Alphabet is a franchise in the middle of a violent identity change: the cash cow that funded two decades of buybacks has become a capital-hungry AI infrastructure builder that now borrows and dilutes to fund the build, and the equity is being priced like a mid-growth stock because the market has not yet decided which company it is holding.
The most important recent development is the August 5 leadership shuffle and the simultaneous departure of four senior AI researchers, Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le, who left to found Discovery Loop. Alphabet invested in their new company and committed Google Cloud as its compute provider, a move that converts a talent loss into an ecosystem anchor but also signals that the company's most senior technical thinkers no longer see the best path forward as being inside the public-company structure. The stock dropped roughly 5% on the news, and the episode arrived in the same week the flagship Gemini 3.5 Pro model remained unshipped after slipping past its June launch window.
The central tension is capital allocation. Full-year 2026 capex guidance stands at a range up sharply from the prior plan. Free cash flow went negative for the first time since the 2004 IPO. The company has already raised roughly $50 billion in equity and $25 billion in bonds to close the gap. The bear case is that this spend is a sunk cost with no guarantee of returns. The bull case is that the $514 billion cloud backlog provides a contracted pipeline that makes the spend a prepayment for future earnings rather than a bet.
The catalyst to watch is the Q3 earnings report and any update on Gemini 3.5 Pro's general availability. A shipped flagship paired with continued cloud backlog growth would resolve much of the uncertainty that is currently compressing the multiple.
Alphabet operates as a holding company whose economics are dominated by one business: Google. Google Services, the segment that includes Search, YouTube advertising, Network, and subscriptions, generated $94.5 billion in revenue in the most recent quarter, up 15% year over year, and remains the profit engine that funds everything else at Alphabet. Search revenue alone grew to $63.27 billion. YouTube advertising rose to $11.06 billion, helped by the FIFA World Cup, the most-watched in YouTube history. Subscriptions and platforms, including YouTube Premium and YouTube Music, grew 15% year over year. The strategic logic of the past decade has been to use the cash generated by Search to fund long-horizon bets: Google Cloud, Waymo, DeepMind, and the Other Bets portfolio. That logic is still intact, but the funding mechanism has changed. In the prior fiscal year, Alphabet repurchased $45.7 billion of its own stock. In the first half of the current year, it repurchased none. Instead it sold roughly $50 billion across common and preferred shares, plus a further $24.8 billion of debt, in a single quarter. The company that used to be the market's most reliable cash recycler is now a net capital distributor, and that shift is the single most important structural fact for an investor evaluating the stock.
Google Cloud is the second pillar and the one that has changed fastest. The segment generated $24.8 billion in revenue in the most recent quarter, up 82% year over year. Operating margin expanded to 35.6% from 20.7% a year earlier. Cloud is now close to a fifth of total revenue, up from a rounding error a few years ago. The backlog, a gauge of contracted future revenue, reached $514 billion, up more than $50 billion in a single quarter, with just over half expected to convert to revenue within two years. That backlog is the bull case in its purest form: it is contracted demand, not a forecast, and it gives the capex program a visible payback path. The strategic question is whether the company can convert that backlog into sustained margin at scale without the compute cost structure eroding the very profitability that funded the build.
The third pillar is the portfolio of Other Bets, of which Waymo is by far the most consequential. Waymo now operates in 14 U.S. cities with a fleet past 4,000 vehicles and more than 500,000 paid rides per week. The company raised $16 billion in an earlier round this year at a valuation of roughly $126 billion. It has stated a target of 1 million weekly rides by year-end. Waymo is still a small fraction of consolidated revenue, but it represents a genuine option on a market that may not exist in a form investors can yet value. The strategic context for the entire company, then, is one of three businesses at three different stages of the S-curve: Search is a mature cash generator, Cloud is a hypergrowth asset with a visible backlog, and Waymo is an early-stage bet on autonomous driving. The portfolio only makes sense if the first business continues to fund the second and the third, and the capital structure shift described above is the signal that the funding relationship has inverted.
