2026-06Claude Opus 5: 164.7 ECI · released 2026-07Claude Fable 5.1: 167.1 ECI · released 2026-09-01GPT-6 Astra: 169.2 ECI · released 2026-09-03GPT-6 AstraClaude Fable 5.2: 174 ECI · released 2026-11-09GPT-6.2 Astra: 174 ECI · released 2026-11-17Fable 5.2 / GPT-6.2 Astra · 174Mar 2023Nov 2026

New highs at release, our reading of Epoch data; selected milestones. Dashed line: our own straight-line fit to Epoch's frontier records (+16 points per year, in line with the pace Epoch reports for reasoning models). By our own estimate from Epoch data, the two new models sit about 2–3 points above that line. This is our arithmetic, not an Epoch figure. Scores carry about ±4 points of uncertainty and can be revised. Epoch ECIOur calculation

The benchmarks

FrontierMath Tier 4 (v2) · hardest research maths
GPT-6.2 Astra100%
Claude Fable 5.2100%
Gemini 4 Carbon95.1%
GPT-6 Astra (Sept 2026)97.6%
Claude Opus 5.595.0%

Three models now score 100%: GPT-6.1 Sol (Sept) and, new in November, Claude Fable 5.2 and GPT-6.2 Astra. GPT-5 (released Aug 2025) scores about 22% in Epoch's current Tier 4 v2 evaluation. Epoch benchmarks

ARC-AGI-3 · standard harness
GPT-6.2 Astra70.3%
Claude Fable 5.281.2%
Gemini 4 Carbon65.1%
GPT-6 Astra (Sept 2026)62.7%
Claude Opus 5.5no score yet

ARC Prize standard harness, each model at its best setting. On the provider-adapter harness most frontier models score 97–100%. ARC Prize

ARC-AGI-2
GPT-6.2 Astra95.8%
Claude Fable 5.293.8%
Gemini 4 Carbon94.2%
GPT-6 Astra (Sept 2026)95.0%
Claude Opus 5.5no score yet

The top models are within about 2 points of each other: this test is nearly maxed out. ARC Prize

GPQA Diamond · PhD-level science (Epoch AI's own runs)
GPT-6.2 Astra96.8%
Claude Fable 5.296.0%
Gemini 4 Carbon96.5%
GPT-6 Astra (Sept 2026)95.8%
Claude Opus 5.594.1%

The top three are statistically tied, close to the benchmark's ceiling. Epoch benchmarks

FrontierSWE V2 · very hard coding tasks
GPT-6.2 Astrano score yet
Claude Fable 5.269.9%
Gemini 4 Carbon61.2%
GPT-6 Astra (Sept 2026)65.5%
Claude Opus 5.562.3%

Error margins are large (about ±5–9 points). FrontierSWE

Remote Labor Index · real paid freelance projects finished to client standard (Scale AI / CAIS)
GPT-6.2 Astrano score yet
Claude Fable 5.241%
Gemini 4 Carbonno score yet
GPT-6 Astra (Sept 2026)21%
Claude Opus 5.521%

Best agent when the index launched (Oct 2025): 2.5%. GPT-5: 1.7%. Remote Labor Index

APEX-Agents · professional agent tasks passed first try (Mercor)
GPT-6.2 Astra72.2%
Claude Fable 5.280.1%
Gemini 4 Carbon84.2%
GPT-6 Astra (Sept 2026)64.7%
Claude Opus 5.573.5%

Margins are about ±4–5 points, so neighbouring models are statistically close. Mercor

≈4 monthsdoubling time since 2023 · 129 days (range 104–158) METR horizons
How long a task an AI agent can finish on its own (METR, 50% success)
0.19 hClaude 3.5 Sonnet · Jun 2024
2.0 ho3 · Apr 2025
3.4 hGPT-5 · Aug 2025
3.7 hGemini 3 Pro · Nov 2025
4.9 hClaude Opus 4.5 · Nov 2025
5.9 hGPT-5.2 · Dec 2025
6.4 hGemini 3.1 Pro · Feb 2026
≈12 hClaude Opus 4.6 · Feb 2026
5.7 hGPT-5.4 · Mar 2026
16+ hClaude Mythos Preview (early version) · published May 2026

METR: an early version of Claude Mythos Preview handles tasks that take a skilled human at least 16 hours, at 50% success; METR warns results above 16 hours are unreliable (raw point estimate 17.4 h, 95% CI 8.5–55 h). Opus 4.6 has wide uncertainty (5 to 60 h). Latest METR data: May 2026; METR stopped updating the page on 8 Sep 2026. No METR horizon exists for Fable 5.2 or GPT-6.2 Astra. METR's Frontier Risk Report (May 2026, data from Feb–Mar 2026): the most capable model labs shared with METR had a 50% point estimate of 16–20 hours, past the roughly 16 hours its task suite can reliably measure. METR horizonsMETR risk

AI is now speeding up AI

1.9×

Anthropic's models now improve about twice as fast as before: 1.9× on benchmarks since Oct 2025 (95% CI 1.5–2.8×), per independent evaluator METR, based on Anthropic's internal data. METR guesses that most of this acceleration, perhaps around 80%, comes from Anthropic's own use of AI. METR acceleration

≈ +15≈ +29points per year on Anthropic's internal index (AECI), Anthropic models only, not Epoch's ECI
10×

Ajeya Cotra estimates Anthropic's researchers may now get roughly a 10× boost from AI. She compares it to living ten days for every day outside, and calls the estimate very uncertain.

