Who’s really winning the AI race: OpenAI, Anthropic, or Google?

ChatGPT made OpenAI the face of modern AI. Close to a billion people use it every week, and for many, it’s synonymous with artificial intelligence itself. But in the part of the race that actually pays the bills, OpenAI is no longer in front. Anthropic, the company behind Claude, now brings in more revenue, and Google may be best positioned to win the long game.

To understand why, you have to stop looking at user counts and start looking at the economics behind AI: who pays, who can leave, and who controls distribution.

The old software rule AI just broke

Traditional software has been a dream business. Once you build the product, every extra user costs almost nothing to serve. That means growth quickly turns into profit.

AI breaks this rule. Every answer from a large language model burns real computing power. That means every interaction has a direct cost. When most of your users are on free plans, they’re not assets – they’re expenses.

This is why raw user numbers are now misleading. The company with the most users might be holding the biggest bill, not the biggest advantage.

OpenAI’s consumer problem: a billion users, few customers

OpenAI dominates consumer mindshare. ChatGPT has close to a billion weekly users, and it still earns most of its revenue from individuals paying for ChatGPT subscriptions.

But only about 1 in 20 users pays anything at all. On top of that, OpenAI has been nudging users off the $20/month plan toward cheaper tiers that include ads. It’s also building shopping into the chat experience and quietly sharing limited ad-targeting data with marketing partners.

The strategy is clear: turn a massive free user base into a real business. The challenge is that ads in chatbots are unproven, and switching costs are tiny. If a better or cheaper option appears, most people can leave with a single tap.

OpenAI’s consumer lead is also under pressure on two fronts:

• Google’s Gemini: ChatGPT’s share of chatbot traffic reportedly fell from about 90% to roughly 55% in a year, with most of that loss going to Google’s Gemini. Not because Gemini is dramatically better, but because Google can push it into Search, Gmail, and Android for billions of users who never went looking for it.

• Trust and backlash: When Anthropic publicly refused a broad U.S. government request for unrestricted access to Claude (including for mass surveillance and autonomous weapons), it took a short-term hit in contracts but a long-term win in public trust. ChatGPT uninstalls in the U.S. reportedly spiked, and Claude passed ChatGPT in downloads for the first time.

Meanwhile, OpenAI is burning large amounts of cash, with reported multi-billion-dollar quarterly losses and huge data center commitments to fund. That’s why it’s racing to push coding and business tools into ChatGPT and chase enterprise customers – the segment where the real money sits.

Why Anthropic’s smaller audience makes more money

In the enterprise market, the picture flips. Claude is now the top choice among many businesses, and this is where Anthropic’s strategy shines.

Consumer users can churn overnight. Businesses can’t. Once a company wires an AI model into its workflows, products, and internal tools, ripping it out becomes a months-long project that nobody wants to fund or manage.

That makes enterprise revenue “sticky” – it compounds over time as companies expand their usage. Anthropic doesn’t have the biggest crowd, but it has some of the stickiest revenue in the industry.

Here’s how the shift played out:

• Enterprise share flipped: In 2023, OpenAI reportedly had about half of all enterprise AI spending, with Anthropic around a tenth. Today, surveys of enterprise buyers and real spend data from tens of thousands of businesses show Anthropic in front.

• Revenue lead: Anthropic’s revenue is now running at around $47 billion annually, up from roughly $4 billion a year earlier. OpenAI sits closer to $25 billion. The “runner-up” in public perception is earning nearly double the revenue of the perceived leader.

• Profit trajectory: Anthropic is approaching its first profitable quarter, which would make it the first of the frontier AI labs to actually turn a profit. OpenAI, by contrast, is still burning cash.

About 80% of Anthropic’s revenue comes from businesses and developers, not casual chat users. Fewer users, each worth far more, on contracts that renew and grow. That’s the difference between a cost-heavy consumer play and a high-value enterprise strategy.

How Anthropic pulled ahead: code, trust, and focus

Anthropic’s lead in enterprise AI comes down to three main factors.

1. Winning over developers with code

Claude has built a reputation for handling long, complex programming tasks particularly well. Its dedicated tool for engineers, Claude Code, went from almost nothing to billions in usage in under a year. Today, an estimated 4% of all public code pushed to GitHub is written with it.

There’s a quiet rule in enterprise software: when the engineers choose a tool, the rest of the company usually follows. If your developers are building with Claude, your business quickly becomes a Claude customer by default.

2. Turning trust into a business advantage

For high-value customers like banks, hospitals, and government agencies, trust is not a nice-to-have. It’s a requirement. They can’t risk handing sensitive data to a vendor they don’t fully trust.

Anthropic’s public decision to walk away from a large defense contract rather than compromise its policies signaled something powerful to these customers: this is a company that will protect data even when it’s expensive to do so.

That kind of stance can cost money in the short term, but it wins long-term enterprise deals – especially with organizations that think in decades, not quarters.

3. Skipping the viral app race

Anthropic never chased a viral consumer app as its main goal. It went straight for long-term contracts and deep integrations. Today, eight of the ten largest companies in the U.S. are reportedly Anthropic customers.

