The Google Capital Company
Stratechery by Ben Thompson · 2026-06-02
Berkshire's $10B stake in Google is a bet that Google is replaying the See's-funding-BNSF playbook at planetary scale — using Search's high-margin cash to buy the compute capacity that will define who wins AI.
The logic turns on a compounding loop: cash buys compute, compute generates more cash, and once global compute supply runs short, cash capacity becomes the binding constraint on who can keep playing. Google Services throws off the margins; Google Cloud is the capital sink large enough to absorb them productively — the same shape as See's funding BNSF, where a small high-margin business fed a giant low-margin one that ultimately produced far more absolute profit. Berkshire, sitting on $373B in cash with nowhere obvious to put it, is effectively naming Google as the company best positioned to convert dollars into compute into more dollars.
Search advertising's high-margin cash flow is increasingly being deployed into a lower-margin but vastly larger AI-and-cloud opportunity — the same pattern Berkshire ran with See's funding BNSF.
The new question is what happens when there isn't enough compute in the world to buy — at that point, the company with the most cash-generating capacity wins the compute, which lets it generate more cash in a compounding loop.
Berkshire is sitting on $373B in cash and $25B in annual free cash flow with few places to deploy it; investing in Google flips the analogy so that Berkshire itself becomes the cash-gushing See's and Google the capital sink that can absorb it productively.
By taking the $10B stake, Berkshire is effectively answering which company is best positioned in a world where cash capacity translates into compute capacity translates into more cash — Google.
Berkshire redeployed See's cash into businesses like BNSF Railway, which consumes billions in capex but generates more absolute profit in a single year than See's has produced in its entire history under Berkshire.
Open
- · What happens to competitors without comparable cash-generating capacity once compute supply tightens?
- · Can Google Cloud actually absorb Search's cash flow at returns that justify the redeployment?
Pipeline
- source kind
- url
- generated by
- anthropic+voyage
- candidates
- 21 (selected 5)
- embeddings
- voyage-3.5
Coverage
100% covered
Each block is one paragraph of the source. Darker means the decomposition captures it well; lighter means it was left out — the part of the document the summary doesn’t cover.
Considered candidates (16)
Below top-k · 13
- evidenceGoogle Cloud's trajectory from money-loser to profit enginec 0.70
Cloud went from $2.6B revenue and a $1.2B loss in Q4 2019, to first profitability in Q1 2023, to $20B revenue and $6.6B profit in Q1 2026 — now 22% the size of Services by revenue and 16% by profit.
- contextGoogle just sold $10B in equity to Berkshire Hathawayc 0.70
Bloomberg reported a new ATM equity program, including a $10B issuance to Berkshire Hathaway — Google's first big capital raise of this kind, near all-time-high prices.
- mechanismGoogle's optionality across the AI stackc 0.70
Google benefits from AI through Services, competes at the model layer with Gemini, and sells compute capacity to frontier labs — multiple ways to win regardless of how the AI race resolves.
- mechanismTPUs give Google a structural cost advantage in commoditized computec 0.70
If compute becomes a commodity, Google's custom TPU stack means it can produce that compute at lower cost than rivals — positioning it as the hyperscaler with the most durable margin.
- claimGoogle Search is arguably the most beautiful business model ever builtc 0.60
Google's search advertising business has free supply, customers who bid each other up on price, and users who decide which customers get to pay — yielding extraordinary margins on minimal infrastructure spend.
- caveatEquity is a strange choice when Google has so much debt capacityc 0.60
With $126B cash against $81B debt and capex funded from free cash flow, Google could easily issue more tax-advantaged debt — making the choice to issue equity notable and in need of explanation.
- mechanismTwo readings of the equity raise: bullish demand or hedged riskc 0.60
Either Google plans massive additional debt issuance and is signaling that compute demand is wildly underestimated, or it is uncertain about capex ROI and wants to share risk and upside with new equity holders.
- contextWall Street has rewarded asset-light skimming, not absolute dollarsc 0.55
The market has historically prized tech companies precisely because they avoid participating in the markets they enable, taking a high-margin skim — valuing relative profitability over absolute profit dollars.
- evidenceBuffett watched GEICO pay Google $10+ per click and called it unmatchedc 0.50
At the 2017 Berkshire meeting, Buffett recalled GEICO paying roughly $10–$11 per click to Google and described it as a business almost without peer — yet he never invested.
- mechanismAggregators trade relative quality for absolute quantityc 0.50
Google maximizes absolute value at the expense of per-unit value: each visitor, click, or result is worth less individually, but the total volume more than compensates.
- contextBuffett's hard-won lesson: wonderful businesses beat cheap onesc 0.50
After being stuck owning the failing Berkshire Hathaway textile business, Buffett concluded it is far better to buy a wonderful company at a fair price than a fair company at a wonderful price.
- contextEarlier thesis: products win compute, not the reversec 0.50
Stratechery had argued in Mythos, Muse, and the Opportunity Cost of Compute that distribution and transaction costs being free means the labs with the best products attract the cash that buys the compute — not the ones with pre-locked supply.
- evidenceSpaceX-Anthropic deal confirmed compute follows willingness to payc 0.50
When SpaceX sold expensive supply to Anthropic shortly after the original argument, it validated the view that high-demand labs can always source compute by paying up — at least so long as supply exists somewhere.
Redundant with selected · 3
- contextAdvertising has a ceiling; AI plausibly does notc 0.60 · sim 0.82
Google Services' margins are inherently high but advertising is only a fraction of GDP, while Cloud's AI-driven growth points at a market that could absorb a much larger share of the economy.
overlapped with: Google Services is becoming Google's See's, funding Google Cloud's BNSF
- exampleSee's Candies as the prototype cash-gushing wonderful businessc 0.55 · sim 0.83
Berkshire bought See's in 1972 for $25M; it threw off enormous pre-tax returns on only $8M of invested capital, generating cash with nowhere productive inside the business to put it.
overlapped with: See's profits funded BNSF — a capital-heavy, lower-margin, bigger-dollar business
- implicationThe deal's real value may be as a mutual signalc 0.55 · sim 0.85
$10B is small for both companies, so the primary function is signaling: Google validates extreme compute demand by raising equity, and Berkshire's participation validates Google's bet.
overlapped with: Berkshire's bet names Google as the winner of the cash-for-compute race
Janitor
Non-content spans (acknowledgements, references, footnotes, headers, boilerplate) are dropped before the decomposition runs.
- total spans
- 37
- kept
- 37
- dropped
- 0
- outliers
- 2
- content · 37