Google Cloud +50%, Meta +31%, AWS Triple Digits: Q1 2026 AI Earnings

Meta posted its strongest revenue growth quarter since 2021. Alphabet hit $107 billion in revenue with Google Cloud accelerating past 50% growth. Amazon confirmed its custom AI chip business crossed $20 billion annualized revenue and is growing triple digits. Microsoft's Azure, the most closely watched number, came in at 35% growth — below the 37-38% guidance — but CFO Amy Hood made the constraint explicit: if all new GPUs had gone to Azure instead of being split with Copilot and M365, growth would have been above 40%. The bottleneck is not demand. It is supply allocation.

Four companies. Four different AI infrastructure bets. One common message: demand is outrunning everything built so far, and the spending to catch up is accelerating, not slowing.

Meta: $55.5 Billion, Best Quarter Since 2021

Meta reported $55.55 billion in Q1 2026 revenue, up 31% year-on-year — the strongest growth rate since Q4 2021. EPS came in at $6.82 against consensus of $6.67. The 14-quarter revenue beat streak continues.

The number behind the number: Meta maintained its $115-135 billion 2026 capex guidance rather than raising it. That is slightly disappointing relative to expectations for an increase. But the revenue acceleration tells a simpler story — the AI infrastructure Meta has already deployed is generating returns. Every percentage point of Reels and Feed algorithmic improvement driven by MTIA-powered inference is showing up in ad pricing and engagement.

MTIA Gen 2 (Meta Training and Inference Accelerator) is now handling a significant share of Meta's recommendation model inference across Instagram, Facebook, and Threads. The exact deployment percentage was not disclosed on the call, but management confirmed MTIA is "in broad production across key inference workloads." For Nvidia, this is the clearest signal yet that one of the world's largest AI inference operations is systematically moving off Nvidia GPUs.

Meta also secured a 20-year nuclear power deal with Vistra Energy and confirmed AWS Graviton5 chip access for some workloads — Meta is diversifying its infrastructure across its own silicon, AWS, and third-party compute in a deliberate supply chain hedge.

Alphabet: $107 Billion, Google Cloud at ~50% Growth

Alphabet hit $107.03 billion in Q1 2026 revenue, up 19% year-on-year, against a $106.9 billion consensus. EPS of $2.63 matched consensus but was roughly 6% below year-ago levels due to the depreciation load from accelerating AI infrastructure investment.

Google Cloud is the story within the story. Revenue came in around $18.4 billion, representing approximately 50% year-on-year growth. This is the second consecutive quarter above 45% growth. Google Cloud is taking real market share, and the mechanism is clear: TPU v6 (Trillium) gives Google vertically integrated inference economics that AWS and Azure cannot match without their own silicon. Gemini 2.5 and Gemini 3.1 are both in production and driving API consumption.

Alphabet also updated its 2026 capex guidance to $175-185 billion — the largest single-year infrastructure commitment in the company's history. The majority goes to data center buildout and TPU capacity. Google is building for 2028 and beyond, not just 2026 demand.

For developers: Google Cloud's 50% growth rate at Alphabet's scale means something structural is happening, not just an AI boom cycle. Gemini 3.1 Flash-Lite at $0.25 per million tokens with 1 million token context is attracting high-volume API workloads. Google Cloud TPU access via Vertex AI is materially cheaper per FLOP than equivalent Nvidia GPU instances on competing platforms for compatible workloads.

Amazon: $177 Billion Revenue, AI Chips Hit $20 Billion

Amazon reported $177.17 billion in Q1 2026 revenue against $177.2 billion consensus. AWS came in at approximately $36.79 billion, representing about 25% growth. Neither number surprised.

The number that matters: CEO Andy Jassy confirmed Amazon's custom AI chip business (Trainium for training, Inferentia for inference) crossed $20 billion in annualized revenue and is growing triple digits year-on-year. In context: this business was essentially zero three years ago. Amazon has built the fastest-growing AI chip business in the world by being the first hyperscaler to bet seriously on custom silicon for third-party workloads.

The practical meaning for developers: Anthropic's Claude models are trained on Trainium 2. Bedrock inference for Claude, Llama, and Titan runs on Inferentia 3. If you are calling Bedrock, your inference is likely running on Amazon silicon. AWS is not just a reseller of Nvidia GPU access — it is building an alternative compute platform, and that platform is now generating $20 billion in revenue.

Amazon also confirmed $200 billion in projected 2026 capex, which makes it the largest infrastructure spender among all five Big Tech companies this earnings cycle.

Microsoft: Azure at 35%, the Supply Story

Microsoft's Azure grew 35% in Q3 FY26 — below the 37-38% guidance range. The miss is the most important data point from the entire earnings cycle.

