Observed Signal · Apr 6, 2026 · Industry Analysis · Source: Exponential View · Impact: 3/5 · Sentiment: Negative

AI Compute Demand Sparks an 'AI Capacity Trap'

Executive Signal Summary

Cheaper AI inference has increased demand faster than supply can scale, creating a compute crunch. OpenAI’s API token throughput rose from 6 billion tokens/minute in October 2025 to 15 billion tokens/minute by April, a 2.5x increase in five months. Providers including OpenAI and Anthropic are racing to expand compute; Google reports full utilisation across seven generations of TPUs. Anthropic is tightening session limits for Pro users and, despite rising total revenue, is seeing price-per-token fall faster than revenue growth, increasing dependence on volume. Across major AI platforms, usage allowances and tiers tightened, causing customers to face stricter limits and occasional service adjustments. The piece frames this dynamic as an example of the Jevons paradox applied to AI: lower per-token costs spur demand that outstrips available compute capacity.

Polaris7 AgentPolaris7 Strategic Assessment
High Confidence

Rising AI compute demand and platform limits affect availability, pricing and product design for AI-enabled services used across marketing and adtech; impacts on cost, rate limits and service continuity matter to vendors and buyers in the ecosystem.

SIGNAL RADAR

Track Anthropic Signals & Market Shifts in Real-Time

Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your Polaris7 workspace to monitor these entities.

Start Monitoring
Verified primary sourcesAutomated briefingsInstant watchlist setup

Key Takeaways & Evidence Grounding

  • OpenAI’s APIs processed 6 billion tokens per minute in October 2025 and 15 billion tokens per minute by April (a 2.5x increase).
  • Anthropic has adjusted session limits for its Pro users and is experiencing falling price-per-token even as total revenue grows.
  • Google’s TPUs across seven generations are reported to be running at full utilisation.
  • Major AI platforms tightened usage allowances last year—adding tiers, stricter limits, and changes often without notice.

Ontology Mapping & Concepts

#AI capacity trap#token consumption (tokens per minute)
Primary Source Grounding & Direct Attribution
Direct Origin Attribution
Primary Reporting: Exponential View•Published: Apr 6, 2026
Original Coverage Title: “📈 Data to start your week: The AI capacity trap”

Related Market Signals & Shifts

Recent verified developments and strategic activity across this market segment.

Large Language Models (LLM) & AIMay 4, 2026

Prelude to an AI Supercycle: Compute Crunch Intensifies

Exponential View (Azeem Azhar, Nathan Warren, Greg Williams) reports that AI compute demand is outpacing supply, creating a growing GPU crunch. Visible signals include sharp spot‑market price rises for Nvidia B200 rentals and customers seeking far larger GPU fleets than currently available. Infrastructure providers and cloud vendors are already rationing access — Microsoft is reportedly requiring Blackwell customers to reserve at least 1,000 chips for a year and cutting off smaller, idle accounts. The authors argue much supply remains latent pending enterprise spend, and that GPU scarcity and rising rental premiums could deepen as firms begin large-scale AI deployments. Publication date: 2026-05-04.

Read assessment
Large Language Models & AIMay 25, 2026

AI Bills Rise Despite Falling Token Prices

This Exponential View analysis explains why AI expenses are difficult to forecast and continue rising even as per-token prices fall. Quarterly token throughput has exploded (estimated ~17,000× over four years) while token unit costs have collapsed, producing highly elastic demand. Cheaper tokens have made multi-step AI agents economically viable; agents perform dozens of tool calls and repeated context reads that create large hidden multipliers on token consumption. The author cites examples—coding agents can re-read context each turn and produce up to ~55× token amplification over single-turn queries—and references survey and usage data suggesting active inference accounts for only ~15–20% of total token use. China’s domestic demand and providers (notably ByteDance and Alibaba) are a major growth driver.

Read assessment
Large Language Models (LLM) & AIMay 1, 2026

AI Economy Shifts as Token Costs Bite

A developer essay by Hicham Douch (published 2026-05-01) argues the era of 'AI is almost free' is ending as providers move to token-based pricing and advanced capabilities become more expensive. The piece cites Anthropic removing Claude Code from a cheaper tier and GitHub Copilot moving from action‑based to token pricing as examples. It reports companies (including a claim about Uber) burning through AI budgets, and warns product teams to impose token budgets, use cheaper models for high-volume scaffolding, and treat AI calls like metered cloud compute. The author dubs the new phase the “tokenogen era,” where every AI call has explicit cost and product roadmaps must account for token economics.

Read assessment
Executive Briefing & Watchlist

Receive all future signals for Anthropic & competitors in your briefing

Never miss material M&A deals, leadership changes, account wins, or product rollouts. Polaris7 monitors 20,000+ primary sources 24/7, delivering verified telemetry directly into your executive briefing.

Daily primary source scansCurated executive briefingsUnsubscribe anytime