In August 1995, Netscape Communications staged one of the most electrifying IPOs in Wall Street history. A company with almost no revenue, selling a product it gave away for free, opened at $28 per share and closed its first day at $58. The market cap hit $2.9 billion before the champagne was warm. Netscape’s Navigator browser commanded between 70 and 85 percent of the market. Its co-founder, Marc Andreessen, appeared on the cover of Time in bare feet. The company had not merely built a product, it had invented the future.

Within four years, Netscape was being absorbed by AOL for a fraction of its peak valuation. Within ten, it was shut down entirely.

The story is taught in business schools as a cautionary parable about first-mover advantage – how it can be spectacular, brief, and ultimately meaningless if the strategic foundations beneath it are hollow. Harvard Business School’s case by David Yoffie and Mary Kwak on the browser wars identified the key mechanism: Microsoft didn’t out-innovate Netscape so much as it out-distributed it, bundling Internet Explorer into Windows until Navigator became an irrelevance. Technology leadership mattered far less than control of distribution. The better product lost.

This spring, OpenAI is valued at somewhere north of $300 billion. Sam Altman is the new Andreessen – prophet, provocateur, and fundraiser of almost supernatural ability. ChatGPT has 900 million weekly active users. The consensus, once again, is that we are in the presence of an inevitable winner.

History suggests a little more scepticism is warranted.

1. The Cash Burn Is an Existential Condition

OpenAI lost approximately $5 billion in 2024 on $3.7 billion in revenue. In the first half of 2025 alone, the net loss was $13.5 billion. The company’s own internal projections, leaked to the Wall Street Journal, forecast operating losses ballooning to roughly $74 billion in 2028, a figure that would represent roughly three-quarters of that year’s projected revenue. Cumulative cash burn through 2029 is estimated at $115 billion. Profitability is not expected until 2029 or 2030 at the earliest, contingent on revenue growing tenfold from current levels.

These are not the financials of a software company. Gross margins have already collapsed from around 40 percent in 2024 to 33 percent in 2025, squeezed by inference costs – the bill incurred every time a user types a query – that quadrupled in a single year. OpenAI loses money on every paying customer. As its subscriber base grows, its losses grow with it.

The parallel with dot-com era pioneers is uncomfortable and specific. AOL, at its peak, was spending lavishly to acquire users through the now-legendary CD distribution strategy with some 300 million discs mailed to American homes in a race to achieve scale before revenue structures could sustain it. The gamble worked, briefly, and then spectacularly didn’t. The $165 billion merger with Time Warner in 2000, heralded as the media deal of the century, became one of the largest destructions of shareholder value in corporate history, a case study now used in courses from Wharton to London Business School to illustrate the dangers of valuation inflation and strategic overreach.

OpenAI’s bet is structurally similar: sacrifice profitability now to achieve the scale from which profitable dominance will eventually flow. What the dot-com era demonstrated is that this logic works only if you get there before the capital runs out – or before a better-capitalised competitor resets the rules of the game.

2. When Everyone Has a Frontier Model

AltaVista launched in 1995 and was immediately, genuinely extraordinary. It could index 20 million web pages at a time when competitors managed one or two million. Users were dazzled. For several years, it was simply the best search engine on the internet. Then Google arrived with a better algorithm, PageRank and within a few years, AltaVista’s dominance had evaporated. The search engine that had defined the category became unrecognisable, lurching from owner to owner and strategy to strategy, before being quietly shut down by Yahoo in 2013.

The strategic lesson, taught in technology management courses across business schools, is the danger of confusing being first with being permanently best. AltaVista’s technical leadership was real. It was also temporary. And once the gap closed, it had built no switching costs, no ecosystem advantages, no proprietary distribution channel that could hold users in place.

OpenAI faces the same structural problem at breathtaking speed. When DeepSeek released its R1 reasoning model in January 2025, performing at parity with OpenAI’s own models but reportedly trained at a fraction of the cost the market reaction was immediate and visceral. Nvidia lost nearly $600 billion in market value in a single day. The event raised an acute question that OpenAI’s $300 billion valuation struggles to answer: if comparable AI capability can be built for tens of millions rather than tens of billions, what exactly is the moat?

The answer, so far, is not obvious. Anthropic, Google DeepMind, Meta, Mistral, xAI, and a growing Chinese cohort are all competing at the frontier. Meta’s approach of releasing powerful models openly is directly analogous to Microsoft’s browser bundling strategy against Netscape: if you can’t win on product, you change the nature of the game. An open-source model ecosystem is to OpenAI’s closed, premium proposition what Internet Explorer bundled with Windows was to Navigator priced on a per-copy.

3. The Talent Exodus and the Mission That Got Misplaced

Institutions, as Clayton Christensen at Harvard Business School argued in his foundational work on disruptive innovation,tend to fail not because their technology becomes obsolete, but because their values do. The values that drive decisions during a company’s founding often cannot survive contact with commercial scale. OpenAI’s history over the past three years reads like a case study written to his specifications.

Ilya Sutskever, the co-founder widely regarded as OpenAI’s most important technical mind, left the company in May 2024 after a failed attempt to remove Sam Altman from the CEO role. His 52-page memo to the board, later made public through legal proceedings, described Altman as exhibiting “a consistent pattern of lying.” The chief scientist who had co-created the technology at OpenAI’s core departed to found Safe Superintelligence, a direct competitor now valued at $32 billion.

He was not alone. Jan Leike, who co-led the Superalignment team, resigned the same month with a parting statement that has become something of a eulogy for OpenAI’s founding vision: “Safety culture and processes have taken a backseat to shiny products.” Miles Brundage, who led the AGI Readiness team, followed in October 2024, writing that “neither OpenAI nor any other frontier lab is ready.” In February 2026, OpenAI disbanded its Mission Alignment team entirely,just sixteen months after creating it. The safety researchers OpenAI had used to justify its nonprofit structure and its claim to be a responsible actor in existential AI development were, one by one, gone.

