The strangest number in the AI race is not a market share. It is the $1.4 trillion in compute commitments OpenAI has signed up to over about eight years.
At the same time, ChatGPT's share of the chatbot market has been shrinking, Google's Gemini app has passed one billion monthly users, and both OpenAI and Anthropic have filed to go public. Put those together and one question keeps coming up: is ChatGPT actually losing the AI race?
The short answer: ChatGPT is losing share, not users. Google has closed the model gap and owns the distribution. Anthropic has moved ahead on run-rate revenue. And OpenAI is carrying the heaviest bill in the industry.
Below are 17 facts, each traced to the tracker, earnings call, filing or launch post it came from. Where two sources disagree, I explain why, because Similarweb, Apptopia and Sensor Tower each measure something different.
Last updated: September 19, 2026. Figures come from third-party trackers and company statements. They are estimates, not audited data, and nothing here is investment advice.
- Fact #1: ChatGPT's Share Has Dropped About 20 Points, but the Number Depends on the Tracker
- Fact #2: Gemini's App Went From 650 Million to 1 Billion Monthly Users in Under a Year
- Fact #3: The "Code Red" Memo Showed How Seriously OpenAI Took Gemini
- Fact #4: Gemini 3 Beat GPT-5.1 at Launch, but the Scoreboard Has Since Moved
- Fact #5: Anthropic's Revenue Run Rate Passed OpenAI's in April and Kept Climbing
- Fact #6: The Apple and Google Deal Changed the Distribution Math
- Fact #7: Google's Own Chips and Cash Give It Room to Keep Spending
- Fact #8: OpenAI Carries a $1.4 Trillion Compute Bill, and Ads Are Part of the Answer
- Fact #9: The Race Looks Different in the U.S. Than in the Rest of the World
- Fact #10: Enterprise Is a Different Race, and Anthropic Leads Most Estimates
- Fact #11: Both OpenAI and Anthropic Have Filed for IPOs
- Fact #12: Google Is Playing a Different Game With Agents and Ambient AI
- Fact #13: Users Are Multi-Homing, and It Is Reshaping the Market
- Fact #14: Google's Own Stumbles Make the Comeback More Remarkable
- Fact #15: ChatGPT Still Holds a Speed Record Nobody Else Has Matched
- Fact #16: The "Two-Horse Race" Is Already a Five-Horse Race
- Fact #17: OpenAI's Code Red Response Arrived in Weeks
- What to Watch Next
- So, Is ChatGPT Actually Losing the AI Race?
- Frequently Asked Questions (FAQ)
Fact #1: ChatGPT's Share Has Dropped About 20 Points, but the Number Depends on the Tracker
Start with the stat everyone argues about. According to Similarweb's generative AI data, ChatGPT's share of worldwide AI website visits fell from about 76% in June 2025 to roughly 53% by May 2026. Gemini climbed from under 9% to around 27 to 28% over the same stretch.
The slide started earlier. In early January, the tracker showed ChatGPT at roughly 64 to 65% of chatbot web traffic, its lowest level since 2023, down from about 86% a year earlier.
Mobile tells a sharper story. Apptopia's app data shows ChatGPT's U.S. share falling from 69.1% to 45.3% between January 2025 and January 2026, while Gemini rose from 14.7% to 25.2% and Grok jumped from 1.6% to 15.2%. The overall chatbot app market grew 152% in that time.
Note the window. That is January to January, not mid-2026, and it covers U.S. mobile only.
Then there is Sensor Tower's State of AI 2026 report, which combines apps and web worldwide. It put ChatGPT at about 46% in May 2026, below 50% for the first time.
- ❌ Similarweb (worldwide web visits): about 76% to about 53%, June 2025 to May 2026
- ❌ Apptopia (U.S. mobile app usage): 69.1% to 45.3%, January 2025 to January 2026
- ❌ Sensor Tower (global apps plus web): about 46% in May 2026
- ✅ ChatGPT is still first on every tracker above
Three trackers, three different rulers, one direction. What none of them show is ChatGPT losing users, which is the point of Fact #15.
