"The Google Capital Company" — Ben Thompson

Why this is in the vault

Thompson's "cash is the ultimate commodity" thesis is the closest external articulation of RDCO's own capital-cycle investing framing, and the Alphabet $80B raise / Berkshire $10B deal is a fresh datapoint on hyperscaler capex demand and the compute-supply battle.

The core argument

Thompson reads the news (via Bloomberg) that Alphabet is raising ~$80B in equity — a $40B at-the-market program, $30B in underwritten offerings/mandatory convertible preferred, and a $10B deal with Berkshire Hathaway — to fund AI capex. He builds to it through two contrasting business archetypes:

• The "beautiful" asset-light business (Google Services/Search): free supply, customers bidding against each other to raise prices, near-zero marginal cost, ~45% margins. Wall Street rewards relative (skim) margins. Buffett admired but never bought it.
• The capital-intensive business that consumes cash but generates large absolute dollars. Thompson uses Berkshire's own arc: See's Candies (tiny capital, huge ROIC, but capped absolute profit) funded BNSF Railway (capital-hungry, lower-margin, but far larger absolute dollars). He notes BNSF's single-year net income (~$5.5B) likely exceeds Berkshire's cumulative ~$3B from See's.

The pivot: Google Cloud is becoming Google's BNSF. Cloud revenue has grown from 6% of Services (Q4 2019) to 22% (Q1 2026, $20B rev / $6.6B profit), with worse but faster-expanding margins. Services may end up the See's that funds the absolutely-larger Cloud/AI business.

On the financing question — why equity not debt, given Google has ~$126B cash vs ~$81B debt? Thompson's Occam's-razor read: Google will also issue a lot of debt soon, signaling demand for compute is being underestimated; the bearish read is Google is hedging uncertain capex ROI by sharing risk/upside. He frames the Berkshire deal as primarily a signaling mechanism ($10B is small for both): Google signals demand exceeds expectations; Berkshire (now run by Greg Abel, not Buffett) replays the See's→BNSF playbook with Berkshire-as-See's and Google-as-BNSF, deploying its ~$373B cash hoard into arguably the best available high-ROIC home — Google's TPU cost advantage means it profits most if compute commoditizes.

The closing frame ("Cash the Ultimate Commodity") generalizes: when compute supply is constrained, the deciding battle becomes who can bring the most cash to bear — and that advantage compounds (most cash → most compute capacity → sells the surplus → generates more cash). Berkshire's bet names Google as that company. He ties it back to his prior calls that compute access follows willingness-to-pay (Anthropic taking TPU supply; SpaceX supplying Anthropic).

Mapping against Ray Data Co

Strong reinforcement, with one useful tension.

• Direct hit on the "capital is the ultimate commodity" framing. Thompson's "cash the ultimate commodity / advantage compounds" closing is almost verbatim RDCO's capital-cycle priors. This is independent external corroboration that the financing layer, not just the physical (memory/power/fab) layer, is where the AI-infra cycle gets decided. Worth folding into the memory-cycle and power-cycle thesis docs as a demand-side anchor.
• Fresh hyperscaler-capex demand signal. An $80B Alphabet raise specifically to fund AI capex — a hyperscaler choosing to issue equity near all-time highs rather than rely on FCF — is a strong directional read that capex is still accelerating, not topping. That supports RDCO's Phase 2 placement (capacity-build still ramping, not yet down-cycle). The hyperscaler-capex anchor dir is currently empty of briefs; this is a candidate datapoint to log there.
• Tension worth holding. Thompson explicitly surfaces the bear case: equity issuance can mean Google is uncertain about capex ROI and wants to share downside risk. That is the same "are we late-cycle / is ROI rolling over" question the capital-cycle thesis must keep stress-testing. Don't let the bullish read crowd out the signal that a confident FCF-funder switching to equity is at minimum ambiguous. The tell to watch (per Thompson): whether substantial debt issuance follows. If it does → demand-underestimated (bullish). If it doesn't → risk-sharing (bearish).
• TPU cost-advantage point reinforces the supply-chain/NVIDIA-adjacent theses. "Compute commoditizes → the low-cost hyperscaler wins" is the same logic underneath RDCO's memory and supply-chain positioning, just applied to silicon.

Related

• [[2026-05-27-stratechery-spacex-ipo-data-centers-in-space]] — same "willingness-to-pay sources the compute" thread; this article explicitly cites the SpaceX→Anthropic supply deal.
• [[2026-05-26-stratechery-nvidia-earnings-ai-stack-new-reporting]] — adjacent hyperscaler/AI-stack capex reporting from the same author.
• [[2026-05-18-memory-cycle-v1.1]] — the capital-cycle thesis this corroborates on the demand/financing side.
• [[2026-05-18-power-cycle-v1.1]] — companion AI-infra cycle thesis; same "capacity-build still ramping" read.
• [[2026-06-01-stratechery-youtubers-box-office-youtube-bar]] — prior-day Stratechery issue (continuity of the tracked-author feed).