Earnings Highlights: Alphabet (GOOGL) | Q3 FY2026 — Cloud Up 103%, TPUs Now Sold as Systems, a $561B Backlog, and Capex Set to Rise Significantly in 2027
1. Executive Takeaway
Alphabet delivered a quarter of "great revenue, earnings dragged down by non-operating items, and cash flow starting to tighten." The stock fell 10% after hours on an EPS miss and negative free cash flow, but for the AI infrastructure supply chain, three signals from this call matter far more than EPS:
Compute demand is still accelerating, and Google's own capacity can't keep up. Cloud backlog reached $561B, with roughly 55% to be recognized over the next 24 months. The company has started renting third-party compute as "bridging capacity" and openly said this will squeeze Cloud margins in the near term. Supply shortage is no longer an adjective; it is a cost line written into the outlook.
The TPU business model has changed. TPUs used to be Google Cloud's internal weapon that customers could only "rent"; now Google is "selling" TPUs directly as systems, and also delivering them through partners such as Oracle. That moves Google one step from cloud service provider toward "AI systems vendor."
The capex ceiling hasn't arrived yet. FY2026 capex guidance stays at $195B–$205B, but a "significant increase" in 2027 has already been flagged, and the balance sheet carries $924B in future purchase commitments. This is the sentence upstream optical, server, cooling and power suppliers care about most.
2. Revenue and Financials
Metric | Q3 FY2026 | YoY |
Revenue | $127.58B | +24.6% (QoQ +6.5%) |
Gross margin | 59.7% | +0.1 pp |
Operating income | $40.58B | +29.9% |
Operating margin | 31.8% | — |
Net income to common | $26.17B | -25.2% |
Diluted EPS (GAAP) | $2.12 | -26.1% |
Segment revenue
Segment | Q3 FY2026 | YoY |
Google Search & other | $63.35B | +12% |
Google Cloud | $30.71B | +103% |
Subscriptions, platforms & devices | $14.29B | +11% |
YouTube ads | $11.21B | +9% |
Google Network | $7.46B | +1% |
Three numbers deserve a closer look:
Operating income grew 29.9%, faster than revenue (+24.6%). With R&D up 36% and depreciation still climbing, operating margin kept expanding, meaning Cloud's scale economics are now feeding the whole company.
The 26% EPS drop comes from non-operating items. OI&E swung from +$12.76B a year ago to -$8.22B, including an $8.66B loss on equity securities, which the company says cut EPS by $0.55. Add it back and EPS is about $2.67, still below last year's $2.87 because the prior-year quarter booked an investment "gain." In other words, nearly all of the EPS decline is investment-portfolio volatility, not a weaker core business.
Cloud operating income was $11.12B at a 36.2% margin, versus $3.59B and 23.7% a year ago. Cloud now earns close to 30% of what the Services segment does ($39.16B, 40.7% margin).
▲ Alphabet FY2026 Q3 key financials (revenue, gross margin, operating margin, net income, EPS, segment revenue, FY2026 capex guidance). (Image source: Simple Tech Trend original; data from SEC 8-K earnings release and company IR)
Cash flow is the real pressure point this quarter: operating cash flow was $42.29B, but capex was $51.36B, leaving free cash flow of -$9.07B and just $19.75B on a trailing-12-month basis. Beyond capex, two extra cash drains stood out: inventory built for TPU system sales (up from $2.44B at year-end to $13.31B), and a $5.2B cash payment for a 2018 EC fine. Long-term debt rose from $46.55B at year-end to $125.10B. Alphabet is now funding this investment cycle with debt and new equity (about $12.4B raised through the ATM program this quarter).
3. Technology and Business Highlights
Cloud: demand far exceeds capacity
On his first earnings call as Cloud CEO, Thomas Kurian offered a string of numbers that show how real the demand is:
Google Cloud now has 18 product lines each above $1B in annual revenue.
Over the past 12 months, more than 2,600 customers each used at least 100 billion tokens, and nearly 75 used at least 10 trillion.
Nearly 90% of the Fortune 100 use Gemini Enterprise, and Google has sold more than 100 million knowledge-work seats across its AI services.
First-party model APIs now process nearly 32 billion tokens per minute, up 40% QoQ and more than 4x a year ago.
These numbers point to one thing: AI inference is becoming a steady, predictable enterprise workload rather than one-off model-training projects. Inference is long-tail demand that runs every day. For data centers that means "racks stay powered on"; for optical communications it means "interconnect bandwidth keeps scaling."
TPU: from renting compute to selling systems
The most important structural change this quarter is that the CFO named TPU system sales a "meaningful contributor" to Cloud growth. The newly disclosed segment definition is blunt: Cloud's "product revenue" comes primarily from TPU system sales.
The customer list came out in one go: beyond Anthropic (buying directly and via Broadcom) and Apple (scaling foundation-model pre-training on the latest TPUs), there are frontier-model startups such as Physical Intelligence, quant trading firms such as Citadel Securities and Jump Trading, universities such as Caltech, and public-sector service providers. Nearly 100 customers use TPUs today, and Google has signed partners including Oracle to deliver TPU systems to third parties, "with more in the works."
