Google

Alphabet did approximately $350B of revenue in 2024 and about 76% of it was advertising. The mechanism underneath is narrower than that number implies: a user types a query with commercial intent, Google runs a real-time auction among advertisers for placement against it, and the winner pays only if the user clicks. Google's cost of serving that query is a fraction of a cent. The advertiser's willingness to pay is set by how close the query sits to a transaction — "personal injury lawyer" clears above $100 a click, "weather" clears at nothing.

Divide the arithmetic out. Search and other advertising was roughly $198B in 2024 against something on the order of five trillion queries. That is approximately four cents of revenue per search on average, and it is one of the most profitable four cents in commerce because the marginal cost is nearly zero and the inventory refreshes every time a person is curious.

The segments beyond search matter differently. YouTube advertising was about $36B — a second ad business with worse targeting but far better duration. Subscriptions, platforms and devices reached roughly $40B and is the fastest-growing consumer line, mostly YouTube Premium, YouTube TV and Google One. Network — ads Google sells on other people's websites — was about $30B and is shrinking, because the open web that business monetises is itself shrinking. Google Cloud did roughly $43B with about $6.1B of operating income, its first genuinely profitable year. Other Bets did $1.6B of revenue against a $4.4B operating loss, and is best read as a portfolio of options in which Waymo is the only one currently worth pricing.

Google's moat is usually described as the search index or the algorithm. Both are real and neither is the binding constraint. The binding constraint is default placement.

Google pays for it. In 2022 the company paid an estimated $26B in traffic acquisition costs to be the default search engine across browsers and devices, of which something above $20B went to Apple for Safari. Add Chrome, which Google owns outright at roughly 65% browser share, and Android, which it acquired in 2005 for about $50M and now runs on the large majority of the world's smartphones. Google does not merely win the search market. It buys and owns the doors.

A judge has now said so. In August 2024 Judge Amit Mehta ruled that Google unlawfully maintained a monopoly in general search, specifically through those exclusive default agreements. The September 2025 remedies decision declined to order divestiture of Chrome or Android, and — critically — allowed the payments to continue on a non-exclusive basis while requiring some search and syndication data sharing with qualified competitors. That outcome was far better for Google than feared, and it left the economic mechanism intact.

The second moat layer is a genuine data feedback loop: more queries produce more click and refinement data, which produce better ranking, which produce more queries. The third is now silicon. Google has designed its own TPUs since 2015 and is the only hyperscaler training frontier models predominantly on hardware it controls, which gives it a structurally lower cost per token than competitors paying Nvidia's margin.

I call the moat contested rather than wide for one reason. All three layers protect a query box, and a meaningful share of the questions people used to type into a query box are now typed into a chat window instead.

The two decisions that made Alphabet were both distribution purchases, not product decisions. Buying Android in 2005 for approximately $50M is the highest-return acquisition in corporate history by any sane measure — it bought Google the operating system on the majority of the world's phones and pre-empted a future where Microsoft or Apple controlled every mobile default. Buying YouTube in 2006 for $1.65B, then widely considered an overpay for a copyright liability, produced a business now generating over $36B of advertising revenue plus subscription income.

Almost everything after that is a weaker record. Motorola was bought for $12.5B in 2011 and sold to Lenovo for $2.9B three years later, with Google keeping the patents — an expensive way to buy litigation insurance. Nest, Fitbit and the hardware programme generally have not produced a franchise. Other Bets has absorbed something in the order of $30B+ of cumulative losses, of which Waymo may eventually justify a large portion and the rest almost certainly will not.

The current allocation question dwarfs all of it. Alphabet is committing $75–85B a year to AI infrastructure — more than the entire market capitalisation of most S&P 500 constituents — to defend a business that the same infrastructure is helping to disrupt. It is simultaneously buying back roughly $60B+ of stock a year and, since 2024, paying a dividend. Doing all three at once is only possible because search generates cash faster than management can find uses for it. That will not always be true.