a16z measures AI spending for the first time: only 4.5% of Americans pay, top 1% spends $900

Andreessen Horowitz has for the first time calculated how much Americans actually spend on AI, and the results reveal a deep gap between how widespread the technology has become and how willing people are to pay for it. The share of paying users in the US nearly doubled year over year — from 2.3% to 4.5% — but that's still a tiny fraction compared to the nearly half of the population already using AI in some form.

Where the money is concentrated

Within this narrow circle, paying capacity is distributed extremely unevenly. The top 1% of paying users spends an average of $900 a month — more than the bottom half of all paying users combined. The typical paying user has held steady at around $25 a month for several months running. In other words, spending growth is happening only at the top: the top 1% increased its spending by 79% since the start of 2025.

Who are these people? Professionals and power users — designers, developers, content creators. They gravitate toward tools built for creation and automation: n8n, fal, Higgsfield, Figma, HeyGen. Only seven products appear simultaneously in the top 50 of all three a16z rankings (web traffic, mobile downloads, and real payments): ChatGPT, Claude, Perplexity, Canva, plus three others.

The model hierarchy stays intact

ChatGPT dominates across every metric: its web traffic is twice that of Gemini and six times that of Claude. On mobile, the gap is even wider. But the picture looks different in the paying-subscriber segment.

Over the past year, Claude went from not appearing at all in a16z's first web ranking (2023) to third place. In March 2026, it overtook Gemini in the number of paying American subscribers, but in June Google restructured its pricing tiers, and the two models are now running neck and neck. Anthropic monetizes better than its rivals: the share of Claude users subscribed to plans priced at $100 a month or more is significantly higher than for ChatGPT or Gemini.

That momentum broke over the summer. In July and August, Claude's session numbers dropped while ChatGPT accelerated on the back of new model releases. By August, ChatGPT had reclaimed the lead in net new subscribers.

Agents as a new market

Personal AI assistants were out of reach for mainstream users for a long time. Now startups like Instinct and Tomo have attracted hundreds of thousands of users. Big players have responded in kind: Meta launched Muse, OpenAI launched Dots, and xAI launched Grok Bot.

Instinct has shown that agents can themselves become a payment channel: 40% of users add a credit card within three weeks and then spend four-figure sums per month on flights and hotels through integrations with Expedia and similar services.

Muse racked up 250,000 daily active users in its first week and more than 5 million downloads within a month. Threads drew 15+ million downloads in three weeks at the peak of the platform's popularity. Amazon blocked Muse within two weeks, while Shopify, Instacart, and OpenTable provided official integrations.

Why only subscriptions work

Almost every leading AI product monetizes through subscriptions and usage fees. Advertising and transaction commissions make up a negligible share.

This is an inversion of the early internet. In the early days of the web, advertising was the primary revenue source for Meta and Google; users themselves were the product. The high cost of running AI models doesn't allow companies to absorb losses for long while waiting for a mass audience to materialize. Monetization has to happen immediately.

Alternatives are emerging. In August, OpenAI reported an annualized revenue run rate of $1 billion from advertising within ChatGPT. Personal agents could open up a path through affiliate commissions and purchase fees.

Geographically, the market is fragmenting. Yandex Browser with its Alice assistant dominates in Russia, while Deepseek is becoming a bridge between Western and Chinese users.

The bottom line: AI is ceasing to be a standalone product and is instead being woven into services that already exist. a16z's real-payment metrics are its attempt to capture this new reality.