The commercialization divergence among the four leading AI giants in the United States: the one with the largest user base is not the most profitable.
In July 2026, The Walt Disney Company made a quiet yet meaningful decision: starting from August, it will discontinue the use of Microsoft's GitHub Copilot among its U.S. technical teams, switch to OpenAI's Codex, and retain Claude Enterprise and Cursor as auxiliary tools at the same time. The decision was based on a simple metric — tracking tool utilization rate and development speed. The standard for enterprises to procure AI has shifted from "whether to deploy" to "which model can produce high-quality code at lower cost and with less rework".
This shift in decision-making reveals a market differentiation concealed by 1 billion monthly active users: in the AI assistant business, the B-end is becoming an increasingly important revenue engine. The revenue structures of four leading companies clearly reflect this trend. OpenAI's annual recurring revenue (ARR) has exceeded 40 billion U.S. dollars as of August, and Sarah Friar, its Chief Financial Officer, confirmed at the investor meeting on August 14 that enterprise business revenue has surpassed the consumer business dominated by ChatGPT — the ratio was 40:60 (enterprise: consumer) at the beginning of the year, and the two lines have now crossed. Anthropic's ARR has exceeded 650 billion U.S. dollars, about 75% to 85% of its revenue comes from enterprise-level API calls, and subscription revenue only accounts for 15%. Google Gemini realizes B-end monetization mainly through bundled sales on the cloud and Workspace, and Google Cloud's Q2 revenue skyrocketed 82% year-on-year. xAI Grok's ARR exceeds 1 billion U.S. dollars, and its revenue mainly comes from X Premium+ subscriptions, Enterprise API and computing power rental. A relatively high proportion of its revenue came from the C-end at the beginning of the year.
There is no equal sign between user scale and profitability. ChatGPT has more than 1 billion monthly active users, but it was overtaken in revenue by Claude whose user volume is far less than itself. The one with the largest number of users is not necessarily the most profitable; the one with the smallest number of users is instead becoming the most profitable one — this is the commercial differentiation of the top four U.S. AI companies at present.
Programming Battlefield: The Only AI Sub-scenario That Has Achieved Unit Economics
Codex's growth curve perfectly explains Disney's choice. At the beginning of 2026, it had about 5 million weekly active users; it exceeded 5 million on June 2; on July 21, the total weekly active users of Codex and ChatGPT Work reached about 10 million; as of August, Codex's weekly active users have exceeded 20 million, growing more than 30 times in five months. In the same period, the growth rate of ChatGPT's main product has slowed down to single digits. On April 2, OpenAI switched Codex from a subscription model to token-based billing, and extended it to all ChatGPT Enterprise customers. Users in the programming scenario are willing to pay for usage, because every line of AI-generated code can be converted into saved engineer working hours.
Claude Code is another strong proof. After this programming assistant was launched in May 2025, its ARR reached 2.5 billion U.S. dollars in February 2026, soared to 6.3 billion at the end of March, exceeded 10.3 billion at the end of April, and surpassed 14 billion in June. As of August 10, data tracking platforms show its ARR is about 15.12 billion U.S. dollars. However, its growth rate in the past four weeks has dropped sharply from several times in the early stage to 5.2%, while Codex grew by 20.8% in the same period, and the competition has entered a white-hot stage. The essence of the competition between the two is to compete for the only scenario where AI has achieved unit economics at present.
Inversion of User Volume and Revenue: Why 1 Billion Monthly Active Users Cannot Beat 65 Billion Revenue
Looking further away, a more striking comparison emerges. At the end of July 2026, Anthropic's annualized revenue run rate (ARR) has exceeded 650 billion U.S. dollars, an increase of 38% over the 47 billion in May, and more than seven times the level of over 9 billion at the end of 2025. OpenAI disclosed that its ARR figure in April this year was about 25 billion U.S. dollars, and as of August, OpenAI's ARR has also exceeded 40 billion U.S. dollars, about twice that at the end of 2025.
In terms of user volume, ChatGPT's monthly active users have exceeded 1 billion. Claude has never announced its C-end monthly active users. Statistics from Sensor Tower show that the monthly active users of its Claude App in the second quarter of this year are about 56 million. The number of users differs by an order of magnitude, but OpenAI's revenue has been overtaken by Anthropic. The answer lies in the customer structure: 8 of the top 10 Fortune 500 companies are Anthropic's customers, there are more than 1000 enterprise customers paying more than 1 million U.S. dollars per year, and more than 100,000 enterprises run Claude through AWS Bedrock. They don't spend 20 U.S. dollars a month on a chat toy, but embed AI into their core workflows, issuing seven and eight-figure bills.