If you’re deciding which AI platform to build your company on, this is the key question: how painful would it be to leave? In the enterprise segment, the winner isn’t always the smartest model – it’s the one your organization can’t afford to rip out.

If you’re exploring how to bring this kind of AI into your own organization, it’s worth pairing this perspective with practical guidance like how to use AI in your business in 2026 without turning into a full-blown tech company.

The China shock: when AI intelligence got cheap

While U.S. labs were fighting for model quality and market share, a different kind of shock came from China.

In 2025, a relatively unknown Chinese lab called DeepSeek released a model that matched the performance of top U.S. models – and claimed it had trained it for only a few million dollars, while American giants were spending billions.

The announcement triggered a brief panic in markets, wiping hundreds of billions off Nvidia’s value in a day. DeepSeek itself faded from public attention, but the real impact was deeper: it reset expectations about how cheap it could become to train powerful models.

Today, free and open-source models – many of them Chinese – account for roughly a third of all usage on major AI marketplaces developers rely on. This isn’t mainly a security story; it’s a price story.

As the cost of raw “intelligence” falls and models start to look more alike, being the absolute smartest model for a few months matters less. If your entire business is just “we have a model,” a free alternative is a direct threat. If your business is distribution, trust, or services, those free models become tools you can weaponize in your favor.

The three questions that actually matter

In a world where models are getting cheaper and more similar, the advantage shifts to things that don’t commoditize as easily:

• Who controls how the AI reaches people?

• How hard is it for the best customers to leave?

• Does it get cheaper or more expensive to run as it grows?

Run OpenAI through those questions and you see exposure: rented distribution (via partners like Microsoft and Apple), low switching costs for most users, and high variable costs for every extra answer.

Run Anthropic through them and you see a moat: deep enterprise integrations, high switching costs, and revenue concentrated in customers who plan years ahead.

But there’s a third player that scores even higher on one of those questions: Google.

Why Google may win the decade

Google’s Gemini doesn’t have to be the best model to matter. What Google has is distribution on a scale nobody else can match.

Google can embed AI into products people already use every day: Search, Gmail, Docs, YouTube, Android, and now even parts of the Apple ecosystem. It can sell AI to people who never went looking for it, as a feature inside tools companies already pay for.

Some key advantages:

• Default status: Gemini can be the default assistant on over 2 billion Android devices, and it’s now moving onto iPhones too.

• Apple partnership: At a recent developer conference, Apple rebuilt Siri on top of Google’s Gemini and dropped the ChatGPT handoff it had been using, in a deal reportedly worth around $1 billion per year. That means many users who think they’re using “AI” in general will actually be using Google’s models by default.

• Bundled value: For businesses, Gemini can be bundled into Google Workspace and cloud services they already pay for, making AI adoption feel like an upgrade rather than a new, separate line item.

When models become commodities, distribution beats brilliance. On that dimension, nobody is close to Google.

The spending bubble: when the bill comes due

All of this is happening against a backdrop of enormous spending. The four biggest tech companies alone are on track to spend around $725 billion on AI infrastructure this year. That’s pushing their free cash flow to the lowest levels in a decade.

By the end of the decade, the industry could be roughly $800 billion short of the new revenue it would need to fully justify that level of investment. Both OpenAI and Anthropic are heading toward public markets under this pressure.

• Anthropic is filing for an IPO with revenue surging and profitability in sight.

• OpenAI is filing while reportedly missing some of its own user and revenue targets, and with warnings from its finance chief about funding the data centers it has already committed to.

Going public turns optimistic stories into audited numbers. One company arrives close to profitable; the other arrives still burning large amounts of cash.

But even the current leader isn’t safe. If enterprises decide they’re not seeing enough return on their AI investments and pull back, the very segment that powers Anthropic’s rise is the first to feel it. No one in this race is immune.

For founders and executives, this is a reminder that the most profitable part of AI today often isn’t the model itself, but the services, integration, and workflow redesign around it – a theme explored in depth in why the most profitable AI business isn’t software, it’s services.

So who’s really winning?

The answer depends on the time frame you care about.

• Right now, on the scoreboard that pays the bills: Anthropic is ahead. It earns more revenue than OpenAI, is closer to profitability, and has deeply embedded itself in the enterprise.

• Over the next decade, on reach and distribution: Google is best positioned. It can ship AI to billions of people who never ask for it, across devices and apps they already use.

• The one with the most to lose: OpenAI. It has the most famous brand, a massive free user base, and huge infrastructure commitments – but less control over distribution and a business model still in flux.

The bigger shift is this: the crowd and the money have come apart. User counts no longer tell you who’s winning. Economics do.

The next time you see a headline crowning a new AI winner, run three questions:

• Who controls how this AI reaches people?

• Could its best customers leave tomorrow without much pain?

• Does it get cheaper or more expensive to run as it grows?

The first real reckoning in AI won’t be about who built the smartest model. It will be about who can afford to run the ones they’ve already promised – and who can turn that intelligence into durable, profitable businesses.