Alphabet Beat Everything and Fell 5%: The AI Spending Bill Comes Due

Alphabet delivered a quarter almost nobody could fault. Revenue of $119.80 billion against the $116.93 billion the market wanted. Google Cloud up 82% year on year, which is an acceleration on the 63% posted in the first quarter and comfortably ahead of Azure and AWS. Adjusted earnings per share of $2.85 against $2.89 expected, which is a rounding error in a quarter with numbers this size.

The stock fell 5% after hours.

It fell because Anat Ashkenazi, Alphabet’s finance chief, told the call that 2026 capital expenditure would now land between $195 billion and $205 billion. Analysts polled by Visible Alpha had modelled roughly $188 billion. FactSet consensus was $187.1 billion. In April, the guidance had been $180 billion to $190 billion, itself an increase on the $175 billion to $185 billion given before that. Alphabet has now raised its capex guidance three times in roughly six months, and the top of the range has moved by $30 billion.

This is the trade of 2026 in a single evening. Beat on everything that measures the business today, get punished for what you intend to spend tomorrow. The market has stopped rewarding the AI narrative on faith and started asking for the receipts.

The arithmetic that worries people

Take the numbers apart and the concern becomes clear. Alphabet spent $35.7 billion on capex in the first quarter, already up 107% year on year. To reach even the bottom of the newly raised range, quarterly spending has to climb substantially from there for the rest of the year. Free cash flow fell roughly 47% year on year in the first quarter, to $10.1 billion. Trailing twelve-month capex now represents about 63% of cash flow from operations. That is not a company funding growth from surplus. That is a company approaching the point where the surplus disappears.

Aggregate across the four largest hyperscalers and the figure lands somewhere around $700 billion to $725 billion for 2026, roughly double 2025, which was itself roughly double 2024. Bank of America projects the combined figure passes $1 trillion in 2027. Morgan Stanley has produced a similar estimate. Sequoia’s David Cahn estimated a $600 billion annual gap between AI infrastructure spending and AI ecosystem revenue in 2025, and nothing published since suggests it has narrowed.

Allianz Research puts the divergence between capex growth and revenue growth at roughly 46%. The comparison they draw is with the 2001 telecom overbuild, which peaked at a 32% divergence and remains the reference case for capital misallocation in the technology sector. If the 46% figure is accurate, the current cycle has already exceeded it.

Then there is the part that does not appear in the capex line at all. Moody’s reported in early 2026 that hyperscalers hold approximately $662 billion in signed data centre lease commitments that have not yet commenced. Under GAAP’s lease commencement standard those obligations sit off balance sheet, meaning they are absent from the figures analysts scrutinise. That off-balance-sheet total is larger than the combined on-balance-sheet debt of the same companies.

The winners and the losers are not who you would guess

Here is where it gets interesting for anyone trading the sector rather than holding an index. The money being spent has to go somewhere, and where it goes is not where it is being raised.

AMD has spent the past twelve months turning its data centre business into the story. Full-year 2025 data centre revenue reached $16.6 billion, up 32% on 2024, and the pace has since accelerated: first-quarter 2026 data centre revenue hit a record $5.8 billion, up 57% year on year from $3.7 billion, on demand for its fifth-generation EPYC processors and Instinct MI350 GPUs. Group revenue of $10.3 billion was up 38%, and the segment now accounts for well over half the company. Management has guided server CPU revenue to grow more than 70% year on year in the second quarter and signed a deal with Meta to deploy up to 6 gigawatts of Instinct GPUs across future generations. The point for traders is not that AMD is beating NVIDIA, it is not, but that the pool of companies monetising hyperscaler capex is widening, which is exactly what you would expect at this stage of a build cycle.

NVIDIA reported first-quarter FY2027 revenue of $81.6 billion on 20 May, up 85% year on year, with data centre revenue of $75.2 billion up 92%. That quarter marked the first with a near-equal split between hyperscale customers at around $38 billion and everyone else at around $37 billion, which suggests AI demand is broadening beyond the big four. Analysis of the compute-to-revenue bridge indicates NVIDIA now captures around 57 cents of every dollar of hyperscaler capital spending, up from 39 cents four quarters earlier. Guidance for the following quarter was $91.0 billion.

So the arms dealer is doing rather well out of the arms race. The armies, less obviously so. The hardware enablers, the power infrastructure providers, the cooling specialists and the memory manufacturers are booking revenue from spending that shows up as a cash outflow on somebody else’s statement. That is the divergence within the sector that matters, and it is the reason a broad technology position and a targeted one have produced quite different outcomes this year.

The supply constraint nobody solved

One detail from Alphabet’s call deserves more attention than it will get. Ashkenazi said the company plans to expand its use of third-party cloud capacity in the third quarter as a bridging strategy while it builds internal capacity, describing the environment as supply constrained and noting the company has said as much for multiple quarters running.

Read that carefully. Alphabet, which is about to spend up to $205 billion on infrastructure, is renting capacity from competitors because it cannot build fast enough. Microsoft has disclosed an $80 billion backlog of Azure orders it cannot fulfil due to power constraints. This is the counter-argument to the overbuild thesis, and it is a serious one: you do not have an overcapacity problem while you are turning away paying customers.

It is also, awkwardly, an argument for spending even more. Which is precisely what the market punished.

What actually happens next

Alphabet was first of the big four to report. Microsoft, Amazon and Meta follow, and each will face the same question in the same form: what are you spending, and what is it earning. Ashkenazi’s answer was that Google will keep investing as long as the return looks attractive, which is either reassuring or circular depending on your priors.

The genuine uncertainty is timing rather than direction. Almost nobody disputes that AI infrastructure will eventually generate returns commensurate with the investment. The question is whether the conversion happens fast enough to service a capital base heading toward a trillion dollars a year, funded increasingly through debt markets rather than operating cash flow. Consensus expects free cash flow growth to resume across the hyperscalers by 2027 or 2028. That is a long time to be asked to hold your nerve.

For traders, the practical read is that this sector has stopped moving as a bloc. Hardware and infrastructure names are capturing the spending. The companies doing the spending are being marked down for it, regardless of how good the underlying quarter looks. A 5% drop on a comprehensive revenue beat is not a market that has lost interest in AI. It is a market that has started asking a different question, and the answer will not arrive this quarter.

Common questions