The Google Capital Company | Stratechery by Ben Thompson

ELI5 / TLDR

Google’s search-ads business is one of the best money machines ever built: the supply (web content) is free, advertisers bid each other up, and users decide who gets to pay. But that beautiful business can only get so big, because advertising is only a slice of the economy. So Google is doing what Buffett did with Berkshire — taking the cash from its perfect, capital-light business and pouring it into a capital-heavy one (cloud and AI) that earns lower margins but could be enormously bigger. The twist: Berkshire just bought $10 billion of Google stock, and Ben Thompson reads the whole deal as a bet that whoever has the most cash will end up with the most compute, and that advantage compounds.

The Full Story

The most beautiful business model ever

Thompson opens by describing a business that sounds too good to exist. Your supply is free. Your customers compete against each other to raise your prices. And your users get to decide which of those customers earns the right to pay you. You build a little infrastructure, pay a bit of depreciation on it, and collect some of the fattest margins in commercial history.

That business is Google search advertising. It was so good that even Warren Buffett, who passed on it, marveled at it. As a GEICO advertiser, he watched the meter run:

“Anytime you’re paying somebody 10 or 11 bucks every time somebody just punches a little thing where you got no cost at all, that’s a good business unless somebody’s going to take it away from you… you’ve almost never seen a business like it.”

The engine here is what Thompson calls aggregation. An aggregator wins by trading relative quality for absolute quantity. A visitor Google sends to a website is worth less than someone who shows up directly — but Google sends so many that the total is larger. One paid click outweighs thousands of useless ad impressions. The aggregator floods the system with volume and trusts that the absolute total beats the higher-quality trickle.

Here’s the irony Thompson points out: Wall Street loves these companies for the opposite reason. Investors prize asset-light businesses that take a thin skim off a market they don’t actually participate in, then keep almost all of it. The market has historically cared about relative money — fat margins — more than absolute dollars.

Buffett’s two lessons: See’s Candies and BNSF

To set up where Google is heading, Thompson reruns Buffett’s own evolution as an investor.

Berkshire Hathaway itself was a mistake — a dying textile mill Buffett bought because it was cheap, then got stuck with. The lesson he drew, in his own words:

“It’s far better to buy a wonderful company at a fair price than a fair company at a wonderful price. Charlie understood this early. I was a slow learner.”

The wonderful company was See’s Candies, bought in 1972 for $25 million. It threw off enormous profits on almost no reinvested capital — earning 60% pre-tax on invested capital, and over the decades returning $1.35 billion in earnings while only needing $32 million plowed back in. A perfect, capital-light cash machine.

But a See’s has a problem: there’s nothing to do with all that cash. You can only sell so much candy. So Berkshire used the See’s cash to buy other businesses — including one at the far opposite end of the spectrum, BNSF Railway. Railways are capital monsters; BNSF ate $3.8 billion in capital last year. But it also made $5.5 billion in net income on $23.4 billion of revenue. Berkshire has earned more from BNSF in a single year than it has from See’s in its entire history.

So, which is the better business?

Thompson lets that question hang. The point: the capital-light jewel funded the capital-heavy giant, and the giant turned out to mint more total dollars.

Google Services is the candy; Google Cloud is the railway

Now the parallel snaps into focus. Thompson lines up Alphabet’s numbers across seven years:

Q4 2019: Google Services made $43.2B revenue / $13.5B profit. Google Cloud made $2.6B and lost $1.2B — 6% the size of Services.
Q1 2023: Cloud turns its first profit. Services $62B rev / $21.7B profit. Cloud $7.5B rev / $0.2B profit — 12% of Services’ revenue, 1% of its profit.
Q1 2026: Services $89.6B rev / $40.6B profit. Cloud $20B rev / $6.6B profit — now 22% of Services’ revenue and 16% of its profit.

Two things jump out. Services has more than doubled revenue and tripled profit in seven years — and it’s vastly more scalable than any candy company. But Cloud is growing faster, and its margins, while thinner (33% vs. 45%), are expanding more quickly.

The ceiling question is everything. Advertising, by definition, is only a fraction of the whole economy. Cloud’s growth, by contrast, is AI — which people variously hope, fear, or expect might swallow the entire economy. So Thompson poses his central analogy as a question:

Might we one day look back and realize that Google Services provide the cash flow to build a business with relatively worse margins, but absolutely higher dollars, much like See’s helped fund BNSF?

The $80 billion raise and the Berkshire twist

The trigger for the whole essay is a Bloomberg report: Alphabet is raising $80 billion to fund its AI spending — a $40B at-the-market share program, $30B in underwritten shares and convertible preferreds, and a $10 billion deal with Berkshire Hathaway.

Thompson asks two questions.

Why issue equity at all? Normally debt is the smarter tool — you borrow, invest, pay off the loan, and existing shareholders keep all the upside (plus the interest is tax-deductible). Google has ~$81B in debt but $126B in cash, so it has plenty of room to borrow more. Thompson’s Occam’s-razor read: Google is also about to issue a lot of debt, because demand for compute is even bigger than people think and they’ll use every instrument available. The bearish read: Google isn’t sure the AI capex will pay off and wants to share the risk along with the upside. If no big debt issuance follows, that’s the tell.