The competitive landscape has sharpened in a way that did not exist two years ago. OpenAI and Anthropic have released models that outperform Gemini on several benchmarks, and the delay of Gemini 3.5 Pro, which slipped past its June launch window and remained unshipped as of early September, has raised questions about whether Google can still hold the frontier. The loss of Jeff Dean, who had been Google's chief scientist for much of the modern AI era, and three other senior researchers to Discovery Loop adds a second-order risk: the people who built the company's technical advantage are now building something outside it, with Alphabet's money in the round and Google Cloud as their infrastructure provider. The strategic position is still strong, but the margin of safety around it has narrowed.
The moat at Alphabet is not a single product; it is a stack. At the bottom sits the infrastructure: the data centers, the TPU chips, the networking, and the storage that power everything above. Alphabet has been one of the few hyperscalers to design its own AI silicon in volume, and the TPU ecosystem is now a meaningful differentiator in cloud economics because it gives the company a cost structure on inference that competitors relying on third-party GPUs cannot easily match. The most recent quarter included the first revenue from TPU system sales, a milestone that signals the chip business is moving from an internal cost center to an external revenue line. The middle of the stack is the model layer: Gemini, developed by Google DeepMind, which is the engine behind AI Mode in Search, the Gemini app, and the enterprise AI offerings in Cloud. The top of the stack is the distribution layer: Search, YouTube, Android, Chrome, and the Play Store. Each layer reinforces the others. Search generates the data and the cash that fund the models. The models improve Search and the app. The distribution layer gives the models a user base of billions. The moat is the flywheel, not any single component.
The Search moat is the one under the most direct pressure, and the pressure has two sources. The first is the antitrust remediation regime. In September 2025, Judge Mehta ruled that Google held an illegal monopoly in general search and ordered the company to share its search index, user interaction data, and certain syndication services with competitors, including generative AI companies. The order became legally final in December 2025 and took effect in early 2026. Google appealed in May, and the D.C. Circuit appeal is now in briefing, with the government's cross-appeal filed in July. The practical effect of the remedies, even if the appeal succeeds, is that the data advantage that has compounded Search's quality edge for two decades is now partially available to rivals. That is a slow leak, not a flood, but it changes the long-run trajectory of the moat. The second source of pressure is structural: AI assistants are beginning to intercept queries before they reach the search box. Each query that an AI assistant answers without a click to a search results page is a query that does not generate an ad impression. The 17% growth in Search revenue in the most recent quarter shows the business is still compounding, but the growth rate has decelerated from 19% in the prior quarter, and the direction of travel is the more important signal.
The Cloud moat is different from the Search moat. It is not a data advantage or a distribution advantage; it is a scale and cost advantage, and it is currently the strongest of the three. The 82% revenue growth in the most recent quarter is driven by new customer acquisition, deeper usage from existing customers with commitments exceeded by more than half, and partner-driven transactions. Marketplace volume grew 7x year over year. Nearly 90% of the Fortune 100 use Gemini Enterprise. The backlog of $514 billion is the quantitative expression of that advantage: it means the company has contracted demand that exceeds its current revenue base by roughly two and a half times, and the backlog is growing faster than revenue. That is a signal of sustained future demand, and it is the most concrete evidence in the company's favor that the capex program is a prepayment, not a bet. The risk is that the backlog is concentrated in a small number of large enterprise contracts, and that the compute cost structure could erode the margin as the fleet scales. The 35.6% operating margin in the most recent quarter is a strong start, but it is a margin built on a compute base that is still in the early stages of the buildout.
The Waymo moat is a data and geographic moat. The company has completed more than 20 million trips and driven over 220 million autonomous miles, a safety record that is a genuine differentiator in a market where a single serious incident can reshape the regulatory environment. The expansion to 14 cities this year, including Denver, which tests the system in winter conditions, widens the geographic data set in a way that competitors cannot replicate quickly. The 500,000 weekly paid rides represent a tenfold increase from levels two years ago, and the trajectory toward 1 million weekly rides by year-end is ambitious but not implausible given the fleet growth and the city-by-city rollout cadence. The Hyundai supply agreement, which calls for 50,000 vehicles by 2028, secures the hardware pipeline for the next phase of expansion. The moat is real but narrow: it is a race, not a moat, and the competitors, including Tesla and Amazon's Zoox, are adding cities and vehicles in parallel. The data advantage is the lead, but the lead is measured in months, not years.
