IPO race: Why AI companies are building a business on habit

There is a pattern to how the internet's most powerful businesses were built, and it has almost nothing to do with technology. Search engines did not win because their algorithms were permanently superior. Social networks did not endure because they had better engineers. They endured because they became habits, woven into the texture of daily life so completely that switching felt less like a product decision and more like a personal disruption. 

Artificial intelligence, it now appears, is learning the same lesson, and learning it faster than any technology before it.

In the span of a single fortnight in June 2026, the AI industry served notice that it has moved well beyond its founding chapter. Anthropic, the maker of Claude, filed a confidential S-1 with the US Securities and Exchange Commission on June 1, days after closing a staggering $65 billion Series H funding round that pushed its valuation to $965 billion, a figure that briefly eclipsed its fiercest rival. 

OpenAI, whose ChatGPT reaches roughly 900 million users every week, followed within days with its own confidential IPO filing, signalling a September 2026 listing window at a valuation hovering between $850 billion and $1 trillion. Perplexity, the AI search disruptor, has confirmed a 2028 IPO timeline, positioning itself as the third act in what is shaping up to be the most consequential public-market test in technology history.

Read more: Open AI’s confidential IPO filing

But to read these filings purely as capital-market events is to miss the more consequential story unfolding beneath the headlines. What the IPO race reveals is a fundamental strategic pivot, one where the contest for AI supremacy is no longer being fought on the model leaderboard. It is being fought inside the daily workflow, the morning browser tab, the code editor left open overnight. The technology, in other words, was always the door. What these companies are now building, with considerable urgency, is the room you never want to leave.

From Benchmarks to Behaviour

Anthropic's revenue trajectory tells a story that no benchmark score can. Its annualised run-rate revenue grew from $9 billion at the end of 2025 to $47 billion by late May 2026, a fivefold surge in under six months, driven primarily by enterprise adoption and its agentic coding tool, Claude Code, which alone crossed $2.5 billion in annualised revenue within months of its public launch. OpenAI, meanwhile, sits on approximately $20 billion in annualised revenue and is burning through capital at a rate analysts estimate at roughly $14 billion this year. The numbers are extraordinary, and yet the most revealing figure in the entire AI economy is far smaller. Only about 5.5 percent of OpenAI's 900 million weekly users actually pay.

Read more on Anthropic’s IPO

Sanjeev Narsipur, Managing Director and Lead for Digital, AI and Technology Services at Alvarez and Marsal, frames the competitive picture with precision. "The moat is shifting from 'best model' to 'most embedded model,'" he says. "How embedded your model is in a day-to-day individual workflow, or business workflow, will define stickiness." 

The shift matters because the traditional defences that technology companies relied upon, proprietary intellectual property and infrastructure advantages, are eroding faster in AI than in any previous technology cycle. 

Research is published openly, talent migrates freely between labs, and model capabilities can be replicated through a process called distillation, where smaller models learn from larger ones. The gap between a frontier model and a capable fast-follower is now measured in months, not years. When your strongest classical moats become leaky, you are forced to build defensibility somewhere else entirely.

That somewhere else is the compound of daily habit, accumulated personal context, memory, chat history, custom instructions, connected data, and workflow embedding. Nimit Chaudhry, Founder and CEO of strongmetrics, has watched this dynamic play out with particular clarity in advertising. "People are opening ChatGPT and Claude before they open Google. Every single day," he observes. "The technology was just the door. The data that walked through it every day was the actual business. AI is doing the same thing, just faster and deeper than anything before it."

The New Economics of Intelligence

The IPO filings are, at their core, a financing necessity rather than a declaration that the technology race is finished. Frontier AI training remains extraordinarily capital-intensive, and public markets offer a new runway at a moment when private valuations have stretched to near-trillion-dollar territory. OpenAI has candidly acknowledged that its listing timeline remains undecided, noting there are "things we want to do that are likely easier as a private company." The frontier capability race continues in parallel with the distribution battle, not instead of it.

What has changed is the terrain on which commercial winners will be decided. Ajay Verma, Co-founder of 0101.Today, points to a structural shift in how AI creates financial dependency that goes well beyond the subscription model. "Unlike SaaS subscriptions, AI introduces a new operating expense: token consumption," he notes. "The first AI war was about getting users. The second AI war is about monetising dependency." Token consumption, the unit by which AI companies bill for each individual prompt, code review, agent action, and customer interaction, creates a recurring cost that deepens with usage rather than remaining fixed. Recent research indicates that agentic AI workflows, where AI takes sequences of actions autonomously, can consume up to 1,000 times more tokens than traditional chat interactions. The more embedded an AI becomes in a workflow, the more tokens flow, and the more revenue accrues to the platform that owns the relationship.

Narsipur identifies the most telling signal that an AI platform has crossed from experimental tool to genuine habit. "The most strategically revealing indicator is the decoupling of usage from the leaderboard," he explains. "When a competitor ships a model that benchmarks better and users don't move, that's the moment you know habit and embeddedness, not capability, are driving behaviour." Anthropic's decision to file ahead of OpenAI is therefore not merely a capital-market manoeuvre. It is a statement about embeddedness, a bid to set the sector's pricing benchmarks and anchor public-market expectations before the larger rival can shape the narrative.

What this means for brands, investors and startups

The implications of this platform shift extend well beyond the two companies racing toward Wall Street. For brands, the AI answer layer is fast becoming the new gatekeeping surface, the equivalent of the early search engine results page. Visibility is migrating from ranking on a search results page to being the source that an AI agent cites, recommends, or acts through. 

This raises what strategists call intermediation risk, the danger that a brand's direct relationship with its customer is mediated, filtered, or even replaced by an AI layer that makes decisions on the customer's behalf. Owning first-party data and maintaining direct customer relationships becomes more commercially valuable, not less, in this environment.

For investors evaluating these IPOs, Narsipur argues that the discipline must shift accordingly. "The commoditizing frontier means benchmark leadership is a depreciating asset," he notes. "Embeddedness and the monetization gap are where durable value is won or lost." The question is not which company has the most capable model today. It is which company has so thoroughly threaded itself into daily workflows that switching would cost more than the alternative is worth.

For AI startups, the picture is equally clarifying. Competing with frontier labs on raw model quality is, by most accounts, an increasingly unwinnable fight. The defensible position lies in vertical depth, proprietary data, and genuine workflow ownership, becoming so indispensable to a specific set of tasks that the general-purpose giant cannot justify the disruption of displacing you.

Perplexity's decision to maintain its 2028 IPO timeline, regardless of how the Anthropic and OpenAI listings are received, reflects a considered bet on exactly this kind of vertical clarity. As an AI-native search product, it is building habit among users who reach for it to answer questions rather than to generate content, a distinct and growing daily behaviour with its own monetisation logic.

The industry trend that these filings collectively illuminate is one that has no clean historical precedent and yet feels oddly familiar. The most durable businesses in the digital age were not built on the best technology. They were built on the strongest daily habits. AI is arriving at this realisation with unusual speed, and the public markets, when they open their doors to these companies later this year and into 2028, will not be valuing intelligence. They will be valuing dependence.