Alphabet Deep Dive: The Franchise Is Exceptional; the Price Assumes the New Capital Is Too

Where the profit sits matters for what follows. In FY2025 the Google Services segment earned $139.4 billion of operating income, Google Cloud $13.9 billion, and the remainder was lost by Other Bets, which includes Waymo, at $7.5 billion and by the costs Alphabet holds at the corporate level at $16.8 billion. Search and other revenue was $224.5 billion, up 13.4%; YouTube advertising $40.4 billion, up 11.7%; subscriptions, platforms and devices $48.0 billion, up 19.1%; Google Network $29.8 billion, which fell 1.9%; and Cloud $58.7 billion, up 35.8%. The growth is broad, the profit is concentrated, and the fastest-growing line is the one that needs the most capital.

The capital is where the picture changes. Operating cash flow was $164.7 billion in FY2025 and capital spending $91.4 billion, which is 22.7% of revenue against 9.6% in FY2021; free cash flow was $73.3 billion, of which $45.7 billion went to repurchases and $10.0 billion to dividends. Over the last four quarters capital spending rose to $132.4 billion and free cash flow fell to $53.3 billion, and in the second quarter alone spending was $44.9 billion against operating cash flow of $39.1 billion. Depreciation, the cost that turns this spending into reported profit, was 5.7% of revenue over the last four quarters, and it grew 42% in the second quarter from a year earlier. Equipment bought this year will be depreciated over the next several years, so depreciation trails spending, and the margin of the next few years carries the spending of this one.

The price has moved with the earnings and then ahead of them. The Class A close was $88.23 at the end of 2022 and $313.00 at the end of 2025, and it was $340.92 at the last close, on 09-29-2026. Earnings per share excluding gains on equity securities went from $4.76 to $9.27 over the same three years, so the shares rose by more than the earnings did: the multiple of trailing earnings excluding gains went from 18.5x at the end of 2022 to 33.8x at the end of 2025, and it is 32.6x on the last four quarters at the last close. Earnings roughly doubled and the multiple a holder pays for each dollar of them roughly doubled as well. That is what a market does when it decides a franchise has more growth in it than it had assumed, and it is the reason the question in this report is about the price and not about the business.

What the price now rests on is a change in the kind of company. Alphabet earned about 50% in FY2025 on the capital that built Search and its other franchises, and it is now adding capital at a rate that has no precedent in its history. It has guided to $195 billion to $205 billion of spending in 2026 and more in 2027, it has a backlog of $519.5 billion that it says is mostly long-term and committed, it has paid for part of this with new equity and with debt of about $100 billion against $16 billion a year ago, and its free cash flow was negative in the second quarter. The later chapters test that change one line at a time. This chapter's point is narrower: at $340.92 the market has already paid for the transition to go well, and the burden of proof has moved from the franchise to the capital.

Reported net income of $112.2 billion and diluted earnings of $9.11 a share include $99.0 billion of net gains on equity securities before tax, almost all of them unrealized gains on holdings in SpaceX and in one private company, which the quarterly filing does not name. The company states the effect after tax: the gain raised the provision for income tax by $21.9 billion, net income by $77.1 billion and earnings per share by $6.26. Net income without the gain is therefore about $35.1 billion, and earnings per share about $2.85. The before-tax and after-tax figures describe the same gain and should not be confused: $99.0 billion is what was booked above the tax line, and $77.1 billion is what reached net income. For earlier periods, where the company gives no such figure, our history of multiples removes the gains at an assumed 22% tax rate; for this quarter that assumption would give $77.2 billion, close to the company's own number, and the two are kept apart. The gains are real accounting profit and they are not operating earnings. They are marks on stakes that have not been sold, part of them restricted from sale, and they sit below operating income, which is why operating income of $40.8 billion is the better measure of what the business earned. We carry no such gains in any forecast of earnings and we value the holdings separately, at carrying value less the tax on the gain and with a haircut, in the bridge from enterprise value to equity. The quarter's earnings without the gain were about $2.85 a share, and the trailing four quarters, on the same treatment, about $10.45, which puts the price at 32.6x earnings on that measure. The third line of the table is the one that concerns us most: in the same quarter the business generated $39.1 billion of operating cash flow and spent $44.9 billion on capital, so free cash flow was $5.9 billion negative.