The income statement in the most recent quarter tells two different stories depending on which number you look at first. The headline GAAP net income of $112.1 billion includes a $99.03 billion gain on equity securities, the bulk of which is tied to the June public offering of SpaceX and a revaluation of the company's stake in Anthropic. Stripping out that one-time item, underlying net income was roughly $35 billion. Underlying diluted EPS was approximately $2.85, up 23% year over year but slightly below the consensus estimate. The operating story is clean. Revenue of $119.8 billion grew 24% year over year. Operating income grew to $40.8 billion.
The cash flow statement is where the story changes character. Operating cash flow in the quarter was $39.1 billion. Capital expenditures of $44.9 billion pushed free cash flow into negative territory for the first time since the 2004 IPO. The capex figure more than doubled year over year and reflects the scale of the AI infrastructure buildout. The company ended the quarter with $242.5 billion in cash and marketable securities, which is a fortress balance sheet, but the trajectory is what matters. Full-year capex guidance stands at roughly $200 billion, up from the prior plan. The CFO indicated that spending is expected to keep rising into the next year with no ceiling. In the prior fiscal year, Alphabet generated $164.7 billion in operating cash flow. It spent $91.4 billion on capex, leaving $73.3 billion in free cash flow. The swing from positive FCF in the prior year to a pace that implies negative FCF for the current year is the most significant financial event in the company's recent history, and it is what drove the stock to its May high and then off it.
The financing section of the cash flow statement captures the response to that swing. In the most recent quarter, Alphabet raised $49.6 billion in net proceeds from the issuance of common stock, Class C stock, and preferred shares. It issued $24.8 billion in new debt, roughly doubling long-term borrowings to about $98 billion in a matter of months. The company also put in place an at-the-market program to sell up to a further $40 billion of stock, none of which had been drawn at quarter-end. In August, the company completed a $25 billion ten-tranche bond offering with maturities stretching to 2066. It drew approximately $115 billion in investor orders, an oversubscription of more than four times. The bond deal carried Aa2/AA+ ratings, which is a signal that credit markets have not yet priced in any meaningful risk to the company's ability to service the debt. The financing profile of Alphabet in the current year is unrecognizable from the financing profile of the prior year, and the market has been slow to reprice the stock for that change.
The segment mix is shifting in a direction that matters for the long-run multiple. Google Services generated $94.5 billion in revenue in the most recent quarter, up 15%, and produced the majority of the operating profit. Google Cloud generated $24.8 billion, up 82%. It produced $8.8 billion in operating income, up from $2.8 billion a year earlier. Cloud's share of consolidated revenue rose to approximately one fifth from about one seventh a year earlier. Advertising, the traditional core, still generated the bulk of revenue. Its share of the total declined to 68% from a higher prior level. That mix shift is the financial expression of the strategic repositioning described earlier, and it is what makes the company harder to value on a single multiple. A pure search advertising business at a high operating margin is a different asset class than a company that is also building a cloud infrastructure business at a 35.6% segment margin and a robotaxi fleet that is still pre-revenue at the consolidated level.
The forward picture for Alphabet in the current year and the next is dominated by one variable: the return on the capital being deployed. Full-year capex of $195-205 billion, rising into the next year, is a number that has no precedent in the company's history. The management commentary on the earnings call framed the spend as a response to demand, and the $514 billion cloud backlog provides the quantitative anchor for that claim. But the backlog is not the same as the demand that justifies the full capex program. The backlog represents contracted compute and enterprise AI workloads, and it is growing, but the capex program also includes spending on general-purpose data centers, TPU fabrication, and the broader infrastructure that supports the company's internal AI workloads, including the Gemini model training runs. The question that the market has not yet resolved is what fraction of the $200 billion spend is directly attributable to contracted backlog versus what fraction is a bet on future demand that has not yet been contracted.
