Does OpenAI still have a moat?
Every funding round, every investor, and what the labs promised in return, in one document. Have a look ↓
Summary
Yes, OpenAI still has a moat, but by October 2026 it sits almost entirely in ChatGPT’s distribution, paid audience and growing workflow layer. It no longer rests on a lead in model intelligence.
The model lead is gone. As of October 20, 2026, Google’s Gemini 4 Argon tops Arena’s overall text ranking, Anthropic’s Claude Opus 5.5 leads Artificial Analysis and the coding and agent rankings, and Meta is back near the top with Muse Wonder. OpenAI’s GPT-6 is competitive but leads none of these boards.
ChatGPT is the asset rivals have not reproduced. OpenAI reported more than 1.2 billion weekly users in early October 2026. That audience lets it launch agents, ads, shopping and work tools without buying users each time.
That audience is habitual, not locked in. Similarweb estimates ChatGPT’s share of visits to major chatbot websites fell from 78% to 57% in the year to August 2026, while Claude rose from 2% to 9.6%. ChatGPT has stabilized since spring, but it is now a leader, not a monopoly.
The raw API is becoming a commodity. Open-weight models carried 56% of tokens on Vercel’s AI Gateway in August 2026 but only 14% of spending. Volume is moving to cheap models; the money stays with the best closed ones, and Anthropic takes most of it.
Anthropic has shown that a rival can win without OpenAI’s consumer reach. It reported $16.8 billion in Q3 2026 revenue to investors, and its Q2 revenue of $11.6 billion was well above OpenAI’s reported $6.7 billion, although the two firms count cloud-partner revenue differently.
Codex is OpenAI’s best route to deeper lock-in, but it is still the challenger. JetBrains found 39% of professional developers using Claude Code at work in mid-2026, against 16% for Codex.
Google is the clearest structural threat. It owns the places where context already lives, and it launched Gemini 4 Argon in October 2026 at an introductory $2 per million input tokens.
Compute is both a barrier and a liability. OpenAI expects about $750 billion of computing spending through 2030, according to the Wall Street Journal, roughly 15 times its reported annualized run rate of about $50 billion in September 2026. Anthropic’s reported commitments are now of a similar size, so scale alone no longer sets OpenAI apart.
Profitability is the open question. OpenAI has not published a break-even date, and it is reportedly seeking at least $30 billion at about a $1.4 trillion pre-money valuation, a round that has not yet closed.
OpenAI’s moat looks real but unfinished. It becomes durable only if ChatGPT turns its reach into recurring work that is hard to replace and cheap enough to run at a profit.
Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us
We keep an up-to-date report on frontier AI labs. Have a look ↓
Sample pages from an earlier edition. The report has been updated with new data since.
What would an OpenAI moat actually mean now?
Today, OpenAI’s moat comes from the people, habits and paid work built around ChatGPT, not from its models.
A benchmark lead does not last long enough to protect anyone. In October 2026 alone, Google and Anthropic each took first place on a major leaderboard, Meta came second on two, and OpenAI led none. Different models win on coding, research, speed or price. A short lead may help sales for a quarter, but it gives customers little reason to stay for years.
The harder assets to copy sit above the model. ChatGPT has more than 1.2 billion weekly users, more than 50 million paying consumers, millions of business seats and products that remember context, connect to company systems and complete longer jobs. Those assets can survive a period when another lab has the better model, which is exactly the period OpenAI is in now.
A moat also has to pay. OpenAI could stay hugely influential while struggling to cover training, inference, infrastructure and partner payments. The real test is whether its products can fund all of that without a larger funding round every year.
Why does OpenAI’s moat look weaker in 2026?
OpenAI’s moat looks weaker because top-level intelligence is now available from at least four labs, and buyers have learned to switch.
GPT-4 gave OpenAI a clear period when serious alternatives were scarce. In October 2026, Anthropic leads coding and agent work, Google tops the overall text ranking with a model built into products people already use, Meta is back at the frontier, and Chinese open-weight models come close at a fraction of the price.
