As AI companies race to build computing infrastructure, investors are looking beyond revenue growth to a harder question: can future earnings justify the capital being committed today?
The artificial intelligence boom is increasingly becoming a test of corporate finance.
For several years, investors have rewarded technology companies for expanding their AI capabilities, gaining customers and building the infrastructure needed to train and run increasingly powerful models. Data centres, specialised chips and cloud computing capacity have become central to the competition.
But building this infrastructure requires enormous upfront investment. The costs arrive before all the potential revenue does, and the economic life of the assets is uncertain. Companies must commit capital today while trying to predict how quickly demand will grow, how rapidly technology will change and how much customers will ultimately be willing to pay.
That gap between spending and returns is becoming one of the most important questions for financial markets.
The issue came into sharper focus on October 26, when Anthropic disclosed financial information in its US IPO filing. Reuters reported rapid revenue growth alongside steep losses and rising costs, making the company a significant test of investors’ willingness to finance the next stage of the AI race.
The question is no longer simply whether AI has commercial potential. It is whether the businesses building it can generate returns that justify the cost of doing so.
Growth is impressive. But what does it cost?
Anthropic’s disclosed figures illustrate the scale of the opportunity and the difficulty of evaluating it. According to Reuters, the company’s revenue reached $4.59 billion in 2025, up from $386 million in 2024. Revenue for the first six months of 2026 was reported at $16.29 billion, compared with $1.16 billion in the same period a year earlier.
Those figures indicate exceptionally rapid expansion. They do not, by themselves, establish whether the business is economically sustainable.
AI companies face costs associated with computing capacity, model development, employees, energy and distribution. Some costs rise as products become more widely used. Others require long-term commitments before demand is fully known.
Anthropic’s reported losses also require careful interpretation: Reuters noted that a significant non-operating accounting item related to convertible bonds affected the reported result. That distinction matters because an accounting loss, cash operating costs and cash flow are different measures. Investors need to understand each one rather than treating a single headline figure as a complete picture of the business.
Revenue growth is evidence of demand. It is not proof that the company can convert that demand into lasting profits.
The infrastructure race changes the risk
The AI investment cycle extends far beyond the companies developing models.
Cloud providers, data-centre operators, chipmakers, energy suppliers and financial institutions are all exposed to different parts of the expansion. Some sell equipment or services immediately. Others commit capital to facilities that may take years to generate sufficient cash.
This creates a chain of financial dependencies.
A cloud provider may build capacity in anticipation of future demand. A specialised infrastructure company may borrow to acquire expensive chips. A lender may finance a project based on assumptions about utilisation, customer contracts and the resale value of equipment. An investor may then value the company based on projected revenue several years into the future.
If demand develops as expected, the chain can support strong growth. If customers delay spending or computing prices fall faster than costs, the same commitments can become a burden.
The risk is not necessarily that AI demand disappears. It is that the infrastructure is built too quickly, at too high a cost or in the wrong places.
Reuters reported on October 19 that S&P Global had warned that major technology companies could have more AI-related liabilities than are readily visible, including obligations arising from complex financing structures. That makes it increasingly important to examine not only reported debt but also leases, guarantees, joint ventures and other contractual commitments.
A company’s balance sheet may tell only part of the story.
Why financing matters as much as profitability
When a company funds investment from its own operating cash flow, the relationship between spending and its existing business is relatively direct. When it relies more heavily on debt, outside investors or complex contractual arrangements, more parties become exposed to its future performance.
Borrowing is not inherently a problem. Debt can be an efficient way to finance productive assets when future cash flows are sufficiently predictable. The difficulty arises when debt obligations are fixed but the revenue expected to support them remains uncertain.
AI infrastructure is particularly sensitive to this mismatch. Facilities and equipment require substantial investment, while demand, pricing and technology can change quickly. A new generation of chips may improve performance but also affect the value of existing hardware. A customer may commit to capacity but later change its requirements. Competition may reduce the prices providers can charge.
The result can be pressure on margins and cash flow even if overall demand for AI continues to increase.
For investors, free cash flow is therefore an important measure to watch alongside revenue and operating profit. It helps show how much cash a business generates after capital expenditure, although it should be interpreted in the context of the company’s growth stage and investment cycle.
A business that spends heavily today may still create substantial value. But the case depends on future cash generation, not spending alone.
The danger of confusing an industry boom with a good investment
One of the most difficult aspects of the AI cycle is that the technology can be transformative while some investments still disappoint.
Railways, telecommunications and the internet all changed economies, but not every company that financed or built the infrastructure earned attractive returns. Competition, excess capacity, financing costs and valuations determined how much value ultimately accrued to shareholders and creditors.
AI may follow a similar pattern without repeating any historical episode exactly.
Some companies could establish durable businesses because they own scarce infrastructure, have strong customer relationships or can deliver useful products at competitive prices. Others may struggle to recover their capital expenditure if the market becomes crowded or customers gain more bargaining power.
Investors must also distinguish between different layers of the AI economy. A chip supplier, a cloud infrastructure company and a model developer have different cost structures, revenue models and capital requirements. Strong results in one part of the market do not guarantee similar economics elsewhere.
Valuation adds another layer of risk. Even a successful company can disappoint shareholders if its growth falls short of what its share price already assumes.
What would make the investment sustainable?
There is no single metric that can settle the question, but several indicators can help distinguish durable growth from an expensive race for capacity.
First, revenue quality. Investors need to know whether demand comes from recurring customer usage, long-term contracts or temporary experimentation. Growth supported by customers who continue to pay for measurable benefits is more convincing than growth driven primarily by initial trials.
Second, capital efficiency. Revenue and gross profit should be considered alongside the investment required to generate them. If each additional unit of revenue requires disproportionate spending on computing capacity, the economics may prove more difficult than headline growth suggests.
Third, financing resilience. Debt maturities, lease commitments, interest costs and access to capital all matter. A company that can survive a period of slower growth has more flexibility than one that must continually raise new funds to meet existing obligations.
Finally, evidence of productivity gains. Businesses adopting AI must see enough value—through higher output, lower costs or better products—to justify ongoing spending. If customers cannot demonstrate that value, enthusiasm may eventually give way to tighter budgets.
These indicators are relevant to investors, but they also matter to the wider economy. If infrastructure spending supports productivity growth, the benefits may extend well beyond the technology sector. If capital is allocated inefficiently, the consequences can include weaker returns, write-downs and pressure on lenders and investors.
The next phase will reward financial discipline
The AI investment boom does not have to end in a collapse to produce winners and losers. A more selective market could continue funding the strongest businesses while demanding better evidence from companies whose plans depend on uncertain future demand.
That would mark a shift from valuing ambition alone to examining the economics behind it.
Anthropic’s IPO filing offers investors another opportunity to scrutinise the relationship between rapid growth, costs and financing. It is not a verdict on the entire AI industry. It is a reminder that the ability to attract capital and the ability to earn an attractive return on that capital are two different achievements.
The defining question for the next phase of the AI race may therefore be less about which company builds the most capacity or develops the most advanced model.
It may be about which companies can turn that investment into sustainable cash flow before the cost of staying in the race becomes too high.