The execution risk around the model layer is the second variable. Gemini 3.5 Pro, the flagship model that was promised at the May developer conference, remains unshipped as of early September. The delay, which Bloomberg attributed in July to coding performance that fell short of internal targets, has now become a recurring theme in the coverage of the company's AI position. The competitive implication is direct: OpenAI and Anthropic have released models that outperform Gemini on several public benchmarks, and each month that 3.5 Pro is not in general availability is a month that enterprise customers are evaluating competitors. The release of Gemini 3.6 Flash and two other variants in July was a real product event, but it did not answer the question that 3.5 Pro was supposed to answer. The start of Gemini 4 pre-training, announced at the same time, is a signal that the team is moving forward, but it also means that the next flagship is still further out. The execution risk here is not that the model is bad; it is that the cadence has slipped, and in a market where the competitive frontier moves on a monthly basis, cadence is the metric that matters.
The talent and organizational risk is the third variable, and it is the one that is hardest to model. The August 5 reshuffle, in which Demis Hassabis moved to a chair and chief scientist role while Koray Kavukcuoglu took over day-to-day operational control of DeepMind, and in which four senior researchers left to found Discovery Loop, is a structural event, not a personnel one. The departure of Jeff Dean, who had been the company's chief scientist and was involved in the technical direction of Gemini, TPU, and the broader AI infrastructure stack, is the single most significant talent loss in the company's AI history. The fact that Alphabet is an investor in Discovery Loop and that Google Cloud is its infrastructure provider mitigates the risk somewhat, because it means the company retains a financial and technical connection to the work. But it also means that the next generation of breakthroughs may happen in a company that is, formally, a competitor. The retention of Hassabis, who is the public face of the company's AGI ambitions, is the counterweight. But the arithmetic of one promotion announced and four researchers gone is not a personnel event that resolves itself with time.
The regulatory environment adds a fourth variable that is outside the company's control. The D.C. Circuit appeal of the search monopoly case is in briefing, with both Google and the government appealing aspects of Judge Mehta's September 2025 remedies ruling. The government's cross-appeal, filed in July, argues that the court erred in not requiring a complete ban on default-placement payments. A ruling that goes against Google on that point would change the economics of the Search distribution agreements, which are the source of the company's largest single revenue line. The EU DMA fine of 890 million euros, imposed in July, is a smaller financial event but a larger signal: it is the first significant DMA penalty for a U.S. company, and it means that the regulatory cost of operating in the EU is now a recurring line item rather than a one-time risk. The combined effect of the U.S. and EU proceedings is not a single number that can be added to the P&L; it is a slow change to the rules of the game under which the Search business operates, and the valuation multiple for the Search segment should reflect that change even before the legal proceedings are resolved.
The most direct downside scenario is a capex-to-returns mismatch that becomes visible in the earnings reports of the next year. The $195-205 billion capex program for the current year, rising into the year after that, implies that a very large fraction of the company's future operating cash flow is already committed to depreciation on compute assets that may not generate the revenue needed to justify the spend. If cloud revenue growth decelerates from 82% to a more modest 35% range as the base grows, and if the backlog does not convert to revenue at the pace management has implied, then the capex program becomes a drag on free cash flow rather than a prepayment for future earnings. In that scenario, the stock would likely trade at a multiple in the mid-teens on forward earnings, reflecting a growth profile that is no longer at the frontier of the hyperscaler group. The balance sheet is strong enough to absorb the spend without financial distress, but the equity holders would bear the full cost of the lower return on invested capital.
