ChatGPT just lost its majority. Here's who's winning what's left.
For three and a half years, ChatGPT owned more than half of every AI conversation on earth. That ended in March 2026. By May, its share had settled at 46.4% — the first sub-50% reading in the platform's history, according to Sensor Tower's State of AI 2026 report. The market didn't shrink. It fragmented. Gemini now holds 27.7%. Claude holds 10.3% — with the highest subscription conversion rate of any platform at 13%. And Fable 5 is still offline on day nine. The race is no longer about who has the most users. It's about who has the most valuable ones.
LEAD STORY
ChatGPT hit a billion monthly users faster than any app in history. TikTok took nine years. YouTube took eight. Instagram took six. ChatGPT did it in roughly three. And in the same month it crossed 1.1 billion monthly users, it lost its majority market share for the first time.
That paradox — a platform that is simultaneously breaking growth records and losing competitive ground — is the defining story of AI in mid-2026.
Sensor Tower's data shows ChatGPT held 65.3% of global AI assistant users in December 2024, fell to 52.8% by December 2025, and reached 46.4% by May 2026 — a drop of nearly 19 percentage points in 18 months. The crossing below 50% actually happened in March; the May figure is where share settled after the threshold broke.
Gemini now holds 27.7% with 662 million monthly users, and Claude holds 10.3% with 245 million. Grok, Perplexity, DeepSeek, and Meta AI collectively account for the remaining share — all single-digit players, all growing.
The causes are structural, not cyclical. Three forces are driving the fragmentation.
First, distribution. Gemini's growth is largely driven by its integration with Google's broader ecosystem — Search, Android, Workspace — giving it default placement across billions of devices without requiring a user decision. That's not winning on quality. That's winning on plumbing.
Second, values alignment. ChatGPT uninstalls surged measurably following OpenAI's February agreement with the Department of Defense — suggesting that users are making platform choices based on a company's political and ethical posture, not just its model quality. That's a new competitive variable that didn't exist in 2023.
Third, monetization efficiency. Claude leads all AI platforms in subscription conversion at 13% of its 245 million users paying — the highest rate across all platforms. ChatGPT has 1.1 billion users and is now running ads to 17% of them daily to compensate for conversion rates that trail Claude's. The platform with more users is monetizing less efficiently per user than the platform with fewer.
For OpenAI's IPO at a target valuation of $850 billion to $1 trillion, these numbers create a specific problem. Revenue is real — $20 billion annualized and growing. But the market share trajectory is pointing the wrong direction at the moment institutional investors are being asked to write billion-dollar checks. A company that crosses a billion users in record time while losing 19 percentage points of market share in 18 months is either in the middle of a temporary competitive fluctuation, or at the beginning of a structural decline. The S-1 will have to tell a convincing story about which one it is.
Meanwhile, GPT-5.6 is expected to launch before month-end, targeting agentic coding capabilities where GPT-5.5 has trailed Claude Code and Fable 5. Multiple ChatGPT Pro users are already reporting noticeably faster, more capable responses — suggesting the model is running in limited deployment before the official announcement. A GPT-5.6 launch recaptures benchmark leadership and gives the IPO roadshow a growth narrative to counter the market share data.
THE MONEY ANGLE
1. The real competition is now conversion, not users. AI assistant spending is on pace to reach $4.2 billion in H1 2026 — nearly double the $1.83 billion recorded in H1 2025. The platforms that win this monetization wave are not the ones with the most monthly active users. They are the ones with the highest willingness-to-pay among their users. Claude's 13% subscription conversion rate against ChatGPT's scale means Anthropic is extracting more revenue per user from a smaller base. As the market matures from growth to monetization, that efficiency advantage compounds. The IPO that should be priced on revenue-per-user efficiency rather than raw user count is Anthropic, not OpenAI — and the market hasn't fully priced that distinction yet.
2. Gemini's distribution moat is the most durable advantage in the market. Google didn't win 27.7% market share by building a better model. It won by making Gemini the default AI experience inside products that 3 billion people already use daily. That distribution advantage is structurally harder to replicate than any benchmark improvement. Apple's $1 billion deal to embed Gemini in Apple Intelligence extends that moat to 2 billion more devices. The implication for investors: Google's AI revenue will grow even if Gemini never wins a head-to-head benchmark comparison, because default placement drives usage that model quality cannot override. Alphabet's AI revenue line is the most underpriced in the sector relative to its structural position.
