The question has changed

The AI trade has not disappeared; it has become more discriminating. For much of the early generative-AI cycle, investors rewarded companies for announcing models, partnerships, data-centre projects and ever larger computing budgets. That approach made some sense while the central question was whether businesses and consumers would adopt the technology at all.

By August 2026, adoption is no longer the sole issue. AI is widely deployed, spending on infrastructure is substantial and the largest technology groups are reporting tangible growth in cloud and AI-related services. The more demanding question is whether that growth can continue at a rate sufficient to justify the capital committed to servers, chips, energy and data centres.

That is the essence of a “show me” phase. It does not mean investors have concluded that AI is unprofitable or overhyped. It means narrative alone is less likely to be rewarded equally across the market. Investors are increasingly separating companies that can identify demand, monetisation and operating leverage from those whose AI strategy remains largely a promise of future relevance.

Evidence of commercial demand is accumulating

There is meaningful evidence that AI infrastructure is already producing revenue for major suppliers. Amazon reported that AWS revenue rose 37% year on year to $42.2 billion in the second quarter of 2026, while AWS operating income increased to $16.6 billion. The company also said its AI and chips businesses had each exceeded a $25 billion annual revenue run rate. These figures matter because they link AI demand to a large, established cloud business rather than to experimental products alone.

Microsoft’s latest fiscal-year results show a similar pattern. Azure and other cloud-services revenue grew 43% in the June quarter, and Microsoft said demand continued to exceed available capacity. The company reported that the additional capacity brought online during the quarter was quickly monetised. Its commercial remaining performance obligation reached $678 billion, an indicator of contracted future revenue, although investors must still account for the timing and composition of that backlog.

Alphabet’s second-quarter results offered another strong data point. Google Cloud revenue grew 82% year on year to $24.8 billion, with the company attributing the acceleration to demand for AI infrastructure, enterprise AI solutions and core cloud services. Alphabet’s overall operating margin also rose to 34%, suggesting that, at least currently, spending on AI is being absorbed by a broad and highly profitable business.

NVIDIA remains the clearest beneficiary of the infrastructure boom. In its first quarter of fiscal 2027, it reported data-centre revenue of $75.2 billion, up 92% from a year earlier, and forecast total quarterly revenue of about $91 billion for the following quarter. The results demonstrate that the supply side of the AI build-out remains exceptionally strong. They do not, on their own, settle whether all purchasers of the equipment will earn attractive long-term returns.

Capex has made cash flow the key test

The challenge for investors is that revenue growth and investment intensity are rising together. Capital expenditure is not merely an accounting line: it represents a commitment to physical assets that must be used productively over time.

Microsoft spent $41 billion on capital expenditures in the June quarter, with roughly two-thirds allocated to short-lived assets such as CPUs and GPUs. It expects calendar-year 2026 capital expenditure of about $175 billion following an accounting-related lease classification change, while maintaining that underlying investment expectations are unchanged. The company remains free-cash-flow positive, but the scale of spending raises the hurdle for future utilisation, pricing and margin performance.

Amazon’s figures make the tension still clearer. Its trailing-12-month operating cash flow increased to $161.4 billion at the end of June, but trailing-12-month free cash flow turned to an outflow of $7.6 billion. Amazon attributed the change primarily to a $66.1 billion year-on-year rise in purchases of property and equipment, largely reflecting AI investment. That does not necessarily signal financial weakness; Amazon has the operating cash generation to invest heavily. It does mean that investors must judge future returns from the investment rather than treating cloud growth as a cost-free opportunity.

Meta illustrates a different version of the same issue. It reported second-quarter capital expenditure, including principal payments on finance leases, of $31.08 billion and forecast full-year expenditure of $130 billion to $145 billion. Meanwhile, quarterly free cash flow was only $784 million. Meta’s core advertising business continues to provide the cash engine for its AI strategy, but the gap between cash generated and capital deployed makes the company particularly exposed to questions about the revenue and efficiency gains its AI systems will deliver.

What investors now want to see

The market’s standard is becoming more practical. A credible AI case increasingly requires several connected signals rather than an isolated product announcement.

First, companies need to show demand that is observable in financial reporting: cloud consumption, paid subscriptions, usage-based revenue, backlog, customer renewals or an identifiable uplift in a core business such as advertising.

Second, they need to demonstrate that increased usage can eventually support durable margins. Higher AI usage may lift revenue but can also raise inference and infrastructure costs. The strongest results will show efficiency improvements, better hardware utilisation and pricing models that align revenue with consumption.

Third, investors want evidence that capital spending responds to customer demand rather than competitive anxiety. Management teams do not need to promise immediate payback on every new data centre, but they need to explain the duration of assets, the proportion committed to customers, and the cash-flow implications if demand growth slows.

Finally, the case for AI will broaden beyond infrastructure providers. AI adoption is widespread: Stanford’s 2026 AI Index found that 88% of surveyed organisations used AI and 70% used generative AI in at least one business function. Yet broad adoption is not the same as broad profitability. The most valuable evidence for software, industrial, financial and consumer companies will be measurable productivity, higher revenue per customer, improved conversion, lower service costs or a defensible new product category.

A sorting mechanism, not an end to the boom

The “show me” phase should be understood as a maturing of the investment cycle rather than a verdict against AI. Infrastructure spending can be rational even when its returns take years to emerge, particularly for companies that already operate large cloud platforms and can spread fixed costs across many services.

But this stage is likely to create greater divergence among stocks. A company can report rapid revenue growth and still disappoint if capital expenditure, depreciation or cash consumption is running ahead of what investors expected. Conversely, evidence that new capacity is being promptly monetised, that usage is rising and that margins can be protected may justify continued investment.

Wall Street has therefore entered a more exacting phase of the AI story. The core debate is no longer whether artificial intelligence will matter. It is which businesses can turn it into recurring revenue, sustainable cash generation and returns that exceed the extraordinary cost of building the underlying infrastructure.

Sources