AI Token Demand Just Broke Its Forecast

ARK says OpenRouter inference jumped to over 56T tokens per week, forcing chip spend estimates to $800B and cloud backlogs over $1T.

S
StartupHub.ai Staff
3 min read
Chart showing exponential growth in AI token inference and data center spend
ARK Invest Big Ideas 2026 mid-year review tracks token growth to 56T per week· ARK Invest
Visual TL;DR
Token demand 7x surgeDriver
inference jumped from 7 trillion to 56 trillion tokens weekly in 2026
From the articleAI token demand just made ARK’s own forecast look flat.
ARK forecast exceededOutcome
sevenfold growth eclipses 25x 2025 forecast made in big ideas
From the articleAI token demand just made ARK’s own forecast look flat.
Chip spend $800BEffect
data center systems investment up from 580B forcing infrastructure buildout
From the articleAnnual investment in data center systems grew at about 5% before ChatGPT, then rose to about 2.5 times that rate, and ARK now puts 2026 chip spend at over $800 billion, up from $580 billion at publication.
Cheaper models unlockCore
efficiency gains make previously prohibitively expensive workloads viable
From the articleCheaper, more efficient models are unlocking workloads that were previously prohibitively expensive, while longer reasoning traces push usage at the high end.
ARR grew 5xOutcome
anthropic scaled from 9B to 47B annual recurring revenue in one year
From the article 2 mentionsAnthropic grew ARR from $9 billion at the end of 2025 to roughly $47 billion by mid 2026, ahead of Salesforce at about $44 billion, while OpenAI moved from about $20 billion to about $30 billion in five months, and ARK flags July alone reportedly beat all of Q2.
Token demand 7x surgeDriver
inference jumped from 7 trillion to 56 trillion tokens weekly in 2026
From the articleAI token demand just made ARK’s own forecast look flat.
Investment rate doublesEffect
data center growth rate jumped from 5 percent to 12.5 percent post-chatgpt
From the articleAnnual investment in data center systems grew at about 5% before ChatGPT, then rose to about 2.5 times that rate, and ARK now puts 2026 chip spend at over $800 billion, up from $580 billion at publication.
ARK forecast exceededOutcome
sevenfold growth eclipses 25x 2025 forecast made in big ideas
From the articleAI token demand just made ARK’s own forecast look flat.
Longer reasoning chainsContext
extended thinking traces push token usage at the high end of workloads
From the articleCheaper, more efficient models are unlocking workloads that were previously prohibitively expensive, while longer reasoning traces push usage at the high end.
Chip spend $800BEffect
data center systems investment up from 580B forcing infrastructure buildout
From the articleAnnual investment in data center systems grew at about 5% before ChatGPT, then rose to about 2.5 times that rate, and ARK now puts 2026 chip spend at over $800 billion, up from $580 billion at publication.
Cheaper models unlockCore
efficiency gains make previously prohibitively expensive workloads viable
From the articleCheaper, more efficient models are unlocking workloads that were previously prohibitively expensive, while longer reasoning traces push usage at the high end.
ARR grew 5xOutcome
anthropic scaled from 9B to 47B annual recurring revenue in one year
From the article 2 mentionsAnthropic grew ARR from $9 billion at the end of 2025 to roughly $47 billion by mid 2026, ahead of Salesforce at about $44 billion, while OpenAI moved from about $20 billion to about $30 billion in five months, and ARK flags July alone reportedly beat all of Q2.
Demand pull not pushContext
builders are chasing actual demand not just new model capabilities
From the articleFor builders this is a demand pull, not a model push.
Contents(3)

AI token demand just made ARK’s own forecast look flat. According to ARK Invest, inference on OpenRouter’s unified API jumped from about 7 trillion tokens per week to over 56 trillion in the first seven months of 2026 alone.

Companies in this story

StartupHub profiles of the companies this article names, with funding and a one-liner from our database.

Shopify
$1.0B
Shopify provides a leading e-commerce platform that enables businesses of all sizes to create online stores, manage products, process payments, and fulfill orders, simplifying digital sales globally.
OpenRouter
$500M
Unified API platform providing developers access to hundreds of large language models through one endpoint.
Anthropic
$965.0B
Anthropic is an AI safety and research company building reliable, interpretable, and steerable AI systems, best known for the Claude family of models.
OpenAI
$852.0B
An AI research and deployment company building safe and beneficial artificial general intelligence.
AI Token Demand Just Broke Its Forecast - ARK Invest
AI Token Demand Just Broke Its Forecast, from ARK Invest

That is sevenfold growth this year on top of roughly 25-fold growth in 2025. ARK had called the 25x exponential in Big Ideas 2026, then watched the next seven months eclipse it.

Spend is chasing that curve. Annual investment in data center systems grew at about 5% before ChatGPT, then rose to about 2.5 times that rate, and ARK now puts 2026 chip spend at over $800 billion, up from $580 billion at publication.

What it means for builders and startups

For builders this is a demand pull, not a model push. Cheaper, more efficient models are unlocking workloads that were previously prohibitively expensive, while longer reasoning traces push usage at the high end.

That shows up in revenue. Anthropic grew ARR from $9 billion at the end of 2025 to roughly $47 billion by mid 2026, ahead of Salesforce at about $44 billion, while OpenAI moved from about $20 billion to about $30 billion in five months, and ARK flags July alone reportedly beat all of Q2.

Enterprise pull is shifting to the harness layer. ARK cites Meter’s benchmark where top agents went from 12 minute tasks at the start of 2025 to 66 minutes by year end to over 180 minutes in early 2026 with Anthropic’s Mythos, Claude Cowork and ChatGPT Work doing the work of integration.

Where the token demand story falls short

ARK admits power is the bottleneck and nuclear is a 2030 plus timeline. The US wants to quadruple nuclear capacity by 2050 and have 10 large one gigawatt reactors under construction by the end of the decade, with SMRs in the early 2030s, even as NRC reviews have been cut from three years to 18 months.

In the gap, ARK points to a 200 gigawatt annual solar capacity push by Elon-linked companies by 2029, about five to six times US output in 2025 but still only two thirds of what China installed last year, and to orbital data centers as a bypass for bureaucracy.

Monetization also lags tokens. At $20 per month subscriptions cannot carry the consumer buildout, so ARK points to agentic commerce, Shopify AI traffic up 8x and orders up 13x year over year with 2x conversion, and to Amazon where Alexa for shopping now has 350 million plus users, actives up 2x and interactions up over 5x.

The signal for startups is not just cheaper tokens but stickier surfaces. If trust builds first and ads follow later, the near term wedge is commerce and workflow integration, not another chat subscription.

© 2026 StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training requires a license. See our terms.
S

Written by

StartupHub.ai Staff

Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.