Buyers have adapted. In a September 2026 Stacklok survey, about 92% of sampled North American enterprises used more than one model provider, and 25% used five or more. Consumers keep ChatGPT open while trying Gemini or Claude. Being first is worth less when using several providers is the norm.
So far, OpenAI has lost exclusivity without losing relevance. Menlo Ventures estimated that its share of enterprise model spending fell from 50% in 2023 to 27% in 2025, yet OpenAI’s reported annualized run rate still rose to about $50 billion by September 2026. The old moat around the model has narrowed while a new one forms around distribution and workflow.
If you want more recent data on this point, see the latest figures in our Frontier AI Labs report ↓.
Has OpenAI lost its lead on AI models?
Yes. OpenAI is still one of the best model makers, but in October 2026 it does not lead any of the main independent rankings.
On October 20, 2026, Artificial Analysis placed Anthropic’s Claude Opus 5.5 first on its broad intelligence index at 58, with Meta’s Muse Wonder Full at 56. OpenAI’s GPT-6 Astra scored 53, level with Google’s Gemini 4 Argon. On Arena’s overall text ranking, Gemini 4 Argon is first on a preliminary basis, ahead of Muse Wonder and Claude Opus 5.5. The rankings disagree on who is best, but they agree that OpenAI is not.
The gap is small, and GPT-6 can still win work. It ranks second behind Claude Opus 5.5 on Arena’s web-development board and on Agent Arena. OpenAI also covers more price points than before: GPT-6.1 Sol, released at DevDay on September 29, 2026, lists at $2 per million input tokens and $10 per million output tokens. Its top model, GPT-6 Astra, costs $10 and $50, more than twice Claude Opus 5.5’s $4 and $20.
What has changed is the meaning of a lead. OpenAI’s alumni helped build Anthropic, Safe Superintelligence and other rivals, so know-how does not stay inside one lab. OpenAI can keep returning to the frontier with its compute and its audience, but it can no longer charge a premium simply for being there.
| Current comparison | OpenAI’s position | What it tells us |
|---|---|---|
| Artificial Analysis broad intelligence index | GPT-6 Astra at 53, five points behind Claude Opus 5.5 | The gap is small but now runs against OpenAI |
| Arena web-development and agent rankings | Second behind Claude Opus 5.5 | OpenAI can still win important workloads |
| API list price | $10 / $50 per million tokens for GPT-6 Astra; $2 / $10 for GPT-6.1 Sol | OpenAI’s top model costs more than rivals’ top models |
| Talent diffusion | Many major rivals employ OpenAI alumni | Research knowledge does not stay locked inside one company |
These pages are excerpts from our report. See the rest ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Are open models turning GPT into a commodity?
For high-volume work, yes. Open models now carry most of the tokens on some developer platforms, but the money still goes to closed models.
Vercel’s AI Gateway shows the split clearly. In August 2026, open-weight models processed 56% of its tokens, up from under 10% in December 2025, but took only 14% of spending. Anthropic alone took 64% of spending. On OpenRouter, DeepSeek had the largest share of text requests in the week beginning October 12, 2026, at 24.4%, ahead of Google at 19.7% and OpenAI at 18.5%.
Price drives that shift. DeepSeek’s V4.1 Flash costs $0.15 per million input tokens off-peak in October 2026, and Z.ai’s GLM-5.3 Flash costs about the same. These are cheaper tiers, not equals of GPT-6 Astra, but for extraction, support, classification or bulk coding, a slightly worse model at a fraction of the price is often the better choice. Epoch AI estimated in May 2026 that open-weight models trail the leading closed models by about four months.
The open field is also getting stronger and easier to buy. Meta released Muse Spark Open, an 816-billion-parameter open-weight model, on October 15, 2026. Moonshot AI reached an approximately $50 billion valuation in October, according to Bloomberg, and AWS made its Kimi K3 generally available on Amazon Bedrock in September. Microsoft Foundry lists DeepSeek, Kimi and GLM models too.