Claude's Moat: Multi-Cloud Neutrality and High Gross Margin
Anthropic got one thing right: multi-cloud neutrality. It is one of the few cutting-edge models that currently provides services on the three major platforms of AWS Bedrock, Google Cloud Vertex AI and Microsoft Azure Foundry, turning cloud vendors from competitors into distribution channels. OpenAI did not change Microsoft's IP authorization from exclusive to non-exclusive until April 2026, but Claude has taken the lead in multi-cloud layout.
Gross margin is a more solid evidence. Data from Semi Analysis shows that Claude's reasoning infrastructure gross margin has jumped from 38% a year ago to more than 70%. In Q2 2026, Anthropic is expected to achieve its first operating profit of about 559 million U.S. dollars, and Semi Analysis expects its Q3 GAAP EBIT to exceed 1 billion U.S. dollars — the scale effect has begun to show operating leverage. Among the four leading products, Claude's unit economic model has taken the lead in getting through. Its risk is no longer survival, but when competing products are focusing on the enterprise side, the renewal rate of 8 of the top 10 Fortune companies determines whether the 650 billion ARR is the starting point or the ceiling.
ChatGPT's Traffic Paradox: The Computing Power Bill Under 1 Billion Monthly Active Users
Back to ChatGPT. 1 billion monthly active users, a dominant number. But do the math: in February 2026, the number of subscribers was 50 million, corresponding to 900 million weekly active users at that time, with a payment rate of about 5.5%; if estimated based on the latest 1 billion monthly active users, the payment rate is about 5%. More than 95% of users are consuming computing power for free. OpenAI's revenue in 2025 was about 13.07 billion U.S. dollars, and its operating loss was about 20.92 billion U.S. dollars. Revenue growth cannot keep up with cost expansion, because every additional free user adds an additional reasoning cost.
OpenAI is not without trying to monetize. Its advertising business launched at the end of 2025 has reached an annualized revenue run rate of 1 billion U.S. dollars by the end of August 2026, less than 200 days after launch. Revenue in Q1 2026 was 5.7 billion, and 6.7 billion in Q2, rising quarter-on-quarter. More importantly, CFO Sarah Friar confirmed in mid-August: enterprise customers grew by 32% in July, the company's overall ARR grew 20% month-on-month, and the proportion of enterprise business revenue has surpassed consumer business (over 50%). But all these, in the face of tens of billions of losses, still prove the astonishing speed of burning money.
1 billion monthly active users is ChatGPT's largest asset and its largest liability. It constitutes a data flywheel, and also constitutes a cost flywheel. Codex is the key path for OpenAI to transform from a "traffic product" to a "value product" — if it can replicate the high ARPU model in the enterprise programming market, user scale makes sense; otherwise, they are just an increasingly expensive computing power bill.
Grok's Survival Game: Stagnant Growth and SpaceX Capital Injection
Grok's story is more stern. 117 million monthly active users — from SpaceX's IPO filing (as of March 31, 2026). But its growth has stagnated: after growing from 35 million in April 2025 to 117 million in March 2026, the growth rate has slowed down significantly. According to data from Apptopia, the daily active users of the Grok App in the U.S. have dropped by 28% since April.
The financial situation is also not optimistic. According to foreign media data, xAI's product ARR is about 1 billion U.S. dollars; but SpaceX's AI business (including xAI, X platform and cloud services) achieved revenue of 2.56 billion U.S. dollars in a single quarter in Q2 2026 — the difference between the two mainly comes from the latter including revenue from computing power rental and other items. In 2025, xAI's operating loss was about 6.4 billion U.S. dollars (revenue of 3.2 billion U.S. dollars), and its annualized capital expenditure was about 30.8 billion U.S. dollars. In February 2026, SpaceX acquired xAI in an all-stock transaction with a valuation of 250 billion, and the combined total valuation reached 1.25 trillion — essentially using SpaceX's cash flow to extend Grok's life. After the acquisition, all 11 co-founders left.
The product strategy is rather sobering: Grok 4.6 scored 61 on the Artificial Analysis Intelligence Index, ranking third in the world alongside GPT-5.6 Sol Max; Grok 4.5 scored 64.7% on SWE-bench Pro, exceeding GPT-5.5's 58.6%; Grok 4 Fast ranked first with 1163 points in the LMArena search test. Providing near-top performance at half the price is a rational choice when falling behind.
But cost performance cannot solve the unit economy. SuperGrok costs up to 30 U.S. dollars per month, X Premium 8-16 U.S. dollars, and the API is billed by volume — these revenues are a drop in the bucket compared to the multi-billion annual burn rate. If it cannot prove its business model before 2027, Grok will always be an affiliated function of the X ecosystem. How long SpaceX's money can last is the biggest variable.