Here is how the three trackers differ, so you know which number to quote:
| Tracker | What it measures? | Region | Time window |
|---|---|---|---|
| Similarweb | Visits to AI chatbot websites | Worldwide | June 2025 to May 2026 |
| Apptopia | Usage share of AI chatbot mobile apps | U.S. | January 2025 to January 2026 |
| Sensor Tower | Combined app and web audience | Global | Snapshot at May 2026 |
Quoting one tracker's number with another tracker's label is the most common mistake in this debate. Always name the source and the date.
Fact #2: Gemini's App Went From 650 Million to 1 Billion Monthly Users in Under a Year
Alphabet put the Gemini app at 650 million monthly users in October 2025 and 750 million by February 2026. Google cited 900 million at I/O and more than 950 million on the July earnings call. On August 11 it announced the app had passed 1 billion, according to SiliconANGLE's report on the milestone.
- ✅ October 2025: 650 million monthly users
- ✅ February 2026: 750 million
- ✅ Google I/O 2026: 900 million or more
- ✅ July 2026 earnings call: more than 950 million
- ✅ August 11, 2026: 1 billion
Alphabet's second-quarter results confirmed the 950 million figure and added that Google had pushed the cost of AI Mode responses to its lowest level since launch.
Google credited its model launch for part of the jump. When the app crossed 750 million, Sundar Pichai called Gemini 3's arrival in AI Mode a positive driver of growth.
Web traffic shows the same climb. In January 2026, Similarweb counted about 2 billion Gemini visits, up 647% in a year, while ChatGPT still drew 5.7 billion.
Fact #3: The "Code Red" Memo Showed How Seriously OpenAI Took Gemini
In a memo to staff on December 1, 2025, Sam Altman said more work was needed on ChatGPT's speed, reliability and personalization, according to AP's report on the memo. The Wall Street Journal broke the story.
One line stood out: "We are at a critical time for ChatGPT," Altman reportedly wrote. OpenAI's COO, Brad Lightcap, later described such alerts as a routine way to sharpen focus.
What got pushed back tells you the priorities:
- ✅ Advertising work: delayed
- ✅ AI agents for health and shopping: delayed
- ✅ The Pulse personal assistant: delayed
- ✅ Speed, reliability and personalization: moved to the front of the queue
A code red is a reallocation of engineers, not proof a product is failing. Still, leaders rarely pause revenue projects unless something is pressing. Facts #8 and #17 show how that pause played out.
Fact #4: Gemini 3 Beat GPT-5.1 at Launch, but the Scoreboard Has Since Moved
Google released Gemini 3 on November 18, 2025. Its Pro model debuted at the top of the LMArena text leaderboard with a preliminary Elo of 1501, the first model to cross 1,500. It also scored 91.9% on GPQA Diamond and 31.1% on ARC-AGI-2, according to VentureBeat's launch coverage. These were Google's own launch-day figures.
A side-by-side benchmark comparison shows Gemini 3 Pro ahead of GPT-5.1 on GPQA Diamond (91.9% vs. 88.1%) and roughly double on ARC-AGI-2 (31.1% vs. 17.6%). GPT-5.1 held level on coding, at 76.3% vs. 76.2% on SWE-bench Verified.
The rollout mattered as much as the scores. Gemini 3 arrived across app and Search at once, a day-one reach a standalone chatbot cannot match.
Leadership on paper has changed hands since. OpenAI shipped GPT-5.2 within weeks (see Fact #17). By April 2026, GPT-5.5 led the Artificial Analysis index, and a GPT-5.5 versus Gemini 3.1 Pro comparison shows 85.0% against 77.1% on ARC-AGI-2. Gemini 3.1 Pro kept a narrow edge on GPQA Diamond, 94.3% against 93.6%.
Benchmarks are snapshots. Treat any single leaderboard result as a moment in time.
Fact #5: Anthropic's Revenue Run Rate Passed OpenAI's in April and Kept Climbing
In April 2026, Trending Topics reported that Anthropic's annualized revenue had passed $30 billion, ahead of OpenAI's roughly $24 to $25 billion. Anthropic ended 2025 at about $9 billion, and its count of customers spending over $1 million a year had doubled to more than 1,000 in under two months.
The climb continued. Anthropic said its run rate topped $47 billion in May, and it passed $65 billion at the end of July, per Bloomberg, as relayed in CNBC's report on the $65 billion figure. OpenAI's run rate was recently about $40 billion. CNBC also put Anthropic's private valuation at $965 billion.