Large companies remain more cautious. Menlo estimated that open models held only 11% of enterprise usage in 2025, and no newer survey of that kind was found. China also proposed export controls on advanced AI models on October 16, 2026, which could complicate use of some Chinese models. The conclusion for OpenAI: ordinary intelligence is now a commodity, so its value has to come from the best models and the products around them.
If you want more recent data on this point, see the latest figures in our Frontier AI Labs report ↓.
Is ChatGPT still the default AI assistant?
Yes. ChatGPT is still the default AI assistant in October 2026, and this remains OpenAI’s strongest moat.
OpenAI reported more than 1.2 billion weekly ChatGPT users in early October 2026, up from more than 900 million in March. Its latest published subscriber count, from March 31, 2026, was more than 50 million paying consumers. Google reported more than 950 million monthly users for the Gemini app in July 2026, but that is a monthly figure against ChatGPT’s weekly one. No standalone rival has built a larger habit around asking an AI for help.
That reach turns directly into revenue. ChatGPT Ads reached a $1 billion annualized rate in less than 200 days, OpenAI said on August 31, 2026. New products land in an app people already open. A rival has to persuade them to download something else, rebuild context and pay again: Meta’s new Muse app had about 2.5 million daily users in mid-October 2026, according to Sensor Tower estimates.
ChatGPT’s lead has narrowed, then steadied. Similarweb estimated that its share of visits to major chatbot websites fell from 78% in August 2025 to 57% in August 2026. A separate Similarweb series shows ChatGPT rising again from 53.7% in April to 56.4% in September 2026, while Gemini slipped for four months. The picture is a clear leader that has stopped losing ground for now, not one that owns the market.
| Consumer measure | Current picture | Why it matters |
|---|---|---|
| ChatGPT weekly active users | More than 1.2 billion (October 2026) | OpenAI reaches people at extraordinary frequency |
| ChatGPT consumer subscribers | More than 50 million (latest count, March 2026) | A huge audience already pays directly |
| ChatGPT Ads | $1 billion annualized (August 2026) | The audience can be monetized beyond subscriptions |
| ChatGPT share of chatbot web visits | 57% in August 2026, down from 78% a year earlier | Rivals are growing without shrinking ChatGPT’s absolute audience |
Prefer verified numbers to headlines? Our report is here ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Are ChatGPT users loyal or just used to it?
Mostly used to it. People keep coming back to ChatGPT, but they can still leave easily, and rivals now hold on to the users they win.
Moving a simple prompt takes seconds. People can ask ChatGPT, Claude and Gemini the same question and keep the best answer. There is no social graph to abandon and usually no data to migrate.
ChatGPT still keeps users better than its closest rival, but the gap is closing. Similarweb estimates that 59% of ChatGPT users were still active after six months, against 45% for Claude’s February 2026 cohort. A year earlier, Claude kept only 25.8%. Habit is no longer something only ChatGPT can build.
Memory, projects and connected tools raise the cost of leaving. A user who has stored preferences, uploaded project files and connected email loses more than a chat box by switching. The friction grows when an agent can continue a project and act across several apps.
OpenAI’s weak point is where context lives. Much of it sits in Google Workspace, Microsoft 365, Slack, GitHub or company databases, which Google and Microsoft reach from products they already control. Loyalty will hold only if ChatGPT becomes the place where work actually happens.
Is Google becoming OpenAI’s biggest threat?
Yes. Google is OpenAI’s biggest threat because it now matches OpenAI on models while owning far more distribution.
The model gap has closed. Google made Gemini 4 Argon broadly available on October 13, 2026, and it took first place on Arena’s overall text ranking, though the lead is preliminary and within the margin of error. Artificial Analysis scores it level with GPT-6 Astra. Its introductory price of $2 per million input tokens and $10 per million output tokens is one fifth of GPT-6 Astra’s; it is due to rise to $4 and $20 later, still well below OpenAI’s top model.