The gap between profit and cash has two parts and they point in opposite directions. The first is depreciation, which will rise. Operating income is struck after depreciation of $7.1 billion in the quarter, and depreciation follows capital spending with a lag that the company's own disclosure describes: much of the equipment bought this year is not yet in service, and servers and network equipment are depreciated generally over 6 years. Both the level of capital spending and the delay in depreciating it flatter the margin of any year in which spending is rising faster than depreciation, which is every year from 2024 to 2029 in our models. Management said plainly on the call that the infrastructure investments will keep putting pressure on the income statement through higher depreciation and data center operating costs such as energy. We read the current margin of 34.0% as a high-water mark that our base case takes down by several points over the next three years and a good deal further by 2035, for the reason that the asset base the company is building is several times what it is depreciating today.

The second part is tax, and it points the other way for a while. Cash taxes paid in FY2025 were $21.5 billion against a provision of $26.7 billion, 81% of it, and the 2025 tax law made permanent a first-year deduction of 100% of the cost of qualifying equipment acquired after January 19, 2025, taken in the year the equipment is placed in service, and allows domestic research and software development costs to be deducted in the year they are paid. When a company places hundreds of billions of dollars of qualifying equipment in service and can deduct it at once, its cash taxes fall below its book taxes for as long as the equipment placed in service keeps growing, and the difference narrows as the deductions of earlier vintages run off against the book depreciation that follows, which is gradual and not a single reversal. The deduction follows the date equipment enters service and not the date it is ordered or paid for, and equipment under a binding contract signed on or before January 19, 2025 does not qualify for the full first-year rate, so much of the large balance of assets not yet in service is deferred benefit and not a deduction already taken. We model that with a schedule calibrated, loosely, to the deferred tax liability the company reports on its equipment, and we bound what we cannot observe: the share of equipment that qualifies, the rate at which the deductions save tax, and the domestic income they meet. In the base case the timing difference is worth $77.1 billion in present value to the firm, about $6.28 a share; it is given back after 2029 as the equipment's deductions run off, and our terminal value carries none of it. A second tax change points the other way: from 2026 the deduction for income from foreign sales falls from 37.5% to 33.34%, which takes the effective federal rate on that income from 13.1% to 14.0%. The deduction was worth $3.9 billion in FY2025, a small number, and the reason we test a long-run cash tax rate of 21% against the 16.8% we use is that the lower rate is an outcome of the current law, not a constant.

Three smaller items complete the picture. Stock-based compensation was 6.2% of revenue in FY2025 and is an operating cost in our cash flows and not an add-back, which is the single largest difference between our free cash flow and the figure the company reports: over the trailing year free cash flow after stock compensation was $25.1 billion. The company has announced that it will meet the tax on employee awards through an at-the-market equity program, which moves a cash cost of the awards into the share count; we count it once, as an expense. Legal and regulatory items are in operating income this year: the quarter's general and administrative expense rose 24%, partly for charges for legal matters, and a Swedish court awarded a price comparison service damages that Alphabet has appealed and recognized in the quarter. The filing describes the award in two places, as about $2.1 billion in the legal note and as $1.5 billion of principal in general and administrative expense with $581 million of interest and costs in other income in the discussion of results; we use the second, which puts 1.3% of the quarter's revenue in operating income. We leave them in, because a company of this size in this regulatory position will have such charges in most years. Finally, the balance sheet: non-current deferred tax liabilities rose from $0.9 billion at the end of 2025 to $22.8 billion at June 30, almost all of it tax the company will owe on the equity gains if it realizes them, which is why we deduct that tax from those holdings and do not treat the gains as free.

Across that grid the shares are worth $29 to $50, against $66.48 in the bear case and $340.92 at the last close. The range is a mechanical result and not a forecast, and it is worth reading for what it does not include: no management response, no sale of the stakes in SpaceX and in a private company beyond what the bridge already counts, and no rescue by a new business. It shows what a finite grid of assumptions gives if the program were written down against a business that kept its Search and YouTube franchises and earned only its cost on everything it built since 2026. It is not a proven floor and not a liquidation value: a worse margin, a higher discount rate or a loss of the franchises themselves would give less.