Gemini's Ecological Gamble: Is AI Overviews Honey or Poison
Gemini's situation is the most complex, it is a layer of capability rather than a single product. Independent App, search AI mode, Workspace, Android system-level integration — Google has filled all of them. The effect is immediate: the monthly active users were about 750 million in February 2026, the financial report at the end of July showed 950 million, and it officially exceeded 1 billion on August 11, becoming the fastest-growing product in Google's history. The search AI mode has 1 billion monthly active users, and AI Overviews has 2.5 billion monthly active users.
But how many of the 1 billion monthly active users are "active users"? Android pre-installation and search diversion contribute a lot. About 90% of the Fortune 100 companies use Gemini Enterprise, but "usage" and "deep adoption" are two different things. Workspace bundled sales make the penetration rate look good, and Google has never disclosed the actual payment rate of AI Pro subscriptions (19.99 U.S. dollars per month).
AI Overviews is the biggest bet. A randomized controlled experiment conducted by the Indian School of Business and Carnegie Mellon University shows that when AI Overviews appears, natural clicks drop by 38%, and the relative increase in zero-click searches is 33%. eMarketer predicts that Google's U.S. search ad share will fall below 50% for the first time in 2026, to 48.9%. But Google's own figures tell another story: search and other ad revenue in Q1 2026 was 60.4 billion U.S. dollars, up 19% year-on-year, and query volume hit an all-time high — Pichai attributed the growth to AI modes driving user return.
The two sets of data are not contradictory, they speak to two temporal dimensions of the same thing. In the short term, AI Overviews has increased total search volume and ad exposure efficiency, and search revenue is rising. In the long run, if "zero-click search" becomes the norm and the content ecosystem shrinks, the information quality foundation of search will be eroded — and search is Google's cash cow. It is still unknown whether AI Overviews is enhancing search or eroding it. Alphabet's capital expenditure on AI infrastructure remains high — capital expenditure in Q2 2026 reached 44.9 billion U.S. dollars, a year-on-year increase of 100%, and the full-year guidance was raised to 1950 billion to 2050 billion U.S. dollars — Google is betting real money that the answer to this question is the former.
Who Is Swimming Naked? The Commercialization Interrogation of Four Companies
Spread out the books, the conclusion is clear:
Who is paying? The B-end is becoming an increasingly important revenue engine. 75%-85% of Claude's revenue comes from enterprises; OpenAI's enterprise revenue has exceeded 50% and is growing faster; Gemini relies on ecological bundling for penetration; Grok is dominated by subscriptions within the X ecosystem, and its commercialization path is still being explored. Personal subscriptions (20-30 U.S. dollars per month) are popular but the payment rate is low. ChatGPT's payment rate of about 5% shows that ordinary consumers have not yet developed the habit of paying separately for AI assistants.
Which business is established? Programming assistants. Only programming can clearly calculate ROI — generated code is converted into engineer working hours, and enterprises are willing to pay for verifiable usage. General conversation cannot do this, and can only rely on low-conversion subscriptions and experience-damaging ads.
Who is swimming naked? The water level has been drawn: Claude's 70%-85% gross margin and high-ARPU enterprise structure are the most resistant to pressure, and its Q3 expected EBIT exceeds 1 billion U.S. dollars, making it the first of the four to approach stable profitability; OpenAI's absolute scale on the B-end (over 20 billion U.S. dollars per year) is also not to be ignored, with enterprises accounting for the majority of the 40 billion ARR, but the overall is still in huge losses; Grok relies on SpaceX for blood transfusion, and the independent unit economy is far from being realized; Gemini has an ecosystem but its independent payment rate is a mystery, and AI Overviews is a sword hanging over the search cash cow.
Anthropic's business model is most dependent on enterprises (the highest proportion), OpenAI is making a major shift to the B-end in the mixed C-end/B-end model, Google follows closely, and xAI is still struggling on the C-end. The ultimate test of this business is very simple: are users willing to open a separate App for AI and pay a separate sum of money? At present, Claude has given the most affirmative answer on the enterprise side, and OpenAI has proved the potential of the mixed model with its B-end overtaking. The rest of the companies are either digesting the computing power bills of free users, or relying on their parent company's money to wait for the tide to recede.
The tide of AI assistants is still rising, but when the tide rises, no one knows who is swimming naked — at least now, the one with the thickest swimsuit has surfaced.
This article is from the WeChat official account "Tech Business", author: Tech Business, published with authorization from 36Kr.