A run rate projects a full year from a recent, shorter period, so it is not the same as booked revenue. Treat the comparison with care: Axios points out that the two firms may not count revenue the same way.
Why the gap? Anthropic sells mostly to businesses and developers. ChatGPT built its base on a huge free tier. Fact #10 covers the enterprise side.
"The rivalry between these two labs goes back years. Here is how the Anthropic vs. OpenAI AI civil war actually started."
Fact #6: The Apple and Google Deal Changed the Distribution Math
On January 12, 2026, Apple and Google announced a multi-year collaboration in which the next generation of Apple Foundation Models will be based on Gemini models and Google's cloud technology, as CNN reported. The models are meant to power a more personalized Siri.
- ✅ Confirmed: a multi-year deal, with Apple's models built on Gemini
- ❌ Not stated in the announcement: how many iPhones are affected
- ❌ Not stated: that the Gemini app itself is on every iPhone
That distinction matters. The statement describes model technology, not a Gemini app on every phone. Apple also still lets Siri hand complex requests to ChatGPT, according to Campus Technology.
Even so, the strategic point stands. A model powering the assistant that ships on hundreds of millions of phones does not need to win a download race. It is already there.
Fact #7: Google's Own Chips and Cash Give It Room to Keep Spending
Google designs its own AI chips, called TPUs, and rents them out. In October 2025, Anthropic announced plans to use up to one million Google TPUs, with well over a gigawatt of capacity coming online in 2026.
Then there is the money. The Financial Times tallied that Amazon, Microsoft, Alphabet and Meta plan roughly $725 billion in 2026 capital spending, up 77% from about $410 billion in 2025. Google Cloud chief Thomas Kurian credited in-house chips, models and products for a cost and research advantage.
Alphabet has since raised its own 2026 guidance to $195 billion to $205 billion, up from $180 billion to $190 billion.
"Spending on this scale has a darker side too, and these AI arms race warning signs explain why."
Money is not only going into chips. It is also chasing talent.
"See how one hire became Meta's $200 million AI researcher."
Spending is not the same as winning. Capex only pays off if demand shows up, which is why the revenue facts in this list matter as much as the chip facts.
Fact #8: OpenAI Carries a $1.4 Trillion Compute Bill, and Ads Are Part of the Answer
The $1.4 trillion from the title is OpenAI's own figure. The list of counterparties is long: Oracle, Nvidia, Microsoft, AMD, Broadcom and Amazon all appear in its deals, set against a company that expected around $20 billion in revenue that year.
The Code Red pause on advertising did not last. In February 2026, OpenAI announced its ad test for logged-in adults in the U.S. on the Free and Go tiers. Plus, Pro, Business, Enterprise and Education plans stay ad-free, and OpenAI says ads do not influence answers.
Microsoft's filings also show the strain. For the quarter ended September 30, 2025, Microsoft reported that OpenAI-related losses cut its net income by $3.1 billion. That is equity-method accounting, meaning Microsoft's share of OpenAI's losses, not $3.1 billion of cash written off.
- ❌ About $1.4 trillion in commitments over roughly eight years
- ✅ Ads test began in February 2026, after the Code Red pause
- ❌ A $3.1 billion quarterly hit to Microsoft's income from its OpenAI stake
The takeaway: a bill this size explains why OpenAI needs ads, enterprise contracts and an IPO option at the same time.
Fact #9: The Race Looks Different in the U.S. Than in the Rest of the World
Global averages hide regional gaps. In April 2026, Momentic's Similarweb-based ranking showed ChatGPT at 54.7% of worldwide web visits, Gemini at 27.4% and Claude at 8.2%. In the United States, ChatGPT held 58.9%, Gemini 19.2% and Claude 12.5%.
Comscore's U.S. desktop numbers point the same way. In March 2026, Comscore's chatbot rankings had ChatGPT at 33.86 million unique visitors, Gemini at 10.66 million and Microsoft Copilot at 5.02 million.
Two things stand out. First, ChatGPT is stronger in the U.S. than worldwide. Second, Claude does noticeably better at home (12.5%) than globally (8.2%).
