Hyperscalers and AI: what the fortnight just revealed

The window from 21 April to 5 May 2026 marked a clear shift in how markets read AI. Investors no longer reward growth promises alone. They now separate companies turning AI into visible revenue, credible backlog and capital allocation discipline from those for whom AI remains mostly a heavy cost line, even when reported results are objectively excellent.

Microsoft: operational quality, constrained capacity

Microsoft confirmed an AI annualised run rate of USD 37 billion, Azure growth of 40% and 2026 capex of around USD 190 billion, while reminding investors that demand still exceeds supply. The stock fell 2.5% over the window, a reminder that even the best franchises can underperform briefly when capex accelerates faster than the market can digest.

Alphabet: the most convincing quarter

Alphabet reported USD 109.9 billion in revenue, Google Cloud up 63% and a backlog above USD 460 billion. Management showed that external AI demand directly fuels growth and that monetisation becomes extensible, from models to hardware. The stock rose 15.3% over the window. The market now places more value on AI sold to third parties, through cloud, inference, services and enterprise tools, than on AI primarily consumed internally.

Amazon: credible monetisation, accepted price tag

The group reported USD 181.5 billion in net sales, AWS up 28% to USD 37.6 billion. In-house chips, Graviton, Trainium and Nitro, now exceed a USD 20 billion run rate and grow at triple-digit rates. Amazon announced about 2 gigawatts of Trainium capacity for OpenAI and up to 5 gigawatts for Anthropic, more than 2.1 million AI chips shipped over twelve months and more than one million NVIDIA GPUs to deploy from 2026. Bedrock spend grew 170% quarter on quarter. Trailing twelve-month free cash flow fell to USD 1.2 billion as a direct consequence of AI equipment purchases, yet the stock rose 8.9%, signalling that the monetisation path is still seen as credible.

Meta: strong growth, contested financial reading

Taken in isolation, the numbers are excellent, USD 56.3 billion in revenue, up 33%, and USD 26.8 billion in net income. Yet the market immediately focused on 2026 capex, raised from USD 115-135 billion to USD 125-145 billion on higher component prices. Meta does not yet offer equivalent visibility on direct AI sales to enterprises. The link between capex dollars and incremental revenue dollars remains more indirect. The stock lost 8.7% over the window after dropping more than 6% after-hours on the release.

Three distinctions now visible

First, between AI sold to external customers, cloud, APIs, agents, chips, and AI mainly used to improve an existing business. Second, between capex matched by demand visible in backlog or multi-year commitments, and more speculative capex. Third, between balance sheets able to absorb a prolonged investment cycle and models whose monetisation still depends on a future narrative. This grid explains almost the entire equity hierarchy observed over the fortnight.

For a family office: look for robustness, not the best model

The point is not to pick the best AI model, but to identify where monetisation becomes robust enough to justify the cost of capital. Alphabet and Amazon clearly improved their credibility profile over the period. Microsoft remains a very high-quality file but is sensitive to capex timing. Meta is still investable, provided one accepts higher tolerance for execution risk and a smaller position size.