Elon Musk wrote on X on 1 November that “The cost of serving Grok Bot was unsustainably high relative to revenue”, replying to a user whose usage “ran out really fast this week”. About 74 minutes later, by our arithmetic, Replit chief executive Amjad Masad posted: “If you’re enjoying a subsidized AI subscription get ready to get rug pulled.”
Anthropic’s S-1 shows how large that subscriptionbusiness has become at one lab. Subscription revenue was $3,222 million in the six months to 30 June 2026, against $180 million a year earlier. By our arithmetic that is 17.9 times as much, while consumption revenue from the API grew 13.3 times, so subscriptions rose from 15.5% to 19.8% of Anthropic’s revenue.
The key insight: Elon Musk said on 1 November that serving Grok Bot cost too much relative to revenue, in a week when Microsoft, Anthropic and OpenAI each changed how plans count usage; Anthropic’s S-1 shows subscriptionrevenue up 17.9 times in a year, faster than its API, by our arithmetic.
Musk’s reply went to the account @DogecoinNorway, which had asked at 20:11 UTC: “Any chance for a Grok Bot reset, @elonmusk? Lots of people say their usage ran out really fast this week.” Musk answered at 20:34 UTC. His post gave no figures for cost, revenue or the limits, and did not say whether a reset would happen. It had about 5.77 million views when we read it on 2 November.
Masad’s post quoted Sawyer Merritt’s post relaying Musk’s explanation. Masad had written in June, in a reply to another account: “The rug pull is coming… sooner or later.” Our first report on Musk’s reply is in Musk says Grok Bot’s serving cost was too high for revenue.
Analyst Shanu Mathew called Musk’s reply “a cost anecdote on the personal agent tools” and asked what it means “for future usage limits or if it flips entirely to API spend”.
What Was Posted on X
@DogecoinNorway The cost of serving Grok Bot was unsustainably high relative to revenue
Elon Musk’s reply of 1 November, 20:34 UTC, to a user asking for a Grok Bot reset: “The cost of serving Grok Bot was unsustainably high relative to revenue.” It gives no figures for cost, revenue or limits.
If youโre enjoying a subsidized AI subscription get ready to get rug pulled.
Replit chief executive Amjad Masad, about 74 minutes later by our arithmetic, quoting a post that relayed Musk’s explanation: “If you’re enjoying a subsidized AI subscription get ready to get rug pulled.” He names no company and no date.
Weโre renaming our Max plans:
Max 5x โ Max Max 20x โ Ultra
Max is still $100/month, and Ultra is still $200. If youโre a subscriber, thereโs nothing you need to do.
Weโre also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until theyโre verified, which helps us prevent fraud and abuse. You’ll get an email when it’s removed.
Anthropic’s 30 October post renaming Max 20x as Ultra at $200 a month and removing the 5-hour limit for verified Ultra accounts. It says weekly limits remain unchanged: a change that loosened one limit, not a tightening.
Ahead of Anthropic’s IPO:
Worldwide traffic to Claude’s upgrade page rebounded in September, nearly returning to its May and June levels.
Similarweb, the web-traffic measurement firm, says worldwide traffic to Claude’s upgrade page rebounded to 17.6M in September from 14.7M in August, nearly returning to the 18.6M it recorded in both May and June (July: 16.8M). The piece uses it as a read on demand for paid Claude plans ahead of Anthropic’s IPO.
Anthropic’s consumption-based and subscription-based revenue for the six months to 30 June 2025 and 2026, from its S-1, with the growth multiples and subscription shares as our arithmetic; right, the plan changes announced by Microsoft, Anthropic, OpenAI and Elon Musk between 28 October and 1 November 2026, per their own statements.
Five Days of Plan Changes
Musk’s post came at the end of a week in which three other large sellers of AI subscriptions changed how usage is counted or priced.
On 28 October, Microsoft chief executive Satya Nadella said: “Our business model is expanding from per-seat subscriptions to include consumption, expanding our TAM and ARPU.” He said the number of Copilot customers consuming credits through usage-based billing grew over 75 percent in the quarter, as we covered in Microsoft takes Copilot past per-seat.
On 30 October, Anthropic renamed Max 20x as Ultra, kept it at $200 a month and said it is “removing the 5-hour limit for verified Ultra accounts”. It added: “Weekly limits remain unchanged.” That change loosened a limit rather than tightening one, as we reported in Claude Ultra drops its 5-hour limit.