The AI race is not one war. It is several regional ones, and Google and OpenAI are not winning them at the same pace.
Fact #10: Enterprise Is a Different Race, and Anthropic Leads Most Estimates
Consumer traffic is not where the margins are. In mid-2025, Menlo Ventures' survey put Anthropic at 32% of enterprise LLM usage, OpenAI at 25% and Google at 20%. OpenAI had held 50% at the end of 2023.
Later figures pushed further. Fortune reported Menlo's newer split as Anthropic 40%, OpenAI 27% and Google 21%. OpenAI disputes the numbers, noting that Menlo is an investor in Anthropic, and says more than 1 million businesses use its tools.
In coding, the split was wider still. Menlo's data had Anthropic at 54% of enterprise coding, against 21% for OpenAI and 11% for Google.
These are survey-based estimates, not audited revenue. Still, they explain how a company with a fraction of ChatGPT's audience can lead on run-rate revenue (Fact #5).
Fact #11: Both OpenAI and Anthropic Have Filed for IPOs
Anthropic confidentially submitted a draft registration on June 1, 2026. OpenAI followed with its confidential IPO filing on June 8, a little more than a week later, at a last-reported valuation of $852 billion. Anthropic's private valuation was reported at $965 billion (see Fact #5).
Neither company has committed to a date. OpenAI wrote that "it may be a while," because some of its plans are easier as a private company.
Why does share matter here? A listing invites public scrutiny of growth. Shrinking share does not sink an IPO. But "most used" and "fastest growing" are no longer the same claim, and public investors read that difference closely. This is analysis, not investment advice.
Fact #12: Google Is Playing a Different Game With Agents and Ambient AI
At Google I/O 2026, the company said AI Overviews has more than 2.5 billion monthly users and AI Mode has passed 1 billion, according to Thurrott's I/O coverage. The Gemini app was past 900 million at the same event, and Nano Banana had generated over 50 billion images in a year.
Google is also pushing agents into daily habits. On the Q2 call, Pichai highlighted features such as Daily Brief and a personal agent, and said the app's daily users had tripled, per the full call transcript.
The bet is different from OpenAI's. ChatGPT wins by being the destination. Gemini increasingly wins by being inside products people already open.
"Google's AI answers also change what creators can publish. Read how Google treats AI-written captions."
Fact #13: Users Are Multi-Homing, and It Is Reshaping the Market
People have stopped picking just one assistant. Apptopia found that one in five AI users now regularly uses more than one AI app, a pattern its analysts compared to streaming services.
Similarweb's data adds scale. Between January 2025 and January 2026, generative AI platform visits grew 28.6%, ChatGPT shed 22.2 points of share, and DeepSeek and Grok each crossed 3% for the first time.
Mind the gap between growth figures. The 152% is Apptopia's U.S. mobile market growth. The 28.6% is Similarweb's global web-visit growth. Different rulers again.
If people use several assistants, "losing the AI war" may be the wrong frame. There may be no single war to win.
Fact #14: Google's Own Stumbles Make the Comeback More Remarkable
Google's road here was not smooth. In February 2024, it paused Gemini's image generation of people after the feature produced inaccurate images. Senior Vice President Prabhakar Raghavan wrote that the feature "missed the mark," in Google's own post on the pause.
Three months later, AI Overviews went viral for the wrong reasons. Search chief Liz Reid explained that the glue-on-pizza and eat-rocks answers traced back to forum content and a satirical page.
That history is why 2026's numbers read as earned rather than inevitable. Google had to repair trust before its distribution could help it.
Fact #15: ChatGPT Still Holds a Speed Record Nobody Else Has Matched
Before writing OpenAI's obituary, look at the audience. Sensor Tower estimates ChatGPT has more than 1.1 billion monthly active users, the fastest mobile app to reach one billion, with Gemini at 662 million and Claude at 245 million, per MediaNama's summary of the report. Those are Sensor Tower's estimates, which is why Gemini's number is lower than Google's own 1 billion.
OpenAI's own weekly figure was 900 million in February. On July 29, The Information reported ChatGPT was nearing 1 billion weekly users, about seven months late against OpenAI's internal target.
So growth slowed, but it did not reverse. ChatGPT lost share of a rapidly expanding pie, not users from a shrinking one.