On specs, the eighth-generation TPU splits into two chips: TPU 8t targets training with up to 2.7x the performance of the prior generation; TPU 8i targets inference with 80% better price-performance. We broke down the architecture of both chips in Google's 8th-Gen TPU Dual Architecture: How TPU 8t and TPU 8i Tackle the Agentic AI Compute Bottleneck. Another move to lower switching costs is TorchTPU: built with Meta, it lets PyTorch models train and deploy directly on TPUs, taking down the "TPUs are only for JAX" ecosystem barrier.
TPUs are no longer just Google Cloud's differentiator; they are a product Google uses to compete with NVIDIA for the "full-system budget."
Search and Gemini: record usage, monetization still catching up
Search & other grew 12%, a clear slowdown from 17% last quarter. Management's breakdown: FX was the single largest factor, followed by the EU ending its de minimis exemption, which hit some APAC retail advertisers, and the World Cup pulling media budgets into June in the prior quarter.
The fourth factor is the one to watch long term: AI Mode queries are growing fast, and AI Mode monetizes at a lower rate. AI Mode queries are on average three times longer than traditional searches, one in five uses voice, images or documents, and most are "incremental" queries. Google's response is to take AI Mode ads to international markets and launch multi-turn ads in the U.S. In short, usage is there; the monetization model is still catching up.
On models, Gemini 4 Argon ramped first-week developer usage 3.5x faster than Gemini 3 within two weeks of launch, and the even stronger Carbon was announced on call day, rolling out first to cyber defenders. The Gemini app has reached 1 billion users.
4. Management Outlook
Alphabet doesn't give revenue guidance, but the directional guidance on this call says more than numbers would:
FY2026 capex stays at $195B–$205B; about $132B has been spent through Q3, leaving roughly $63B–$73B for Q4. This range was just raised last quarter; for context see Alphabet FY2026 Q2 Earnings Highlights.
2027 capex will "increase significantly", with details on the Q4 call. This is "direction confirmed, number withheld" language: high certainty, unknown magnitude.
Cloud margins face near-term headwinds: because supply is tight, Google began using third-party compute at the end of Q3 and will ramp it in Q4. That says two things at once: demand exceeds owned capacity, and the 36.2% margin is likely not the Q4 peak.
TPU system sales keep scaling in Q4, with a "meaningful revenue ramp" in 2027. The word is "expect," not "target," implying real order visibility.
Q4 FX headwind slightly larger than Q3, and Services laps the Search acceleration that began in Q3 last year.
ROIC is expected to stay "meaningfully above our cost of capital," but the CFO was explicit that infrastructure spending and TPU inventory builds will keep pressuring free cash flow.
In STT terms: Google is choosing to fill capacity first and worry about cash flow later. It is the same playbook as Meta's Q3 2026: record $34.9B quarterly capex and negative free cash flow. The hyperscalers have collectively entered the "cash flow yields to compute" phase.
5. Supply Chain and Customer Signals
What this call means for the optical and AI infrastructure supply chain breaks down into four signals:
Signal 1: $924B in purchase commitments. The CFO disclosed $924B in future purchase commitments at the end of Q3, plus $152B in pre-commencement leases and other obligations (both are contractual maximums, some with price and volume flexibility). It spans technical infrastructure capacity, inventory, leases and energy. For upstream suppliers, this is far longer visibility than any single quarter's capex. Customers paying up front and signing early to lock capacity is the same signal we saw in Corning FY2026 Q3: customers prepay to fund capacity.
Signal 2: TPU system shipments = rack-level interconnect demand. TPUs are sold as "systems": not just chips, but a bundle of compute, storage, networking and system components. TPU scale-out has always relied on Optical Circuit Switching (OCS) and heavy optical interconnect; for the architecture see OCP 2026: Inside Google's Virgo Network and the Campus-as-a-Computer. As TPUs move from internal use to external sales, even delivered by partners like Oracle, demand for OCS, optical modules and AECs no longer follows only Google's own data centers, but every TPU customer's facilities.
Signal 3: inventory jumped from $2.4B to $13.3B. Inventory grew by about $10.9B in nine months, which the company attributes to TPU system sales. The mix isn't disclosed, but system-level inventory typically includes accelerators, switches, optical components and mechanicals. Upstream shipments may reflect this TPU ramp earlier than Google's revenue recognition does.
Signal 4: Anthropic's dual-track sourcing through Broadcom. Kurian noted that Anthropic works with Google "directly and via Broadcom," one of the few public confirmations of Broadcom as a TPU delivery channel. Together with Apple scaling pre-training on TPUs, the TPU customer base is clearly more diversified than a year ago.
6. Conclusion
Three metrics to track next quarter:
Cloud operating margin: once third-party compute ramps in Q4, how far does 36.2% fall? A small decline would mean TPU system sales carry a healthier margin structure than the market assumes.
The actual 2027 capex number (on the Q4 call): whether "significant" means 20% or 40% directly sets the 2027 ceiling for the optical and server supply chain.
How fast external TPU customers and delivery partners grow: nearly 100 customers, and who comes after Oracle? Every new delivery partner adds a TPU interconnect growth curve independent of Google's own data centers.
The one-line takeaway: Alphabet's EPS was dragged down by its investment portfolio, but the real news is that TPUs have officially become an externally sold hardware business. Google traded negative free cash flow for a $561B backlog and another round of compute buildout in 2027. For the supply chain, that is worth re-modeling far more than any single quarter's EPS.
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments