Why the industry’s next winners may not be the companies building the models.
The Industry Is Asking the Wrong Question
The technology industry has found a new proxy contest: whether OpenAI or Anthropic will become the first major AI company to reach the public markets. The story has acquired the familiar rhythm of a platform race: whoever arrives first is assumed to define the category. But being first to list is not the same as being best positioned to capture the market.
An IPO is a financing event, not a verdict on who will ultimately win AI. It may crystallize a narrative, attract a new class of shareholders, and provide the first detailed look at the economics of the industry’s leading laboratories. But it will not determine where long-term value ultimately accrues.
The first phase of generative AI rewarded the companies that built the most capable models. Systems that could write, code, summarise, reason, and converse at a level once thought impossible attracted extraordinary investment almost overnight. Stanford’s 2026 AI Index reported that generative AI reached 53% population-level adoption within three years, while U.S. private AI investment climbed to $285.9 billion in 2025. That was the right way to judge the first stage of the market, but unlikely that it will be the right way to judge the next.
TCP/IP did not become economically important because companies paid a premium for the protocol. GPS became indispensable because it disappeared into logistics, aviation, and mobile phones. The technology mattered enormously, but the largest commercial opportunities emerged around the systems and services it enabled. The first AI IPO will mark the point at which investors begin asking a different question: not simply who builds the most impressive intelligence, but who makes it indispensable inside the real economy.
Public Markets Change What Success Looks Like
Venture capital is built around potential. Investors back companies long before every commercial question has been answered, betting that a technical lead, rapid growth or an exceptional founding team will eventually become a valuable business.
Public markets are less interested in potential than proof. They eventually want to know whether revenue is durable, margins are defensible, and growth can be sustained without ever-rising capital requirements.
Those questions are especially difficult in frontier AI. Epoch AI estimates that the cost of training frontier language models has increased 3.5-fold per year since 2020. At the same time, the cost of using a given level of capability continues to fall rapidly.
That combination creates an uncomfortable dynamic. Building the best models requires vast and recurring capital expenditure, even as intelligence itself becomes cheaper. A model that is innovative in January can feel ordinary by November. A valuation has to survive that compression.
Every Technology Platform Eventually Becomes Infrastructure
The first AI IPO will arrive just as the industry is beginning to compete on something other than model capability. A technology becomes most powerful when people stop noticing it. That has been true of nearly every major computing platform.
Electricity transformed industry not when dynamos were impressive, but when factories reorganised around motors. The internet became commercially decisive when it turned into a utility for commerce, media, search, payments, and enterprise coordination.
AI is moving along that curve faster than previous platforms. Stanford’s 2026 AI Index notes that capability is outpacing the benchmarks designed to measure it, with some evaluations becoming saturated within months.
This creates a strange effect: the better the technology becomes, the harder it is for most customers to distinguish one impressive demonstration from another. The spectacle of raw performance loses some of its commercial power precisely because performance becomes more widely available.
Maturity often looks like commoditisation to insiders. It is better understood as diffusion. When a technology becomes cheaper, more reliable, and easier to access, it stops being reserved for specialists and begins reshaping ordinary work.
Businesses Don’t Buy Models. They Buy Outcomes.
By the time investors are analysing quarterly earnings, enterprise customers will already be rewarding something different. Enterprise purchasing is less enchanted than public discourse. Boards do not approve budgets because a system writes elegant prose or performs well on a benchmark. Banks are already deploying AI across wealth management, client onboarding, trading, treasury, and internal operations, while keeping human oversight where regulation and client trust require it.
Hospitals and insurers care about triage accuracy, documentation burden, claims processing, patient safety, and liability. Software companies want faster delivery, but not at the expense of maintainability, security or developer trust. Different industries measure different outcomes. None buy AI simply because it performs well on a benchmark.
GitHub’s Copilot research found that developers using Copilot completed a controlled programming task 55% faster than those without it. Stack Overflow’s 2025 Developer Survey found that more developers distrust the accuracy of AI tools than trust them, with only 3.1% reporting high trust in the output.
This is the enterprise dilemma in miniature. A tool can increase speed and still require oversight. It can raise individual productivity while adding review burden elsewhere. Microsoft’s 2026 Work Trend Index found that organisational factors such as culture, manager support and talent practices accounted for more than twice the reported AI impact of individual mindset and usage.
That is why enterprise adoption rarely follows the neat logic of technical superiority. The best system on a benchmark may not be the system a bank, hospital or manufacturer can safely embed into daily operations. The buyer is not only purchasing intelligence. It is purchasing accountability.
The Scarcity Is No Longer Intelligence
The strongest misconception in today’s AI debate is that intelligence itself will remain the scarce commodity. For a period, that was true. Access to a frontier system created an advantage because very few organisations could build or use one. But scarcity in technology rarely stays fixed. It migrates.
As capable systems proliferate, scarcity shifts elsewhere. Distribution, workflow ownership, proprietary data, trust and execution become more valuable because they are far harder to replicate than intelligence itself.
A promising pilot can be built in weeks. A production deployment that survives procurement, security review, compliance testing, staff training and integration with existing systems can take far longer. The work is less cinematic than training a frontier model, but it is often where commercial advantage accumulates.
Gartner has predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. The point is not that AI will fail. It is that implementation will determine commercial success.
This is why commoditisation at the frontier should not be read as the end of opportunity. It may be the beginning of a more durable market. The invention of the database did not eliminate software companies. It enabled thousands more to exist. AI commoditisation is likely to have the same effect.
The companies that benefit most may be those with privileged positions inside work itself: systems of record, workflow platforms, trusted industry vendors, vertical specialists, cybersecurity providers, developer ecosystems and data owners. Their advantage lies less in research breakthroughs than in lower friction, stronger compliance, better data rights and a deeper understanding of customer workflows.
The First AI IPO Won’t Decide Who Wins AI
The first major AI IPO will matter because investors will, for the first time, see the economics of frontier AI in public: revenue quality, customer concentration, compute commitments, margin structure and the cost of staying near the frontier.
This is how platform shifts usually become clearer. The early period is dominated by the makers of the enabling technology. The latter period reveals the organisations that can translate the technology into institutional habit.
The IPO race between OpenAI and Anthropic is therefore important, but not final. It will not decide who wins AI any more than the first internet IPO decided e-commerce, search, social networking, cybersecurity or enterprise software. It will mark the moment when the market starts asking harder questions: not only who can produce intelligence, but who can make it dependable; not only who can demonstrate capability, but who can absorb it into the machinery of real work.
The first decade of AI has been about making machines astonishing. The next will be about making them useful enough to become ordinary. That is when the largest fortunes in technology are often made: not at the moment a platform dazzles the world, but at the quieter moment when the world begins to rely on it.