Fact #16: The "Two-Horse Race" Is Already a Five-Horse Race
Framing this as Google vs OpenAI misses who is gaining fastest. According to Similarweb's June update, Claude's share of worldwide AI web traffic rose from 1.6% to 8.9% in twelve months, while ChatGPT fell from 76.4% to 52.7% and Gemini rose from 8.9% to 27.3%.
The smaller players are stable. In the same panel, DeepSeek, Grok, Copilot and Perplexity sit at roughly 3.6%, 2.5%, 2.0% and 1.1%. This is panel data that no outside auditor has verified.
"China's labs are part of this story too, as covered in our Trump AI China race breakdown."
"Grok's owner has strong views on where all this is heading, collected in these Elon Musk superintelligence quotes."
The market is fragmenting into task-specific winners, each optimized for a different use case, price point or platform.
Fact #17: OpenAI's Code Red Response Arrived in Weeks
OpenAI did not wait. It launched GPT-5.2 on December 11, 2025, less than two weeks after the memo, in Instant, Thinking and Pro versions.
The pace kept up. GPT-5.3 and, on March 5, 2026, GPT-5.4 followed, as one model timeline records. By April, GPT-5.5 was leading the Artificial Analysis index (Fact #4), and ChatGPT crossed one billion monthly users (Fact #15).
That is why betting against OpenAI on speed alone is risky. The company has responded to pressure quickly before.
"OpenAI's next big claim is covered in our GPT-6 Astra AGI claim post."
What to Watch Next?
Three signals will tell you more than any single share number.
First, watch whether the IPO filings become public. A public prospectus would show revenue, margins and spending commitments in audited form, which no tracker in this article can do.
Second, watch the next Similarweb, Sensor Tower and Apptopia releases. If ChatGPT's share keeps falling while its user count keeps rising, the "losing share, not users" reading holds. If both start falling, the story changes.
Third, watch the model cadence. Fact #4 and Fact #17 show that benchmark leadership has changed hands within weeks. Any lead you read about today may be gone by the next launch.
So, Is ChatGPT Actually Losing the AI Race?
Here is the honest answer: ChatGPT is losing ground, not losing.
- ✅ Google closed the model gap and pairs it with distribution OpenAI cannot copy: Search, Android, Chrome, Workspace and a Siri deal
- ✅ Anthropic leads on run-rate revenue and on most enterprise estimates, and both companies are heading toward IPOs
- ✅ OpenAI still has the biggest audience and the fastest-growing consumer app in history, but it also has a $1.4 trillion bill to pay
The real story of 2026 is not Google beating OpenAI. It is the end of the assumption that one company would run away with this the way Google once ran away with search. Three serious competitors, plus a growing field behind them, is a more accurate headline than a simple winner and loser.
"Want the long view on where this goes next? See these AI 2027 predictions."
Frequently Asked Questions (FAQ)
Is Google Gemini better than ChatGPT in 2026?
It depends on the task and the month. Gemini 3 led GPT-5.1 on several benchmarks at launch, but GPT-5.5 led the Artificial Analysis index by April. Gemini has the distribution edge, while ChatGPT has the larger audience, with Sensor Tower estimating more than 1.1 billion monthly users.
Why did Sam Altman declare a "code red" at OpenAI?
Altman sent the memo on December 1, 2025, weeks after Gemini 3's launch. It pushed staff to improve ChatGPT's speed, reliability and personalization, and it delayed advertising, health and shopping agents and the Pulse assistant.
What is the $1.4 trillion figure?
It is the amount OpenAI says it has committed to spend on data centers and cloud services over roughly eight years. It is a spending commitment, not the size of the AI market or a valuation.
Has Anthropic overtaken OpenAI in revenue?
On annualized run-rate revenue, reports say yes since April 2026, and the gap has widened since. Run rate is not the same as booked revenue, and the two companies may not measure it identically.
Will ChatGPT lose its lead completely?
Every tracker still puts ChatGPT first, at roughly 46% to 54% depending on the method. The trend is down, but no current data points to it losing the top spot soon.
Disclaimer: This article is for information only. Market data comes from third-party sources and may change quickly. Nothing here is financial or investment advice.