3. The DoD uninstall signal is a preview of what government contracts cost in consumer markets. OpenAI's February DoD deal triggered a measurable spike in ChatGPT uninstalls — the clearest evidence yet that brand trust and values alignment drive platform switching alongside features. That signal has direct implications for every AI company now in negotiation with the US government. Anthropic is working toward a deal that restores Fable 5 and co-develops a capability evaluation framework with the administration. If that deal is announced publicly and framed as a government partnership rather than a compliance resolution, Anthropic risks the same consumer trust erosion that followed OpenAI's DoD announcement. The framing of the Fable 5 resolution matters as much as the resolution itself.
THE OPPORTUNITY
1. Claude's 13% conversion rate is the most important number in AI investing this week. Claude leads all AI platforms in subscription conversion with 13% of its 245 million users paying. At an average subscription price of roughly $20 per month, that's approximately $640 million in monthly subscription revenue from a user base one-fifth the size of ChatGPT's. When Anthropic's S-1 publishes its actual revenue breakdown, the revenue-per-user figure will be the single most important metric for assessing whether the $965 billion valuation is defensible. A platform that converts at 13% in a market where the leader converts at lower rates has pricing power that most analysts are not yet fully modeling. Investor move
2. Multi-platform AI strategy is no longer optional for enterprises. The policy intended to protect American AI superiority may be accelerating Chinese model adoption among the developer community. With Fable 5 offline and GPT-5.5 under scrutiny, GLM-5.2 from China's Zhipu AI leads GPT-5.5 on FrontierSWE benchmarks — and is available without export control risk. Any enterprise that built its AI stack around a single US frontier model now has documented evidence that geopolitical risk can remove that model from service within hours. The practical response: maintain active credentials and tested workflows across at least two providers. The cost of multi-vendor AI infrastructure is trivially low. The cost of a single-vendor outage is now well-established. Business owner move
3. The Agentjacking threat requires immediate action — not next sprint. Security researchers disclosed a new attack class called Agentjacking this week, exploiting Sentry error-tracking to trick AI coding agents into executing malicious code. The attack had an 85% exploitation rate and affected 2,388 organizations. The attack works by injecting malicious instructions into error reports that AI coding agents — Claude Code, Cursor, Codex — interpret as legitimate debugging guidance. If your engineering team uses AI agents connected to error-tracking platforms, add a human review layer between error reports and autonomous agent execution today. There is no universal patch yet. Business owner move | Career move
QUICK HITS
Agentjacking: The Attack That Exploits Developer Trust Security researchers disclosed a new attack class that targets AI coding agents through Sentry, the widely used error-tracking platform. Attackers inject malicious markdown instructions into fake error reports. When Claude Code, Cursor, or Codex reads the report as part of a debugging workflow, it executes the attacker's commands on the developer's machine. The attack had an 85% exploitation rate and has affected 2,388 organizations. The reason it works: developers have specifically trained themselves to trust their coding agents. When Claude Code tells you to run a command, you run it. ◆ Money angle: An 85% exploitation rate across 2,388 organizations is not a niche security event — it is a systematic vulnerability in how AI agents interact with developer tooling. Every company running AI-assisted development pipelines has exposure until a patch ships.
FERC Orders Grid Reform for AI Data Centers The Federal Energy Regulatory Commission issued targeted show-cause orders to six US regional grid operators, directing them to either defend current interconnection frameworks or propose reforms to speed up grid access for large-load customers — explicitly including AI data centers. FERC Chair Laura Swett called it a "national priority." The orders bypass the years-long standard rulemaking process entirely. Microsoft has added over 4 gigawatts of new capacity in the past 18 months. CoreWeave targets 1.7 gigawatts by end of 2026. ◆ Money angle: Regulated grid operators who move fast on interconnection reform capture AI data center revenue that currently sits in multi-year queues. Utilities in Virginia — where the highest concentration of AI infrastructure is being built — are the direct beneficiaries of faster interconnection approvals.
Fable 5: Day Nine As of June 21, the API at claude-fable-5 still returns errors. The June 20 refund processing cutoff passed yesterday. The June 22 free trial window closes tomorrow. Trump's "not a threat" statement from June 20 has not been followed by any official regulatory action. Kalshi holds at 57% odds of restoration before July 1. Gemini 3.5 Pro remains in limited Vertex AI enterprise preview only, with nine days left in Google's self-imposed June general availability window. ◆ Money angle: June 22 closes the last operational deadline in the Fable 5 crisis calendar. If no restoration announcement arrives this weekend, the next catalyst is GPT-5.6's launch — which hands OpenAI the coding benchmark leadership Fable 5 had been holding.
The AI assistant market is no longer a monopoly. It is a three-platform race with a fragmented tail — and the winner will be determined not by who has the most users, but by who extracts the most value from the ones they have.
See you tomorrow, The Future Geek Team