Ajeya Cotra: “it looks like the 'dramatic acceleration' threshold has already been crossed” (6 Nov 2026). Ajeya Cotra

2027

the year a superhuman AI researcher will likely arrive, per the AI Futures Project's forecast (10 Nov 2026), assuming no deliberate slowdown. AI Futures Project

13×

the cost of reaching a fixed level of AI performance has fallen about 13× a year since 2023 (about 47% per quarter, average across five benchmarks).

Same score on a PhD-level science exam (GPQA Diamond): from about 30 cents a question to about four hundredths of a cent. About 725× cheaper in under 18 months (Epoch AI estimate). Epoch price

02 Sovereign AI

Europe notices a dependency

History shows how access to infrastructure can become a strategic question.

SIGNAL · 01

A signal, and a European answer

May 2000: the US switched off Selective Availability, the deliberate degradation of civilian GPS. In 1999 ESA opened its Galileo Directorate. On 15 December 2016 Galileo initial services went live, about 17 years later. GPS.govESA

SIGNAL · 02

The network is subject to law

March 2012: under EU sanctions, SWIFT cut off EU-sanctioned Iranian banks. March 2022: EU sanctions cut seven major Russian banks, including VTB, off from SWIFT. SWIFT sanctionsEU Council

03 Sovereign AI

Compute and models

The frontier is global. Its inputs and outputs are not evenly distributed.

SIGNAL · 03

Where new AI compute landed

Of the new AI compute that arrived worldwide from 2022 to June 2026, the US received about 72%, China about 10%, and the EU about 5%. Epoch AI estimates about 27 million H100-equivalents arrived worldwide. Estimates outside the US are less certain. Epoch AI imports

SIGNAL · 04

Notable models by origin

Stanford’s AI Index counts 59 notable AI models from the US, 35 from China and 2 from Europe, including the UK, in 2025. Stanford AI Index

SIGNAL · 05

Open weights follow the frontier

As of May 2026, Epoch AI found that since January 2026 the best open-weight models trailed the closed frontier by about 4 months on average, or 8 ECI points. The earlier average was about 3 months from January 2023 to October 2025. Epoch open weights

04 Sovereign AI

The cloud layer

European providers grew, while the market grew faster.

SIGNAL · 06

A smaller share of a larger market

European providers’ share of Europe’s cloud infrastructure market fell from 29% in 2017 to 15% in 2022 and held around 15% through mid-2025. Amazon, Microsoft and Google held 70% in 2025. Europe includes the UK. Synergy Research

05 Sovereign AI

Finance at the frontier

Regulated institutions face both the opportunity and the operational responsibility.

SIGNAL · 07

Generative AI is already in banking

About 85% of the banks the ECB supervises directly already use generative AI, according to ECB supervisory data from September 2026. In July 2026 the ECB told every bank it directly supervises to submit an action plan on frontier-AI risks by 31 October 2026. The ECB is assessing those plans. On 16 November 2026, Google Threat Intelligence Group reported that attackers use agentic frameworks to automate parts of live intrusions. ECBECB letterGoogle Threat Intelligence

SIGNAL · 08

Critical providers enter the frame

In November 2025 the EU’s financial supervisors named 19 technology providers as critical under DORA. This was the first list. DORA list

SIGNAL · 09

The high-risk timetable has changed

The EU Digital Omnibus on AI has been in force since 27 July 2026. It delays AI Act high-risk obligations to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems. EU regulation

06 Sovereign AI

Then Europe builds

From policy to infrastructure, the response is taking shape.

SIGNAL · 10

The gigafactory ambition

July 2026: the EU opened its AI Gigafactories call. Up to seven sites, up to €10B in public money, expected to unlock at least €20B in private investment. More than €30B in total. European Commission

SIGNAL · 11

A capacity proposal

The EU Cloud and AI Development Act, a Commission proposal of 3 June 2026, aims to triple EU data-centre capacity within 5–7 years. CADA proposal

SIGNAL · 12

A Dutch AI factory

Over €200 million for a Dutch AI factory in Groningen: €130 million from the national government and the region, plus about €70 million from the EU, granted October 2025. Dutch governmentEU funding granted

SIGNAL · 13

A vital piece of the supply chain

ASML is the only company in the world that makes EUV lithography machines, according to its annual report filed in February 2026. Annual report

SIGNAL · 14

Output per person

One AI lab reports over $6.6M revenue per employee for the trailing twelve months to September 2026. For IT services, about $63K is our calculation from published figures, annualising a quarter. Revenue is not profit, and the time bases differ. S-1/AOur calculation

Sources

Trace every number

Explore the source behind each figure and chart.