Distribution is the bigger issue. The Gemini app passed 950 million monthly users in July 2026, and Gemini also appears inside Android, Search, Chrome, Gmail, Docs and Cloud. A good-enough assistant built into all of those products can win through convenience. Google can also fund AI from advertising and cloud revenue, while OpenAI has to make the AI product itself pay.
Google is not winning everywhere. Its share of chatbot web traffic fell for four straight months to September 2026, and its share of tokens on Vercel’s gateway dropped sharply as developers moved to open models. OpenAI still has the stronger AI-first brand. But if assistants become background features inside existing software, Google gains and OpenAI’s standalone app matters less.
If you want more recent data on this point, see the latest figures in our Frontier AI Labs report ↓.
Want to know who is really ahead? Our report is here ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Is OpenAI losing developers to rivals?
Partly. OpenAI still has huge developer usage, but it no longer dominates, and developers now spread work across many providers.
OpenAI has reported that its APIs process more than 15 billion tokens per minute, and its platform covers tool use, voice, agents and evaluations. Years of early adoption left real switching costs: prompts, tests, guardrails and monitoring shaped around OpenAI models. In Stacklok’s September 2026 survey, 57% of enterprises estimated that moving to another provider would take four months or longer.
But developer share has fragmented. On OpenRouter, OpenAI handled 18.5% of text requests in the week beginning October 12, 2026, behind DeepSeek and Google. On Vercel’s gateway, Anthropic took 64% of spending in August. OpenAI does hold its own at the top end: GPT-6 Astra took 7.7% of Vercel’s gateway spending in its first twelve days in September.
The tools also make switching easier. GitHub Copilot offers both Claude and Codex, and Microsoft Foundry lists Chinese open models. To keep developers, OpenAI has to make the workflow around the model harder to leave: evals, permissions, persistent agents and deployment controls. The raw API is too easy to swap.
Has Anthropic overtaken OpenAI in enterprise AI?
Yes, on spending and revenue, Anthropic now leads, especially in coding. OpenAI still reaches more companies.
Menlo Ventures estimated that Anthropic captured 40% of enterprise model spending in 2025, against 27% for OpenAI and 21% for Google, and 54% of coding-model spending against 21%. No newer survey of that kind was found by October 2026, but revenue points the same way. Anthropic’s reported Q2 2026 revenue was $11.6 billion against $6.7 billion for OpenAI, and Anthropic told investors in October that Q3 revenue reached $16.8 billion, up about 45%. The comparison is not perfectly clean: OpenAI reportedly books cloud-partner sales net, while Anthropic books some gross.
OpenAI’s answer is breadth. More than 2.5 million businesses use its products, OpenAI said in September 2026, and more than nine million people used paid ChatGPT business seats as of March. Enterprise produced over 40% of OpenAI’s revenue. Most large companies use both: Barclays plans to give Claude Code to half its developers by the end of 2026, while Atlassian expanded its OpenAI deal in October 2026 to run its Rovo agents on GPT-6.
OpenAI is also moving into implementation. Presence, launched in July 2026, packages agents, policies, evaluations and human escalation into a managed deployment. OpenAI says the version running its own English-language phone support resolves 75% of inbound issues without a person; BBVA, SoftBank and IAG are testing it. If that works at scale, OpenAI sells a working business process, not just a model.
| Enterprise measure | OpenAI | Anthropic | What it means |
|---|---|---|---|
| Estimated model-spend share (2025) | 27% | 40% | Anthropic leads the model layer in Menlo’s survey |
| Estimated coding-model share (2025) | 21% | 54% | Anthropic has a clear specialist advantage |
| Reported Q2 2026 revenue | $6.7 billion | $11.6 billion | Anthropic is ahead, though the firms count partner revenue differently |
| Business reach | More than 2.5 million businesses using its products (September 2026) | More than 300,000 business customers (September 2025) | OpenAI has much broader organizational reach |
| Paid business users | More than 9 million (March 2026) | No comparable public figure | ChatGPT has spread far beyond technical teams |
We made a report out of all this. Take a look ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Can Codex catch Claude Code and become a moat?