CFO Amy Hood's explanation: "Had we been able to allocate all of the new GPU capacity to Azure commercial workloads, growth would have been above 40%." Microsoft chose to direct a portion of new GPU procurement toward Copilot and M365 AI features rather than pure Azure cloud capacity. That decision suppressed Azure's reported growth rate while building out the consumer-facing AI product surface.

The implication is important in both directions:

For the bull case: Demand for Azure AI capacity exceeds what Microsoft can currently supply. The constraint is real and confirmed by the CFO. When supply catches up — and Microsoft is spending $80+ billion in 2026 capex to make that happen — Azure growth will re-accelerate. The miss is a supply miss, not a demand miss.

For developers using Azure: The GPU capacity constraint is explicit. Provisioned Throughput Units (PTUs) for GPT-5.5 and GPT-5 access via Azure OpenAI Service remain constrained. Rate limits on Azure AI endpoints are hardware limits, not policy limits. If you are hitting Azure AI rate limits at scale, the solution is not a support ticket — it is a PTU reservation in the right Azure region, booked months in advance.

The Combined $630 Billion Picture

The five companies together — Microsoft ($80B+), Meta ($115-135B), Amazon ($200B), Alphabet ($175-185B), Apple ($65B) — are spending roughly $630 billion on AI infrastructure in 2026. To put that in perspective: the entire US federal non-defense discretionary budget is approximately $700 billion. Five tech companies are spending nearly that much on data centers, GPUs, custom chips, and fiber.

What does this buy? Capacity that will come online in 2027 and 2028. The data centers being built with 2026 capex dollars will start generating revenue in 2027-2028. This is why Microsoft's CFO and Amazon's CFO are both comfortable with the spending pace despite short-term margin pressure — they are building for the demand curve they see in 2028, not 2026.

For developers and infrastructure planners: the capacity crunch of 2025-2026 will materially ease in 2027. If your business model depends on cheap, plentiful GPU access, the structural position improves significantly in two years. If you need that access today, the constraints are real and the queue is long.

What Each Call Signals for AI Developer Costs

Google Cloud (best developer signal right now): 50% Cloud growth at competitive prices with TPU access suggests Google is the most aggressive at buying developer market share through pricing. If you can run on TPU-compatible frameworks (JAX, TensorFlow, or PyTorch with XLA), Google Cloud's economics are the most favorable in the market.

Amazon (best custom silicon momentum): Trainium 2 and Inferentia 3 at $20B revenue growing triple digits is the clearest proof that custom silicon is a viable alternative to Nvidia for production workloads. Bedrock pricing reflects these economics.

Meta (clearest Nvidia displacement signal): MTIA Gen 2 in broad inference production tells you that a world-class AI inference operation has concluded that custom silicon beats Nvidia on economics at scale. This is the proof of concept for every other hyperscaler's custom chip program.

Microsoft (important warning about supply): Azure's GPU allocation to Copilot/M365 over pure cloud capacity tells developers that Microsoft will prioritize its own product surface when supply is tight. If your production workloads depend on Azure AI endpoint availability, that prioritization policy is a risk to model in your architecture.

Key Takeaways

• Meta $55.5B +31%: strongest growth since 2021; MTIA Gen 2 in broad production; $115-135B capex maintained; AI ad engine generating real returns
• Alphabet $107B +19%: Google Cloud at ~$18.4B (~50% growth); $175-185B 2026 capex; TPU v6 Trillium driving cost advantage; Gemini 3.1 Flash-Lite at $0.25/M tokens
• Amazon $177B, AWS $36.8B: AI chip business (Trainium + Inferentia) hit $20B annualized, triple-digit growth; $200B 2026 capex; Bedrock running on Amazon silicon
• Microsoft Azure 35% — below 37-38% guidance: CFO confirmed GPU capacity was diverted to Copilot/M365; without that diversion, growth would have been above 40%; supply miss, not demand miss
• Combined $630B capex: capacity coming online 2027-2028; structural AI compute cost crunch eases in two years
• Developer action: evaluate Google Cloud TPU for cost-sensitive workloads; model Azure AI endpoint availability risk; Bedrock users note Trainium 2 is now the underlying hardware for Claude and Llama inference

For the chip supply constraints behind this compute build-out, read SK Hynix Q1 2026: 71.8% Margin, HBM Orders Eclipse 3-Year Supply. For Amazon's foundational Anthropic investment, read Amazon $25 Billion Anthropic Investment. For the open-source alternative to managed inference at these prices, read DeepSeek V4 Pro: 1.6T Parameters, Beats Claude on Coding.