More numbers from the talk

Reliability: HIGH = primary source or official statistics; MEDIUM = analyst, research firm or secondary source; LOW = speaker perspective.

SourcePublisherDateReliability
Epoch Capabilities Index (ECI) dataFrontier AIEpoch AIUpdated 18 Nov 2026HIGH
ECI record post on Claude Fable 5.2Frontier AIEpoch AI9 Nov 2026HIGH
Benchmark data: FrontierMath and GPQA DiamondFrontier AIEpoch AINov 2026HIGH
FrontierSWE leaderboard (via Epoch AI)Frontier AIFrontierSWENov 2026HIGH
APEX-Agents leaderboardFrontier AIMercorNov 2026HIGH
Remote Labor Index leaderboardFrontier AIScale AI and CAISNov 2026HIGH
How good are Fable 5.2 and GPT-6.2 Astra at real work?Frontier AIEpoch AI17 Nov 2026HIGH
The plunging price of thoughtFrontier AIEpoch AI22 Sep 2026HIGH
ARC-AGI leaderboardFrontier AIARC Prize FoundationFetched 18 Nov 2026HIGH
Time horizons (benchmark_results_1_1)Frontier AIMETRdata to May 2026; page frozen 8 Sep 2026HIGH
Frontier Risk ReportFrontier AIMETR19 May 2026HIGH
Independent assessment of AI R&D acceleration at AnthropicAccelerationMETR5 Nov 2026HIGH
AI is dramatically accelerating AI research right nowAccelerationAjeya Cotra, Planned Obsolescence6 Nov 2026HIGH
Looks like the intelligence explosion will happen in 2027 after allAccelerationAI Futures Project10 Nov 2026HIGH
Our own calculation from Epoch AI data (not an Epoch figure)Frontier AIAragorn Meulendijks / ItsMrMetaverse18 Nov 2026MEDIUM
Selective AvailabilityHistoryGPS.gov2000HIGH
The 30-year journey of European satellite navigationHistoryESA2025HIGH
SWIFT and sanctionsHistorySWIFT2022HIGH
EU sanctions on Russian banksHistoryCouncil of the EU2 Mar 2022HIGH
AI Chip Imports documentationComputeEpoch AIOct 2026HIGH
AI Index 2026, research and development, figure 1.1.2ModelsStanford HAI2026HIGH
The open-closed ECI gapModelsEpoch AI29 May 2026HIGH
European cloud providers local market shareCloudSynergy Research Group24 Jul 2025MEDIUM
Supervisory data on AI use at directly supervised banksFinanceECB22 Sep 2026HIGH
AI-enabled cybersecurity threatsFinanceECB7 Jul 2026HIGH
EU supervisors designate critical ICT third-party providersFinanceMorgan Lewis, citing EU supervisorsNov 2025MEDIUM
Agentic threat activity crosses a significant thresholdFinanceGoogle Threat Intelligence Group and Mandiant16 Nov 2026HIGH
Digital Omnibus on AI, Regulation 2026/1744FinanceEUR-Lex2026HIGH
EU AI Gigafactories callEurope buildsEuropean Commission30 Jul 2026HIGH
Cloud and AI Development Act proposal COM(2026) 502Europe buildsEuropean Commission3 Jun 2026HIGH
AI factory in GroningenEurope buildsRijksoverheidJun 2025HIGH
EU financing for the AI factory in GroningenEurope buildsRijksoverheid13 Oct 2025HIGH
EUV lithography systems, annual filingEurope buildsASMLFeb 2026HIGH
AI lab revenue per employee, trailing twelve monthsOutput per personSEC filing17 Nov 2026HIGH
IT services revenue per employee: our calculation from a large listed IT-services firm's published Q3 2026 results (annualised revenue / staff)Output per personOur calculationOct 2026MEDIUM
Benchmark data archiveTalk numbersEpoch AINov 2026HIGH
Mensa Norway puzzle test resultsTalk numbersTrackingAISep 2026MEDIUM
ARC Prize leaderboardTalk numbersARC Prize FoundationNov 2026HIGH
Finance professionals and AI readinessTalk numbersAICPA & CIMADec 2025HIGH
AI-related publications in 2025Talk numbersNCBI PubMedOct 2026HIGH
AI forecasting: one year inTalk numbersJacob SteinhardtJul 2022MEDIUM