Six days ago, we published a piece arguing that every AI subscription is a ticking time bomb for enterprise. The argument: AI labs are selling compute at a fraction of its cost, companies have built load-bearing workflows on those subsidized prices, and when the correction comes, it will be brutal.
We did not expect the fuse to be quite this short.
On May 14, three days after that article went live, both Anthropic and OpenAI made moves that took the repricing from theoretical to actual. It is happening now, publicly, with competitive aggression that tells you neither company thinks the current model survives the year.
Anthropic announced that effective June 15, it is splitting its Claude subscription into two separate usage pools. Interactive usage, the kind where a human sits at a keyboard and has a conversation, stays covered under existing subscription limits. But automated and programmatic usage, the kind generated by agentic tools like OpenClaw and third-party harnesses that route tasks through Claude around the clock, is being pulled out of the base subscription entirely and placed behind a separate credit meter with fixed monthly API credits.
That is the structural change the original article warned about. The all-you-can-eat model did not survive contact with agentic workloads. Anthropic's head of Claude Code, Boris Cherny, had already said in April that subscriptions "weren't built for the usage patterns of these third-party tools." The June 15 change makes that position official policy. Axios put it well: the industry is rediscovering a lesson from earlier eras of computing. Humans have built-in limits to how much data they can consume. A human might send dozens or hundreds of prompts a day. An autonomous coding agent can generate thousands of requests, run tests continuously, browse the web, and recursively call models. The subscription was priced for the human. The agent showed up anyway.
The reaction from users was immediate and hostile. Critical replies flooded the announcement, with developers calling the changes "gaslighting" and publicly declaring they were switching to OpenAI's Codex. Which is exactly what OpenAI was counting on.
Within hours of Anthropic's announcement, Sam Altman posted that OpenAI would give enterprise users two free months of Codex access if they switched within 30 days, complete with a one-click migration tool that transfers Claude Code prompts, skills, and MCP configurations. Anthropic countered the same day by boosting Claude Code's weekly usage limits by 50% for Pro, Max, Team, and Enterprise users, valid until July 13, a window that overlaps almost perfectly with OpenAI's free trial period.
Call it what it is: a pricing war fought in the open, with each company trying to lock in developers before the IPO window opens in the second half of the year.
And they are not the only ones moving. GitHub announced that all Copilot plans will transition to usage-based billing on June 1, 2026, replacing flat-rate premium requests with token-based AI Credits. Their own blog post was unusually direct about the reason: "Copilot is not the same product it was a year ago. It now powers far more complex, agentic workflows that consume far more compute." Annual plans will not auto-renew. Monthly subscribers will be migrated automatically. The flat-fee era at GitHub is over in two weeks.
The subscription restructuring grabbed headlines. But a quieter change had already been extracting more money from developers for weeks.
When Anthropic released Claude Opus 4.7 in April, it kept the per-token rates identical to Opus 4.6: $5 per million input tokens, $25 per million output tokens. Same numbers on the same pricing page. But Opus 4.7 shipped with a new tokenizer. For the unfamiliar: the tokenizer is the component that sits between your text and the model and decides how many tokens your words are worth. The new one is more aggressive. The same input, the same sentence, the same document, now generates up to 35% more tokens than it did on the previous model. For developers who migrated to the new flagship without benchmarking, bills climbed by as much as 27% with no visible change to the pricing page.
That kind of repricing does not show up in an announcement blog post. It shows up on the invoice three months later.
The consumer subscription changes are the part you can see. Underneath, the enterprise contracts have been shifting since late 2025.
The Register reported in April that Anthropic began renewing enterprise customers under usage-based plans as early as November 2025. By February, the company introduced a single $20-per-employee monthly seat covering Claude chat, Code, and Cowork, with all token consumption billed separately at API rates. Anthropic's support documentation now states explicitly: "Chat-only seats and Standard/Premium seats are no longer available for new contracts. Both legacy plan types are transitioning to the single Enterprise seat at their next renewal." Gizmodo's reporting confirmed it bluntly: the legacy plans are dead.
The transition from subscription to metered billing that the original article identified as inevitable is now official. Anthropic was the first major provider to make it explicit. The flat-fee enterprise AI seat, the thing that made AI cheaper than a single SaaS tool on the P&L, is being retired. And as CNBC noted, OpenAI's Nick Turley acknowledged on the BG2 podcast that "it's possible that in the current era, having an unlimited plan is like having an unlimited electricity plan. It just doesn't make sense."