Why is Berkshire suddenly interested in Google, after all these years, at barely a discount to its all-time high? Partly because Buffett no longer calls the shots — his successor Greg Abel does. But Thompson argues Abel is just running Buffett’s playbook again:

Only this time Berkshire Hathaway is See’s Candies, and Google is BNSF.

Berkshire is sitting on $373 billion in cash and generated $25 billion in free cash flow in 2025. There are vanishingly few places to deploy that kind of money at a high return — and Google may be the best one going. It has optionality: its ads business benefits from AI, it’s a contender at the model layer with Gemini, and it can sell compute to rival frontier labs. Crucially, thanks to its in-house TPUs, Google has a durable cost advantage — so in a future where compute becomes a commodity, Google is the hyperscaler best positioned to profit.

The signal, and the cash-compounding endgame

$10 billion is small change to both companies, so Thompson reads it mostly as a signal. From Google: demand for compute is far bigger than the market believes, and we’ll fund supply by any means, equity included. From Berkshire: this is an endorsement and a validation — and if the signal is right, Berkshire is getting a deal while putting its cash machines to work building the future.

Then Thompson zooms out to his real thesis: cash is the ultimate commodity. He revisits an earlier debate — OpenAI backers argued OpenAI was ahead because it had locked up the most compute. Thompson disagreed: in AI, distribution and transaction costs are still free (the two preconditions for aggregation), so the winners are whoever has the most compelling products. Better products win more users, which generate the cash to buy the compute to serve them. He cites Anthropic taking TPU supply from Google, then SpaceX supplying Anthropic the compute it needed — expensive, but the demand justified the price.

His closing escalation is the punchline. All of that assumed enough compute exists in the world to be bought. But what if one day it doesn’t?

What if the ultimate battle, the one that determines who gets compute, becomes a matter of who can bring the most cash to bear? And what if that advantage compounds, such company with the most cash capacity ends up with the most compute capacity, which you already know they will sell, in addition to using themselves, driving the ability to generate more cash?

In that world, the company with the most cash wins the most compute, sells the surplus, and makes even more cash — a flywheel. Which company is the best bet? Thompson’s last line: we now know which one Berkshire is betting on.

Key Takeaways

Aggregator logic: Google maximizes absolute value (total clicks, total visitors) at the expense of relative value (the worth of any single one). Wall Street historically valued tech for the opposite — high margins on an asset-light skim.
See’s vs. BNSF: Berkshire’s capital-light candy business (60% pre-tax on capital, almost nothing reinvested) funded its capital-heavy railway (eats $3.8B/yr in capital), and the railway now out-earns the candy by a wide margin in a single year.
Google’s mirror: Google Services (ads) is the high-margin, capital-light jewel; Google Cloud (AI) is the lower-margin, capital-heavy giant being funded by it.
Seven-year trajectory: Cloud went from losing $1.2B (Q4 2019) to $6.6B profit (Q1 2026), rising from 6% to 22% of Services’ revenue. Cloud margins (33%) trail Services (45%) but are expanding faster.
The ceiling argument: Advertising is capped at a fraction of the economy; AI/cloud potentially is the whole economy — so absolute dollars may eventually favor the lower-margin business.
The $80B raise: $40B at-the-market shares, $30B underwritten shares + convertible preferreds, $10B to Berkshire. Some ATM proceeds cover tax on employee equity awards.
Equity over debt: Unusual, since Google has $126B cash against $81B debt. Bullish read: more debt is coming too. Bearish read: Google wants to share AI-capex risk.
Berkshire’s deployment problem: $373B in cash, $25B FCF in 2025 — few targets can absorb that at a high return. Greg Abel (post-Buffett) is replaying the See’s-to-BNSF move with Berkshire as See’s and Google as BNSF.
TPU edge: Google’s in-house chips give it a durable cost advantage, positioning it best if compute becomes a commodity.
Thompson’s thesis: Cash is the ultimate commodity. If compute becomes truly scarce, whoever can bring the most cash wins the most compute, sells the surplus, and compounds the advantage.

Claude’s Take

This is Thompson at his most characteristic: take a financial news item (the Berkshire stake), find a clean historical analogy (See’s funding BNSF), and use it to reframe how you should think about a whole industry. The See’s/BNSF mapping is genuinely illuminating — it reframes Google’s AI spend not as a margin-destroying mistake but as the classic Buffett move of routing capital-light profits into a capital-heavy compounder. That’s the kind of frame that sticks.

Where to keep a skeptical eye: the essay’s bullish lean does most of the work, and Thompson is honest enough to flag the bearish read (equity issuance as risk-sharing because Google isn’t sure the capex pays off) before largely setting it aside. The “cash is the ultimate commodity” finale is elegant but assumes a future of genuine compute scarcity that may or may not arrive — and if compute doesn’t become scarce, much of the flywheel argument loosens. The closing rhetorical flourish (“we now know which one Berkshire is betting on”) is persuasion, not proof; a $10B stake is, as Thompson himself notes, rounding error for both firms, so reading it as a grand civilizational bet is a stretch he’s aware he’s making.

Still, the core mechanics are solid, the numbers are real and well-chosen, and the aggregation-theory framing (absolute vs. relative value) is a genuinely useful lens that most coverage of Big Tech misses entirely. Score: 8. Docked from higher because the conclusion leans harder on narrative momentum than the evidence strictly supports, and the whole piece rests on one analogy doing a lot of load-bearing work.