What remains is a company that has updated its founding mission statement, originally premised on being “unconstrained by a need to generate financial return”, to accommodate a $300 billion for-profit restructuring, a Pentagon contract permitting use of its models “for all lawful purposes,” and an advertising tier on ChatGPT. The company that presented itself as too safety-conscious for commercial pressures now sells banner ads.

4. The Microsoft Dependency Problem Nobody Is Talking About

There is a detail at the heart of OpenAI’s business model that deserves considerably more attention than it receives. Microsoft has invested $13.75 billion in OpenAI and holds 27 percent of the restructured company. OpenAI trains and deploys its models on Azure. Its costs are partially subsidised through computing credits that Microsoft provides at discount. OpenAI’s flagship product, ChatGPT, is in direct competition with Microsoft’s own Copilot suite, which is built on OpenAI’s models.

This is not a partnership. It is a dependency relationship in which the more powerful party has every incentive to extract maximum value from OpenAI while simultaneously building the capability to replace it. Microsoft did exactly this to Netscape, using AOL distribution leverage to make Internet Explorer the default before rendering Netscape commercially irrelevant. HBS professor James Sebenius’ analysis of the browser wars for MIT Sloan Management Review described Netscape’s catastrophic miscalculation: the assumption that its technical superiority was so obvious that partners would always prefer it over the competition. “We’re so hot,” AOL’s chief negotiator recalled Netscape’s representatives saying, “we’ll license to everyone, so you better take it.” The arrogance of the dominant.

OpenAI’s relationship with Microsoft carries structural echoes. The moment Azure can offer comparable models under its own label the dependency flips. OpenAI becomes a cost centre for Microsoft rather than a strategic partner. The $13.75 billion investment looks different when you consider that it buys Microsoft access to OpenAI’s research, talent, and IP at preferential rates while binding the company to Azure infrastructure it cannot easily leave.

5. The Valuation Is a Hostage to a Future That May Not Arrive

In the final months before the dot-com bust, Cisco was valued at over $500 billion – briefly the most valuable company in the world. Its CEO, John Chambers, predicted 50 to 70 percent annual growth rates as far as the eye could see. Within two years, Cisco had lost 86 percent of its peak value. The infrastructure build-out that had justified the valuation was real; the revenue model that would monetise it was not.

OpenAI’s $300-plus billion valuation rests on internal projections showing revenue growing from $20 billion in 2025 to $200 billion by 2030 – a tenfold increase in five years, requiring ChatGPT’s paying subscriber base to expand from 50 million to 220 million, enterprise AI adoption to accelerate dramatically, and entirely new product lines such as robotics, hardware devices, e-commerce and advertising to materialise at scale simultaneously.

These are assumptions about a future that will be contested at every step by Google, Anthropic, Meta, Microsoft, DeepSeek, and others. INSEAD professor Nathan Furr, whose research on technology transitions has examined how incumbent leaders navigate disruption, distinguishes between what he calls “exploration” investments – bets on genuinely new capabilities – and “exploitation” investments that simply intensify existing approaches. OpenAI’s Project Stargate, its $600 billion compute commitment through 2030, its vast data centre builds are exploitation investments, doubling down on scale as the source of advantage at precisely the moment when DeepSeek has demonstrated that scale may not, in fact, be the decisive variable.

Nathan Furr, Professor of Strategy at INSEAD and co-author of The Innovator’s Method, has examined how established technology leaders fail to navigate disruption. Writing in the Harvard Business Review with Andrew Shipilov, Furr has argued that the central trap for technology incumbents is not that they fail to innovate, but that they mistake intensification for innovation: “organisations end up spreading their resources too thinly across too many [existing] projects,” he has observed, rather than placing genuinely exploratory bets on emerging paradigms.

Project Stargate, OpenAI’s $600 billion compute commitment through 2030 is not an exploration investment in Furr’s sense. It assumes scale remains the decisive variable in AI. At the precise moment when DeepSeek demonstrated that comparable model performance can be achieved at a fraction of the compute cost, OpenAI’s response was to commit more capital to compute. It is the AltaVista move, doubling down on the thing that made you dominant just as the basis of competition shifts beneath you.

History does not repeat itself, but it plays out in ways that business school curricula are designed to illuminate. The pattern is recognisable: a visionary pioneer captures a category, attracts stratospheric investment, builds a culture of absolute confidence in its own indispensability, and then watches as a more nimble, differently incentivised competitor changes the rules underneath it. AOL’s walled garden was dominant until Google made openness the value proposition. Netscape’s browser was peerless until Microsoft made distribution the weapon. AltaVista’s index was unmatched until PageRank made relevance, rather than scope, the measure of quality.

In each case, the pioneer didn’t lose because it stopped being good. It lost because it kept optimising for the game it had already won, while its successors reinvented the game itself.

OpenAI may yet find the path to the $200 billion revenue year it has promised its investors. The projections are not impossible, just extraordinarily dependent on a version of the future where no competitor manages to disrupt either the consumer or the enterprise AI market in the way that Google disrupted search, or the way that DeepSeek has already partially disrupted the assumption that frontier AI requires frontier capital.

The history of technology is littered with companies that were genuinely transformative, genuinely brilliant, and genuinely dominant for a period. They are remembered today, if at all, as pioneers of the wave that swallowed them. Marc Andreessen understood this better than anyone. After Netscape, he became one of the most successful venture investors in Silicon Valley, backing the disruptors rather than the disrupted.

Andreessen has invested in OpenAI.

Interested in this subject? You might also like…