Codex is OpenAI’s best chance to own completed work rather than model access, but it is still behind Claude Code with professional developers.
Codex is growing fast. OpenAI reported more than five million weekly Codex users on June 2, 2026. Later company figures combine products: 8 million active users across Codex and ChatGPT Work in July, and 40 million across Codex and ChatGPT in early October, with no time window given. Those numbers show momentum but cannot be compared directly. Codex also produced 64% of output tokens across Codex and ChatGPT in OpenAI’s enterprise data in June.
Claude Code still leads where it counts. JetBrains surveyed more than 15,000 professional developers in May–July 2026: 39% used Claude Code at work and 31% called it their main tool, against 16% for Codex. Codex grew from 3% in January, so the gap is closing, but OpenAI does not report Codex revenue, while Claude Code disclosed a run-rate above $2.5 billion as early as February 2026.
The prize is larger than coding. Codex already handles reports, spreadsheets and research, and a workspace that understands files, runs parallel tasks and keeps project history is much harder to replace than a chatbot. It becomes a moat when skills, permissions, reviews and integrations pile up around it. Fast adoption is proven; durable lock-in is not yet.
Does ChatGPT’s scale give OpenAI a data moat?
It gives OpenAI a product-feedback advantage, but October 2026 shows that this does not guarantee the best model.
More than a billion weekly users expose ChatGPT to rare questions, new languages and unusual workflows. Even when conversations are not used for training, OpenAI can learn from behavior: where users retry, abandon a task or correct an answer.
OpenAI’s privacy-protected studies cover millions of conversations across age groups, countries and types of work. Few rivals can watch general-purpose AI use at that scale, which helps OpenAI find failures early and test features on a huge audience.
The enterprise side is more restricted. OpenAI does not train on Business, Enterprise or API customer data by default. That promise wins sensitive workloads, but it means corporate deployments do not feed company knowledge back into the base model.
The limit is now visible. Anthropic leads coding and agent rankings with a much smaller consumer audience, largely by learning from developers and enterprise work. ChatGPT’s data helps OpenAI build better consumer products; it has not kept GPT-6 ahead.
Is OpenAI’s compute empire a moat or a trap?
Both. OpenAI’s compute buildout protects its ability to serve demand, but it also creates the biggest financial risk in the business.
OpenAI has signed capacity deals with Microsoft, Oracle, Amazon, CoreWeave, Nvidia, AMD, Broadcom and Cerebras. In August 2026, Nvidia backed an Ohio campus of up to 8 gigawatts with OpenAI as the planned tenant on a 20-year lease, and the $122 billion March round was completed when Nvidia and SoftBank paid their final installments on October 1. No verified, deduplicated total of OpenAI’s gigawatts exists, because many of these deals overlap.
The bill is huge. The Wall Street Journal reported in July 2026 that OpenAI expects about $750 billion of computing spending through 2030, up from about $600 billion, and repeated the figure in October. That is roughly 15 times its reported annualized run rate of about $50 billion in September 2026. Execution is a concern too: Sachin Katti, OpenAI’s computing-capacity chief, is reportedly leaving in November.
Compute is also less exclusive than before. Anthropic’s draft IPO prospectus reportedly lists more than $600 billion of noncancelable infrastructure commitments, and Meta says its Prometheus cluster is heading toward about one million GPUs. When rivals buy capacity on the same scale, compute stops being a moat and becomes a cost of staying in the race.
Scale still lowers costs when utilization is high. Partners such as Oracle and SB Energy build and finance many sites, but OpenAI carries the long-term payments. The economics turn bad if capacity sits idle or prices fall faster than usage grows, and Gemini 4 Argon’s pricing shows prices are still falling.
If you want more recent data on this point, see the latest figures in our Frontier AI Labs report ↓.
If this is useful, the full report goes much further. Have a look ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Does Microsoft protect OpenAI or hold it back?
Microsoft still helps OpenAI more than it constrains it, but it is no longer a reliable ally against Anthropic.