The second downside scenario is a competitive loss in the model layer that becomes self-reinforcing. If Gemini 3.5 Pro ships in a form that does not close the gap with the best models from OpenAI and Anthropic, and if Gemini 4 follows with a similar gap, then the enterprise AI customers that are currently evaluating the company's offering would shift their commitments to competitors. The cloud backlog of $514 billion is the buffer against that scenario, because it is contracted demand, not a forecast. But the backlog is concentrated in a small number of large enterprise contracts, and the contracts that are being signed today, which is when the competitive dynamic of the model layer matters most, are the ones that drive the growth profile of the segment in the next two years. The loss of Jeff Dean and three other senior researchers, while mitigated by the Discovery Loop arrangement, is a risk factor in this scenario because the people who were responsible for the technical direction of the model layer are no longer in the company, and the replacement leadership has not yet demonstrated the same track record.
The third downside scenario is regulatory. A loss on the D.C. Circuit appeal that results in a ban on default-placement payments would change the economics of the Search business in a way that is difficult to offset. The default agreements, particularly the arrangement with Apple, are the source of a large fraction of the Search revenue, and a prohibition on them would force the company to compete on quality alone in a market where the distribution advantage has been a structural part of the moat. The data-sharing remedies that are already in effect are a smaller, slower leak, but they are real, and they are not subject to the same degree of legal uncertainty as the payment ban. The EU DMA penalties are a smaller financial event but a larger signal of the direction of regulatory travel, and the cost of compliance is a recurring item that does not go away. The combined regulatory risk is not a single number, but it is a discount to the long-run multiple that should be applied to the Search segment in any valuation framework that takes the legal proceedings seriously.
The fourth downside scenario is the capital structure itself. The equity raise in the most recent quarter, the $25 billion bond offering in August, and the $40 billion ATM program in place but undrawn represent a meaningful change to the capital structure. The dilution from the equity raise is real and should be factored into any per-share metric. The interest cost of the new debt, at coupons ranging from 4.5% to 6.5%, is a new recurring expense that did not exist in the company's financial profile a year ago. The ATM program is a live overhang, because it represents a potential source of additional dilution that management can activate at any time, and the market's reaction to an ATM drawdown would likely be negative. The combination of dilution, interest cost, and the ATM overhang is a set of frictions that the market has not fully priced into the stock, and they represent a floor on how cheap the stock can be relative to the company's earnings power.
The valuation of Alphabet is complicated by the fact that the company is now three businesses with three different capital profiles, and the trailing earnings multiple is distorted by a one-time equity gain. A sum-of-the-parts approach is the most useful framework here. It values the Search and advertising business, the Cloud business, and the portfolio of Other Bets separately, then sums the results and subtracts the net debt and applies the dilution from the 2026 equity raises.
For the Search and advertising business, the relevant metric is operating income. The segment generated the majority of the company's $40.8 billion operating income in the most recent quarter, and the run rate annualizes to approximately $140 billion. A mature advertising business with a 34% operating margin and a dominant distribution position is fairly valued at a multiple in the range of 12x to 15x operating income. The lower end of that range reflects the data-sharing remedies that are in effect and the uncertainty around the D.C. Circuit appeal. The higher end reflects the scale of the business and the fact that the regulatory risk, while real, has not yet produced a material change to the revenue base.
For the Cloud business, the relevant metric is revenue, because the segment is in a hypergrowth phase and the margin profile is still maturing. The most recent quarter run rate of $24.8 billion annualizes to approximately $100 billion. A cloud business growing 82% year over year with a 35.6% operating margin is a growth asset. The appropriate multiple is a revenue multiple in the range of 4x to 6x, which produces a value in the range of $400 billion to $600 billion. The backlog supports the higher end of that range, because it is contracted demand that provides a floor under the revenue trajectory. The risk that argues for the lower end is the concentration of the backlog in a small number of large contracts and the uncertainty around the compute cost structure as the fleet scales.
