Alphabet spends $44.9B in Q2 — and the money went into electrons

Last week, the w30 story made a single argument: hyperscalers are no longer priced on the volume of their capital spending, but on its composition — what the money is actually buying. Meta raised guidance by $10B and its stock fell 8.6% the same day, because the market had already shifted to asking "electrons or chips?" This week, Alphabet's Q2 results provided the sharpest answer to that question yet.

Every company building datacenters at gigawatt scale faces the same physical ceiling: grid interconnection queues. At $9–15M per megawatt of capacity, you can have the balance sheet, the chip orders, and the engineering talent, and still find yourself waiting seven to twelve years for your site to receive power from the public grid. The bottleneck was never the GPU. It was always the socket.

Alphabet's Q2 2026 results landed at $44.9B in capital expenditure, up 100% year-over-year, with full-year 2026 guidance reaffirmed at $180–190B. Analysts going into the call had been focused on TPU rollout pace — TPU being Google's proprietary AI chip, its in-house equivalent of an Nvidia GPU — and Google Cloud margin signals. What the quarter confirmed instead was an acquisition: $4.75B for Intersect Power, a clean energy developer whose deal closed on March 13, 2026. Not a PPA — a Power Purchase Agreement is a long-term electricity contract signed with an existing utility, the instrument hyperscalers have used for years to secure electrons without owning the underlying asset. An acquisition. Alphabet bought the energy company outright.

At $9–15M per megawatt, the $44.9B CapEx envelope for a single quarter represents the physical equivalent of roughly 3 to 5 gigawatts of new capacity. The Intersect deal alone — $4.75B — secures several gigawatts of future energy and datacenter projects, co-located on sites Alphabet now owns. The model is what the company calls "Energy Parks": solar generation and servers built side by side, electricity flowing from panel to processor without ever touching the public grid. No interconnection queue. The seven-to-twelve-year problem disappears when you build your own power system.

Alphabet is the first hyperscaler to acquire a major clean energy developer outright rather than sign contracts with one. The market's assessment of that move remains open as of the July 22 after-hours session. What is already readable is structural: the marginal dollar of hyperscaler CapEx has moved past the chip layer. It now goes into the asset that makes chips operable at scale — the electron, the storage, the land, the grid bypass.

This is not a confirmation of the w30 thesis. It is an amplification. The constraint is so acute that the marginal dollar is no longer moving toward PPAs with energy companies — it is moving toward acquiring the energy companies themselves. Meta, Alphabet, Microsoft, Amazon remain the upstream engine, not the investable consequence. That consequence sits one layer down, in the companies that own what the hyperscalers are now buying outright.