The spending data coming out of the enterprise side makes the cost problem concrete.
Uber burned through its entire 2026 AI budget by April. Four months. The company's CTO, Praveen Neppalli Naga, told The Information he is "back to the drawing board, because the budget I thought I would need is blown away already." Uber had rolled out Claude Code to its engineering team in late 2025 and stood up internal leaderboards ranking developers by usage. By February, usage had doubled. By April, 84% of Uber's developers were classified as agentic-coding users in internal telemetry. Monthly API costs per engineer ranged from $500 to $2,000, with 95% of Uber engineers now using AI tools monthly and 70% of committed code originating from AI. The tool proved too successful to afford at scale.
ServiceNow hit the same wall. CIO Kellie Romack described the rapid cost increases as a difficult management challenge, making ServiceNow the second major public company to disclose that it blew through its full-year Anthropic budget in the first few months of 2026. Gergely Orosz of the Pragmatic Engineer newsletter coined a term for what is happening across Silicon Valley: "tokenmaxxing." Great for AI vendors, he wrote. Bad for everyone else.
KPMG's data shows U.S. organizations projecting average AI spending of $207 million over the next 12 months, nearly double what they projected a year ago. Goldman Sachs surveys indicate large companies are already overrunning their AI budgets by orders of magnitude.
The original article quoted Brian Jabarian, an economist at the University of Chicago: "The time for the bill is going to come." The bill is here.
Both OpenAI and Anthropic are on IPO timelines for the second half of 2026. OpenAI completed the largest private funding round in history in April, $122 billion at an $852 billion post-money valuation. Anthropic has reportedly surpassed $30 billion in annualized revenue. Massive numbers, both of them. Also both attached to companies that are still burning cash at extraordinary rates.
Public markets will not tolerate the gap between subscription revenue and compute cost that has defined the past three years. The moment either company files, analysts will demand unit economics that show a path to margin. Usage-based billing is the fastest way to demonstrate that path.
None of this contradicts the repricing thesis. The pricing war is the last land grab before the gate closes. Both companies are spending aggressively now to lock in users whose switching costs will make them sticky when prices rise. OpenAI offers two months free. Anthropic offers 50% more capacity. Both expire in July. What comes after July is the real pricing.
The competitive responses on May 14 revealed something bigger than a pricing spat. Two distinct enterprise AI coalitions are taking shape, and they disagree on something fundamental: how to bill for intelligence.
OpenAI, backed by Microsoft and the Stargate infrastructure alliance, is positioning around unified, unrestricted enterprise plans. The pitch is simplicity: one subscription, no tiered restrictions, no metered surprises. The bet is that volume and ecosystem lock-in will generate enough revenue to cover costs eventually.
Anthropic, aligned with Google, Amazon, and its growing compute partnerships, is moving toward tiered, capacity-controlled subscriptions with per-token billing that reflects actual usage. The bet is that transparent metering will produce cleaner revenue data for public markets and a more defensible business model long-term. CNBC's analysis framed Anthropic's approach as the more prudent strategy, noting that when both companies go public, Anthropic will have cleaner data on what its customers actually value, while OpenAI will have bigger numbers but a harder time proving how much of them are real.
Enterprise buyers choosing between these camps are not picking a vendor. They are betting on which billing model wins. And that bet will determine whether your AI budget is predictable or volatile for the next several years.
The original article laid out a set of recommendations: audit token consumption, model costs at 2x to 10x current prices, build vendor optionality, and have the conversation with the CFO before the CFO has it with you.
All of that still stands. The urgency is different now. A week ago, the repricing was a forecast. Today it is a fact. Anthropic has formally decoupled agentic usage from subscription pricing. Legacy enterprise seats are being retired. A stealth tokenizer change has already raised effective API costs by double digits. Companies are blowing through annual AI budgets in months.
The subsidy era did not end with a bang or a whimper. It ended with a press release on a Wednesday in May, followed immediately by a competitor offering two free months to poach the users who noticed.
The clock we warned about last week? It already went off. The question now is whether your organization heard the alarm.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。