GOOGL

The trailing multiple in the panel understates what investors are actually paying, because a single quarter of unrealised marks on equity holdings inflated reported earnings by a factor that has nothing to do with operations. The forward multiple, which strips that distortion out, sits in the upper half of the five year band even after a double digit drawdown from the high, so the market is still assigning a premium to Alphabet's operating earnings while the headline reads cheap.

The architecture is one very large profit engine funding two very different consumers of capital. Search and its adjacent surfaces generate the cash, at a segment margin that has survived a decade of predictions of its collapse, and that cash pays for the compute that Cloud sells and the compute that Gemini consumes. Note that the shared AI research and development sits outside all three segments, in Alphabet level activities, so the reported Services margin flatters the true cost of the model programme and the reported Cloud margin flatters the true cost of the infrastructure that serves it. Anyone modelling this business needs to hold both facts at once.

Operating leverage originates in two places. The first is Search, where incremental query volume carries almost no incremental cost once the index and the serving infrastructure exist, and where AI Overviews and AI Mode have so far expanded usage without collapsing the advertising unit economics. The second is Cloud, where the segment margin has been climbing steadily as fixed infrastructure absorbs more workload, and where Alphabet sells the same silicon it uses internally. That vertical integration is the structural difference from every rival except Amazon: the tensor processing units that train Gemini also generate third party revenue, so the fixed cost of the model programme is amortised across two demand curves rather than one.

The Network line is a managed decline and should be read as such. Subscriptions are the quiet compounder, growing fastest of the advertising adjacent lines and carrying recurring revenue characteristics that the advertising business does not have. Other Bets remains a cost centre with one asset, Waymo, that has plausible standalone value and no path to consolidated profitability inside the forecast horizon.

Rate the mechanism, not the brand. The YouTube network earns the top rating because it is the only pillar here where the advantage is created by users on both sides and cannot be bought: creators go where the audience and the payout are, advertisers go where the creators are, and no competitor has assembled that loop at scale despite a decade of well funded attempts. The data flywheel rates one notch lower because generative interfaces genuinely change what data matters. Two decades of click logs are worth less when the interface is a conversation, and the remedies order now requires Alphabet to share categories of search and interaction data with rivals, which converts a private asset into a partially public one.

The silicon position is underrated by the market and is the pillar PGS would defend hardest. Designing the accelerator, owning the data centres, training the model and selling the capacity is a cost structure competitors cannot replicate by writing a cheque, and it shows up as a Cloud margin that keeps expanding through a period of extreme capital intensity. Its erosion risk is low because the barrier is accumulated engineering rather than a contract.

Distribution is the weak pillar and the one most exposed to legal outcomes. Exclusive default arrangements are now prohibited and the payments that sustained them are under appellate review from both directions. The honest reading is that distribution was always a purchased moat, and purchased moats erode when the purchase becomes illegal. What protects Alphabet is that the product wins most choice screens on merit, which the September ruling effectively conceded.

Sequencing decides whether any of this matters. Backlog conversion has to work first, because it is the only driver that turns the current capital programme into recognised revenue on a timeline the market will tolerate. Everything else is downstream of it. If capacity arrives and converts, the spending is a build cycle with a visible end; if it slips, the same spending is a write down waiting to be scheduled, and the equity story changes from compounder to capital trap regardless of how well Search performs.

Token consumption is the second gate and the more interesting one, because it is where AI demand becomes a metered utility rather than a subscription. Throughput has been compounding at a rate that few businesses of this size have ever recorded, but per token pricing is falling as fast as volume is rising, which is why management keeps emphasising cost efficiency in each model release. The revenue outcome depends on whether volume growth outruns price decay, and one quarter of disclosure is not enough to settle that.

Search monetisation of AI formats sits third by dependency, not by importance. It is the largest pool by far, and the current evidence, with query volumes at record levels and Search revenue growing faster than the Services average, argues the formats are additive. Subscriptions and Waymo are optionality. Neither changes the investment case inside two years, and Waymo carries capital demands of its own that compete with the compute programme for the same balance sheet.