The April 27, 2026 amendment gave OpenAI more room. Microsoft’s license runs through 2032 on a non-exclusive basis, Azure stays the primary cloud, and OpenAI can sell on any cloud. That freedom explains its large capacity deals with Oracle, CoreWeave and Amazon, each worth tens of billions of dollars or more in announced value.
OpenAI still pays for the relationship. It keeps sharing revenue with Microsoft until 2030, subject to a total cap, at a rate the Financial Times reported at around 20%. The payments no longer depend on progress toward AGI, and Microsoft’s own revenue share to OpenAI was eliminated. Microsoft held about 27% of OpenAI after the October 2025 recapitalization; no updated figure has been published.
Microsoft is also hedging. It made Claude the default in Copilot for Excel and PowerPoint in May 2026, switched Microsoft 365 Copilot’s preferred model back to OpenAI’s GPT-5.6 in July, and since September 30 has enabled both GPT-6.1 Sol and Claude Sonnet 5.5 by default for eligible users. For OpenAI, that means its biggest partner now treats it as one supplier among several, inside products where Microsoft owns the customer.
Can OpenAI turn its scale into durable profits?
Not yet proven. OpenAI has shown that demand is enormous, but its costs are still growing faster than its proof of profit.
Revenue is rising fast. Audited figures reviewed by the Financial Times showed $13.07 billion in 2025. Reported quarterly revenue reached $5.7 billion in Q1 2026, and Reuters reported about $6.7 billion for Q2. The annualized run rate is a different measure: it takes the latest month and multiplies it by 12. OpenAI told investors its run rate was almost $50 billion in September 2026, Reuters and the Financial Times reported. That implies very fast growth since Q2, and it rests on reported, unaudited figures. OpenAI also has a reported target of about $70 billion in annualized revenue by December 2026.
Costs remain heavy. In Q1 2026, OpenAI reportedly had an operating loss of about $9.3 billion, a gross margin of about 39% and cash burn of $3.7 billion. The Financial Times reported projections of about $280 billion of cumulative negative free cash flow through 2030. No verified break-even year has been published.
There are real levers. Ads reached a $1 billion run-rate in August 2026, cheaper models such as GPT-6.1 Sol can lift margins, and Presence and Codex price around outcomes rather than tokens. Anthropic shows it can be done: it reportedly posted an 11% adjusted operating margin in Q3 2026, though it also reported an $80.7 billion first-half net loss, mostly from financing accounting.
For now, outside capital fills the gap. OpenAI deferred its IPO to 2027 at the earliest and is reportedly seeking at least $30 billion at about a $1.4 trillion pre-money valuation, against $852 billion in its last completed round; the higher figure is a proposal, not an achieved price. The moat becomes economic only when cost per useful task falls faster than the price customers pay.
Does OpenAI still have a moat?
Yes. OpenAI still has a real moat in October 2026, but ChatGPT is carrying almost all of it.
The model advantage has gone. GPT-6 is excellent and wins some tasks, yet Claude Opus 5.5, Gemini 4 Argon and Meta’s Muse Wonder rank at or above it on the main independent boards, and open models sell close substitutes for a fraction of the price. Judged on models alone, OpenAI has no moat.
ChatGPT tells a different story. Its 1.2 billion weekly users, paid audience, ads business and enterprise footprint remain unmatched, and Codex and Presence are pulling OpenAI into recurring work. Rivals can copy model features faster than they can rebuild that distribution.
The weak spots are real. Users keep several assistants. Developers route between APIs. Google bundles Gemini across an ecosystem OpenAI does not own, and Microsoft now sells Claude next to GPT. Anthropic leads enterprise coding and reports more quarterly revenue. And OpenAI’s compute bill is large enough to overwhelm even very fast revenue growth.
