AI valuations are reaching dizzying heights in Silicon Valley, with OpenAI and Anthropic approaching the trillion-dollar mark, while even startups without commercial products are attracting billions in funding.
Take Safe Superintelligence, founded by former OpenAI chief scientist Ilya Sutskever. It was valued at $32 billion in 2025, despite having no publicly available product. Or Ineffable Intelligence, founded by former Google DeepMind researcher David Silver. Just months after its creation, it raised $1.1 billion in April 2026 at a valuation of $5.1 billion.
Investors are betting enormous sums on AI's future. But will these companies generate enough profits to justify such valuations?
AI may change the world. But changing the world and making money for investors are two very different things.
First, let us break down the AI business
AI is not one business. It is three.
1. The model builders
Companies such as OpenAI, Anthropic, Google DeepMind, Meta, xAI and France's Mistral AI are developing foundational models that can perform a wide range of intellectual tasks. Their ultimate ambition is artificial general intelligence (AGI)—AI capable of performing a broad range of intellectual tasks at human level or beyond.
2. The application companies
Companies such as Microsoft, Salesforce, Palantir, Harvey, Cursor and India's Sarvam AI use AI to solve specific problems, from coding and legal research to business automation. By integrating AI with proprietary data and customer workflows, they can become deeply embedded in businesses, creating switching costs and pricing power.
3. The infrastructure providers
AI requires enormous computing power, specialised chips, memory, networking, electricity and cooling. Companies such as Nvidia, AMD, Broadcom, TSMC, CoreWeave, Amazon Web Services and Microsoft Azure supply the chips, manufacturing capacity, cloud computing and data centres that power AI.
If demand for AI computing continues to grow, these companies should benefit.
All three segments are well positioned to capture substantial value. But a great business is not necessarily a great investment.
So, how big is the AI premium?
The average enterprise value-to-revenue (EV/revenue) multiple of companies listed on Nasdaq is approximately 15 times, compared with 27 times for companies in the AI industry. Put simply, this measures how much investors are willing to pay for every dollar a company earns in annual revenue, taking into account its debt and cash holdings. AI companies therefore command a premium of roughly 80%.
Consider two examples. Anthropic’s $965 billion valuation against a reported annual revenue run rate of $47 billion implies a multiple of approximately 20.5 times. OpenAI’s reported valuation of $852 billion against annual revenue of approximately $25 billion implies a multiple of 34.1 times.
These are extraordinary valuations. But markets do not value companies on current revenue alone; they anticipate future growth and profits. And the higher the expected growth, the higher the multiple investors may be willing to pay.
Is this the dot-com bubble all over again?
There are obvious similarities. The internet was transformative, and investors recognised its potential. Yet many who invested in internet companies lost huge amounts of money.
AI could follow a similar path. It may transform the world without delivering attractive returns to every investor.
So, what could go wrong? Potentially five things, namely:
1. AI models do not become as capable as expected
Today's valuations assume continued improvements in AI, potentially towards AGI, but I believe this is unlikely to be the principal risk. We have all used these models and seen how effective they can be. They still hallucinate and are far from perfect, but they are improving rapidly.
2. Businesses do not adopt AI
Privacy, security, regulation, reliability, organisational resistance and integration challenges could slow adoption. But this in my view is unlikely to be the principal risk.
The corporate sentiment is clear: companies that do not embrace AI risk being left behind. Microsoft, just this week, has offered voluntary exit packages to around 60,000 employees as it restructures its workforce and pours billions into AI. So AI adoption will vary, but the incentive is strong.
3. AI creates value, but companies cannot capture it
Competition, free open-source models and falling prices could make it difficult for AI companies to charge customers enough to earn substantial profits. After all, why pay $20 a month for an AI chatbot if a rival offers similar services for free?
But I believe this is unlikely to be the biggest risk. Consider Google Search: the more people used it, the better it became at understanding what users wanted, attracting even more users. AI could follow a similar path, as more users generate valuable feedback that helps improve models, attracting still more customers. Over time, a handful of companies could come to dominate the market.
And with AI finding uses in almost every industry, there is an enormous market to capture.
4. AI companies capture value and generate revenue, but fail to generate sufficient profits
Computing, infrastructure, research, talent and operations could absorb a substantial share of revenue. J.P. Morgan estimates that global AI capital expenditure could reach approximately $900 billion in 2026. This is a real risk, as we do not yet understand how much AI companies will need to keep spending to remain competitive. How much infrastructure is required to generate each dollar of revenue? How quickly does expensive equipment become obsolete?
We are still figuring out the economics. As more financial data becomes available, investors will be better able to estimate how much capital AI companies need and whether they can earn sufficient returns on it.
5. AI companies generate profits, but not quickly enough
Suppose models improve, businesses adopt them, companies capture the value, and profits eventually arrive. But what if this takes ten years instead of five?
For investors paying today's extraordinary valuations, this may be the most important risk.
The future belongs to those who build it and the returns belong to those who don't overpay for it.
Consider a company with revenue of $10 million, valued at $500 million—a 50× revenue multiple. Its revenue doubles to $20 million, but investors become less optimistic and value it at just 10× revenue.
Its valuation falls to $200 million—a 60% decline despite doubled revenue.
The company may have executed well, but still the investor still loses.
At high multiples, share prices depend heavily on future growth expectations. Even a modest change in those expectations can cause a sharp correction. The higher the starting valuation, the less room there is for delays or disappointment.
So, is AI overvalued?
I don't know. I suspect nobody really does and definitely not AI itself.
The technology is real. The opportunity is enormous. But the economics are still evolving, and the valuations leave little room for companies to disappointment in their growth trajectory
The two risks that matter most are profitability and timing.
Will AI companies generate sufficient profits after accounting for the enormous investments required? And will those profits arrive quickly enough to justify today's valuations?
I’d say realistically the future may be powered by AI, but for now its valuations seem powered by optimism.