For the Other Bets portfolio, the most meaningful component is Waymo, which raised $16 billion in an earlier round this year at a $126 billion valuation. That is the market's current price for the robotaxi business, and it is the most defensible starting point for a sum-of-the-parts valuation. The remaining Other Bets carry a nominal value that is difficult to justify from an earnings perspective, but they also represent options on outcomes that are not yet priced into the consolidated multiple. A reasonable allocation for the full Other Bets portfolio, with Waymo as the anchor, is in the low hundreds of billions. The sum of the three components, at the midpoint of each range, is approximately $2.8 trillion. Against that, the company has approximately $98 billion in long-term debt. It holds $242.5 billion in cash and marketable securities, a net cash position that adds to equity value. The equity raises and the potential ATM drawdown represent dilution that reduces the per-share value. The current market capitalization of approximately $4.1 trillion implies that the market is already pricing in a premium to the sum-of-the-parts value, which is a function of the trailing P/E distortion from the equity gain and the market's incomplete digestion of the capital structure change. The bear case, quantified, is a stock price below $300, reflecting a multiple on forward earnings that is at the low end of the historical range. The base case is a stock price in the mid-$300s, which reflects the midpoint of the sum-of-the-parts framework with the capital structure changes fully priced in. The bull case is a stock price above $440, which reflects the higher end of the multiple ranges for both the Search and Cloud segments, a Gemini flagship that ships and closes the competitive gap, and a cloud backlog that continues to grow faster than revenue. The current price sits below the base case range, which is the basis for the investment thesis that the stock is undervalued relative to its fundamental profile, with the primary risk being that the bear case is the one that plays out.
The investment case for Alphabet is a question of which company the stock is pricing in, and the honest answer is that the market has not yet decided. The trailing multiple near 16x is a number produced by a $99 billion one-time gain and does not reflect the earnings power of the operating business. On forward earnings, after adjusting for dilution from the 2026 equity raises, the stock trades at a multiple that is at the upper end of the big-tech group, not at the discount that the trailing figure suggests. The market is pricing in the capital structure change, the negative free cash flow, and the model layer uncertainty, and it is not fully pricing in the cloud backlog, the cloud operating margin, or the fact that the company has a net cash position after the raises. That asymmetry is the basis for the view that the stock is undervalued at the current price, but it is a view that is conditional on the cloud backlog converting to revenue at the pace that management has implied.
The strongest counterargument to that view is that the capex program is a bet, not a prepayment, and that the backlog is not the same as the demand that justifies the full $200 billion spend. If the compute cost structure erodes the margin as the fleet scales, and if the model layer gap with OpenAI and Anthropic persists, then the cloud business is a growth asset that is being financed at the expense of the Search business, and the equity holders are funding a bet that may not pay off. The departure of Jeff Dean and three other senior researchers is a data point in that direction, because it is the people who built the company's technical advantage choosing to build outside it, with Alphabet's money in the round. The Discovery Loop arrangement mitigates the risk but does not eliminate it, because the next generation of breakthroughs may happen in a company that is, formally, a competitor. The regulatory risk is a second counterargument, because the data-sharing remedies are already in effect and the D.C. Circuit appeal is unresolved, and the long-run multiple for the Search segment should reflect that.
The judgment, stated plainly, is that the stock is fairly to modestly undervalued at the current price, with the base case being a range of $360 to $400 and the primary risk being that the bear case plays out. The cloud backlog is the single most important variable in the thesis, and it is a contracted pipeline that provides a floor under the revenue trajectory that is not present in the trailing multiple. The model layer is the second most important variable, and the delay of Gemini 3.5 Pro is a real risk that the market is pricing in but not overpricing. The capital structure change is a third variable, and the dilution and interest cost are real frictions that should be factored into any per-share metric. The regulatory risk is a fourth variable, and it is a slow change to the rules of the game that should be reflected in the multiple applied to the Search segment. The net assessment is that the stock offers a reasonable entry point for an investor who is willing to hold through the uncertainty of the model layer and the regulatory proceedings, with the cloud backlog as the anchor for the upside and the capex program as the anchor for the downside. The risk is not that the company is broken; it is that the market is right that the capex program is a bet, and that the bet does not pay off on the timeline that the stock is currently pricing in.