OpenAI’s moat has moved from models to products. It should stay one of the leaders for years if ChatGPT becomes the place where people finish recurring work. Whether that moat lasts depends on making those workflows hard to replace and profitable to run.
| Part of OpenAI’s position | Moat strength now | What it means |
|---|---|---|
| Frontier models | Weak to moderate | Competitive, but no longer leading the main independent rankings |
| ChatGPT distribution | Strong | OpenAI’s clearest defensive asset |
| Consumer habit and memory | Moderate | Useful friction, but users can still multi-home easily |
| Enterprise reach | Strong | Broad adoption gives OpenAI many routes into companies |
| Enterprise model share | Weak to moderate | Anthropic leads high-value workloads and reported revenue |
| Codex and workflow products | Rising | The best route toward deeper switching costs |
| Developer API | Moderate | Large scale, but increasingly interchangeable |
| Compute access | Strong but risky | Hard for most to copy, expensive to carry, and matched by Anthropic |
| Pricing power | Weak | Top GPT-6 pricing sits well above Gemini 4 Argon and open models |
| Economic moat | Unproven | Demand is real; durable margins are still the open question |
If you want more recent data on this point, see the latest figures in our Frontier AI Labs report ↓.
Want the whole picture? Check our report ↓
Sample pages from an earlier edition. The report has been updated with new data since.
Our methodology
This analysis tests whether OpenAI still has a durable moat based on the evidence available as of late October 2026. We separated the question into model performance, consumer distribution, user behavior, enterprise adoption, developer usage, workflow integration, data advantages, infrastructure and economics.
We treated benchmark leadership as a temporary advantage rather than proof of a moat. Frontier rankings can change after one release, so we used independent model, web-development and agent evaluations from Artificial Analysis and Arena, along with API list prices, to judge whether OpenAI remains competitive, not whether it owns an enduring technical lead.
For distribution, we looked at weekly users, subscribers, web-traffic share, retention, paying businesses and paid business seats. ChatGPT and Gemini publish different usage measures, so we used those figures to understand scale and direction rather than present them as a perfectly matched market-share comparison.
We used Menlo Ventures’ enterprise estimates as the clearest available view of how model spending is splitting across OpenAI, Anthropic, Google and open models. Those figures are market estimates rather than audited vendor revenue, so they help compare competitive position more than exact company performance.
We assessed open models separately from full enterprise platforms. Price, control and developer adoption show how quickly ordinary model intelligence is becoming substitutable, while support, security, governance and reliability explain why the same models may spread more slowly inside large companies.
We treated memory, projects, integrations, evaluation systems, permissions and recurring agent workflows as stronger evidence of switching costs than a model API alone. The aim was to separate products that people can replace in minutes from operating layers that become embedded in a team’s work.
Compute commitments were considered both an access advantage and a financial obligation. Planned power, chips and cloud capacity can protect availability and lower unit costs at high utilization, but commitments are not the same as profitable demand or capital already converted into productive infrastructure.
For profitability, we focused on whether recurring revenue can cover inference, research, staff, sales, partner payments and OpenAI’s infrastructure obligations. Fast revenue growth is useful evidence of demand, but it does not establish an economic moat unless cost per useful task falls fast enough to produce durable operating cash flow.
We prioritized recent company disclosures, independent benchmarks, audited financial reporting and market research that added a checkable metric. Key sources include OpenAI’s earlier GPT-5.6 launch material, Let’s Data Science on the GPT-6 rollout in ChatGPT, Artificial Analysis’ Intelligence Index, Artificial Analysis’ Coding Agent Index, Menlo Ventures’ enterprise AI report, Sensor Tower’s AI app research, Alphabet investor reporting, Financial Times reporting on OpenAI’s audited figures, Similarweb on ChatGPT and Claude retention, Vercel’s AI Gateway production index, JetBrains Research on AI coding-agent adoption, TechCrunch on Gemini app users, OpenAI on the next phase of its Microsoft partnership, Microsoft’s Copilot default-model notice, Implicator on OpenAI’s revenue run-rate, and The Wall Street Journal’s reporting on compute commitments.
Additional product and infrastructure sources include OpenAI on Codex, OpenAI on Presence, OpenAI’s business-data controls, OpenAI on Stargate, the OpenAI developer platform, OpenRouter usage and rankings, and Moonshot AI’s Kimi product information.
Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us