On 31 October, OpenAI’s help center said that “Starting in December, Plus, Pro, and Business plans’ usage allowance will be shared across ChatGPT and Codex”, with usage “measured in tokens”. It says that “More advanced models and complex tasks generally use more of your allowance” and that users who reach the limit can “wait for it to reset, upgrade your plan, or buy credits”. We covered the notice in our piece on the shared ChatGPT and Codex allowance.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on usage-based pricing. It explains the general idea only and does not discuss this week’s plan changes.
What Anthropic’s Filing Says About Limits
Anthropic’s S-1 describes its subscriptions as “subject to usage limits that reset over short periods”. Its risk factors say: “We have also changed, and may in the future change, usage limits, rate limits, or model and product capabilities, which could adversely affect our reputation and the perceived value of our offerings.”
The filing’s revenue table splits the business in two. In the six months to 30 June 2026, consumption-based revenue was $13,064 million and subscription-based revenue $3,222 million, for a total of $16,286 million. A year earlier the figures were $983 million and $180 million. Compute and infrastructure expense was $8,945 million in the 2026 half, the filing shows; it does not split that cost between subscriptions and the API.
SemiAnalysis estimated in an article on 5 October that subscriptions were about 10% of Anthropic’s revenue but “can take up over 40% of inference compute”. It called the figures “rough numbers”, and they are its estimates, not Anthropic’s. The S-1 table, filed later, puts subscriptions at 19.8% of revenue in the 2026 half, by our arithmetic; the two figures may not cover the same period.
On 7 October, SemiAnalysis posted a clip of a guest saying: “You can pay $200 for a Claude plan and get $12,000 of Opus 5.5 tokens at API pricing.” By our arithmetic that is 60 times the plan price, on that guest’s workload and at list API prices. Our report on the estimate is SemiAnalysis: Claude plans can take over 40% of inference, and per-token list prices are in AI token prices in 2026.
The Structural Read
As we read the week’s statements, the common thread is measurement: OpenAI will count ChatGPT and Codex usage in tokens, and Nadella described Microsoft’s model as expanding from per-seat subscriptions to include consumption.
As we read the S-1, the subscription line grew faster than the API line, so whatever each plan costs to serve weighs on a larger share of revenue: 19.8% in the 2026 half against 15.5% a year earlier, by our arithmetic.
Amjad Masad, Replit chief executive, on X
“If you’re enjoying a subsidized AI subscription get ready to get rug pulled.”
Three Implications
FOR SUBSCRIBERS OpenAI says users who reach the shared limit can wait for it to reset, upgrade their plan, or buy credits.
FOR INVESTORS IN AI LABS Anthropic’s S-1 lists changes to usage limits and rate limits among the things that could adversely affect its reputation and the perceived value of its offerings.
FOR SOFTWARE BUYERS Microsoft said the number of Copilot customers consuming credits through usage-based billing grew over 75 percent in the quarter.
The framework argues that a plan priced below what its heaviest users consume is doing strategic work: “The negative margin is not a leak. It is a procurement budget.”
As we read it, the week’s plan changes test how long each lab keeps paying that budget at the price it charges today. The framework is the owner’s argument and does not discuss Grok Bot or this week’s changes.
Musk gave no figures for Grok Bot’s cost, revenue or limits, and xAI has published no subscription revenue. Masad did not name a company or say when or how plans would change.
Anthropic’s filing does not say whether its subscriptions lose money, and SemiAnalysis’s compute share is an estimate. OpenAI says it expects “most users won’t need to wait for their usage to reset more often than they do now”, and Anthropic’s Ultra change removed a limit. The week’s changes do not all point the same way.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growth story. This week it started being priced as a credit story.
Musk said serving Grok Bot cost too much relative to revenue, and Amjad Masad told users of subsidized AI subscriptions to “get ready to get rug pulled”. In the same five days, Microsoft, Anthropic and OpenAI each changed how plans count usage. Anthropic’s S-1 shows subscription revenue up 17.9 times in a year, by our arithmetic, faster than its API.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Revenue figures and quoted passages on limits are from Anthropic’s S-1; plan changes are from Microsoft’s prepared remarks, Anthropic’s and OpenAI’s own statements, and Musk’s and Masad’s posts on X. SemiAnalysis’s figures are its estimates. Growth multiples, shares and the 60 times figure are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
