After years of free tiers, flat-rate enterprise contracts, and VC absorbing the difference, the industry is renegotiating its economics. Customers will feel it
Photo by Wally Skalij/Getty Images
Big Tech told investors this week it will spend roughly $700 billion on AI infrastructure in 2026, likely the largest corporate buildout in history. The market mostly celebrated.
Investors may be cheering the shift underneath, not just the spending. After years of free tiers, flat-rate enterprise contracts, and venture capital absorbing the difference, the industry is renegotiating its unit economics. Customers are about to feel it.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
Microsoft $MSFT's GitHub Copilot will move on June 1 from a flat-rate coding-assistant model toward usage-based billing tied to tokens, which are the units of data (usually a few characters of text) processed by AI. The change ends what was effectively all-you-can-eat AI-assisted coding. Anthropic shifted enterprise customers to per-seat plans with usage billed on top, and last month cut off third-party tools that were running $200-a-month consumer subscriptions through autonomous AI agents around the clock, consuming thousands of dollars in compute.
OpenAI's head of ChatGPT has said publicly that unlimited AI plans probably no longer make sense. The company is charging more for its newest model but hasn't restructured enterprise pricing the way Anthropic has.
Token consumption has become the AI industry's favorite demand metric, with developers racing to figure out tokenmaxxing. Meta $META and Shopify $SHOP built internal leaderboards to track employee usage. Nvidia $NVDA CEO Jensen Huang has said he would be alarmed if a $500,000 engineer was not burning $250,000 worth of tokens.
And the fight is increasingly for developers. Anthropic’s Claude Code has become one of the company’s most important growth engines. The company said Claude Code had surpassed a $2.5 billion revenue run rate in February and had more than doubled since the start of 2026; by April, Anthropic overall said its run-rate revenue had topped $30 billion, up from about $9 billion at the end of 2025. OpenAI pledged to mirror that focus and launched its rival Codex tool, recently adding a $100 monthly subscription tier in an effort to lure back power users who had decamped to Claude Code.
The attraction is obvious. Developers burn tokens, run agentic workflows that burn even more, and become hard to dislodge once AI is wired into their daily work. For labs preparing for IPOs as soon as this year and trying to turn usage into durable revenue, coders are not just early adopters. They are the wedge.
That lock-in is more real than most buyers expected. A Zapier survey of more than 500 executives found 90% believed they could move between AI vendors within a few weeks. Only 42% of the ones that actually tried said it went smoothly. APIs, training data, custom tooling, and institutional workflow knowledge do not migrate cleanly.
Other companies have different answers to the changing landscape. Meta launched Muse Spark last month, the first major model from its Superintelligence Lab and a sharp break from the open-source Llama line that built the company's AI brand. Muse Spark is closed-source, and Meta has said it plans to charge for developer access, an inversion of the Llama strategy. Investors aren't sold, and the stock has slipped.
Apple $AAPL, reporting Thursday, went the other direction. Apple is taking a lighter-infrastructure path, treating AI more as an integrated product feature than a standalone infrastructure race, while relying in part on partnerships for heavier model work. Apple's capex is roughly a tenth of Meta's and the stock has been climbing back after a rough 2025.
Even the other hyperscalers are making slightly different bets, leaning into what each does best. Google $GOOGL posted 63% cloud growth and is selling its TPUs to "select" customers. Amazon $AMZN Web Services grew 28% and disclosed a $364 billion contracted backlog. Microsoft's Azure and AI services grew about 40%.
A striking share of the cloud boom is now riding on two startups. Anthropic is one of those select customers, and Amazon announced a $100 billion deal with the company. About 45% of Microsoft's promised future cloud revenue comes from OpenAI.
Wall Street is betting the handoff works. April was the S&P 500’s best month since the pandemic, and the rally again leaned on the Magnificent Seven, which were responsible for more than 40% of the index’s total return last year. Their AI spending plans are no longer just company strategy. They are one of the market’s main engines.
The next phase is less forgiving. The hyperscalers can afford to build because their core businesses still print cash. The labs can raise prices because their best users are already locked in. But customers spent the first phase of the AI boom treating intelligence as abundant and cheap. Now they are about to find out what their AI strategies are worth when the meter starts running.
