AI 2027 Tracker

Predictions Confirmed

16/31

53 total tracked

Speed Ratio

0.70×

~70% of predicted pace

Overall Assessment

Mixed

Directionally plausible, uneven pace

Scenario Speed Ratio

Reality is progressing at ~70% of AI 2027's predicted pace

0× 0.70× Current 1.0×

Status Breakdown

16 Confirmed

3 Ahead

8 On Track

4 Behind

13 Emerging

9 Not Yet Testable

Visual Overview

Status Distribution

Category Coverage

By Category

Emerging Frontier model trained at 10²⁷ FLOP (Agent-0, completes May 2025) May 2025 Not Yet Testable Frontier model uses daily weight updates / online learning January 2027 Confirmed Models shift to continuous/iterative training Late 2025 Emerging IDA achieves superhuman performance at coding Early 2027 Emerging High-bandwidth non-text reasoning (neuralese) deployed Early 2027 Confirmed OSWorld benchmark reaches 65% by mid-2025 Mid 2025 (65%), Early 2026 (80%) Not Yet Testable 10²⁸ FLOP training run completed March 2027

On Track Frontier model variant 10x cheaper released Late 2026 Emerging Global AI power consumption reaches 38GW End of 2026 Emerging AI consumes 2.5% of US electricity End of 2026 On Track Global AI capex reaches $1 trillion cumulative Late 2026 Confirmed Massive datacenter buildouts continue Through 2026 Confirmed Leading AI company reaches $45B annual revenue End of 2026 Behind Leading AI company reaches $3T valuation End of 2026 Confirmed Massive AI infrastructure investment continues Ongoing through 2025–2026 Ahead Stock market impact and public backlash from AI job displacement Late 2026 On Track Leading AI company reaches $40B compute costs, 6GW power, $200B capex End of 2026 Behind Stock market rises 30% in 2026 End of 2026

Confirmed Best AI agents cost hundreds of dollars per month Mid 2025 Emerging AI model capable of autonomous self-replication January 2027 Confirmed Agents struggle with long-horizon tasks 2025 Ahead METR time horizon doubles every 4 months ~4 month doubling from 2024+ Confirmed Computer-using agents marketed as 'personal assistants' Mid 2025 Confirmed Unreliable but useful AI agents emerge Mid 2025

Confirmed AI companies focus on AI-for-AI-research 2025–2026 Emerging AI R&D progress multiplier reaches 2x Late 2026 Not Yet Testable AI R&D progress multiplier reaches 3× January 2027 On Track AI R&D progress multiplier reaches 1.5× Early 2026 Not Yet Testable AI algorithmic progress multiplier reaches 4× (~2× overall R&D) March 2027

Emerging Large-scale anti-AI protest (10,000+ people) Late 2026 On Track Public unaware of best AI capabilities Ongoing through 2027 Confirmed Department of Defense scales up AI lab contracting Late 2026 Emerging Nationalization of leading AI lab debated February 2027 Confirmed Continued skepticism from academics and journalists Through 2025–2026

On Track AI provides substantial bioweapon design help Late 2025 Not Yet Testable China's centralized AI facility airgapped Late 2026–Early 2027 Confirmed AI scores 85% on Cybench Early 2026 Ahead AI reaches near-best-human hacking capability Early 2027 Not Yet Testable AI model weights stolen by nation-state February 2027 Emerging Leading AI lab progresses toward RAND SL3 security Mid-to-late 2026 (WSL 3.0); SSL 3.0 not until ~Apr 2027

Emerging China nationalizes/centralizes AI research Mid 2026 Emerging China has ~12% of global AI-relevant compute Mid 2026 Behind Leading Chinese AI lab ~6 months behind US frontier Mid 2026 On Track Chinese domestic AI chips 3 years behind US-Taiwan Mid 2026 Confirmed Export controls impact Chinese AI compute Ongoing Emerging Military tension around Taiwan escalates over AI race Mid 2026–Early 2027 Confirmed Gap between top US labs narrows to 0-2 months Late 2025

Confirmed Coding agents provide significant real-world value Mid 2025 Not Yet Testable Coding fully automated but research taste still requires humans Early–Mid 2027 Not Yet Testable 200,000–250,000 AI coding agents run in parallel March 2027 Behind RE-Bench score reaches 1.3 Early 2026 Not Yet Testable Superhuman coder emerges March 2027 On Track SWE-bench-Verified score reaches 85% Mid 2025

Recently Updated

Updated 2026-06-22 End of 2026

GLOBAL AI POWER 38GW PEAK POWER

Updated 2026-06-22 Mid 2026

They are about six months behind the best OpenBrain models.

Updated 2026-06-22 Late 2026

While the latest Agent-1 could double the pace of OpenBrain's algorithmic progress, Agent-2 can now triple it, and will improve further with time.

Updated 2026-06-22 Early 2026

AI systems contribute enough to AI research that the effective R&D multiplier reaches 1.5× — AI makes AI research 50% faster.

Updated 2026-06-22 Late 2025

We imagine the others to be 3-9 months behind OpenBrain (page 4, Late 2025). By Early 2026, several competing AIs match or exceed Agent-0 (page 7). The 3-9 month gap is the Late 2025 state; near-parity emerges by Early 2026.

Current Assessment

May 2026 Summary

The tracker shows a mixed but still directionally relevant picture. Several qualitative predictions around agents, coding tools, infrastructure investment, lab competition, and institutional response have materialized, while important quantitative milestones remain uneven.

The most interesting divergence is still between capability and scale: METR-style agent horizons and some deployment signals have moved quickly, while frontier training-compute milestones, valuation targets, and parts of the R&D feedback-loop thesis remain harder to verify or behind the original pace.

Bottom line: AI 2027 remains worth tracking closely, but the public evidence supports a sober read: real directional signal, uneven timing, and substantial uncertainty around the most consequential 2026–2027 claims.

Related Resources

Read AI 2027

The original scenario by AI Futures Project

About This Tracker

What we do and why

Methodology

How we score predictions

All Predictions

Browse all 53 tracked predictions

Which Predictions Came True?

Scorecard of confirmed and ahead predictions

AI 2027 vs Reality

The big picture comparison

vs AI Futures Self-Grading

Authors grade their own predictions

vs Metaculus Crowd

Crowd wisdom meets scenario forecasts