Big Tech Q1 2026 Earnings: $630B AI Capex and an Azure Supply Crunch

Microsoft, Meta, Alphabet, and Amazon report Q1 2026 earnings on April 29. Apple follows on April 30. These five companies have collectively committed approximately $630 billion in AI infrastructure spending for 2026 — a number larger than the GDP of most countries. The earnings calls land not primarily as financial events but as supply chain signals: whether AI compute demand is growing faster than the capacity being built, and what that means for developers trying to access frontier models through APIs.

The top line numbers will be covered everywhere. What matters for developers and infrastructure builders are the signals buried in management commentary — specifically, whether capacity is keeping pace with demand, whether pricing is holding, and whether any company is pulling back or accelerating its hardware commitments.

The Numbers Going Into April 29

Microsoft (Q3 FY26, reporting April 29):

Azure guided 37–38% revenue growth for Q3. That would be an acceleration from Q2's 35% growth. The number to watch is whether Azure exceeded guidance — in Q2 FY26, Azure missed guidance slightly and caused a $357 billion market cap wipeout in a single session. Management has since acknowledged that Azure demand is running ahead of supply, meaning they could sell more if they had more data center capacity. The bottleneck is not customers — it is GPUs and data center buildout timelines.

Meta (Q1 2026, reporting April 29):

Meta guided $53.5–56.5 billion revenue. The more interesting number is the capex update: Meta raised its full-year 2026 capex guidance to $115–135 billion, the largest single-year infrastructure commitment in the company's history. Meta is building AI infrastructure specifically to reduce Nvidia dependency — its MTIA (Meta Training and Inference Accelerator) Gen 2 chips are now deployed in production across recommendation model inference. The earnings call will reveal how much of that $135 billion is allocated to MTIA versus third-party GPU procurement.

Alphabet (Q1 2026, reporting April 29):

Consensus expects $92.2 billion revenue with Search growing 17–18% year-on-year. Google Cloud is the key AI signal — Cloud revenue growth rate and TPU utilization commentary will indicate whether Google's self-sufficient chip strategy (TPU v6 Trillium, in production since late 2025) is creating a cost advantage over hyperscalers that are still primarily Nvidia-dependent. Gemini 2.5 and Gemini 3.1 Ultra are both in production; the call will reveal whether API demand is growing faster than TPU capacity.

Amazon (Q1 2026, reporting April 29):

AWS consensus is $36.8 billion, representing 25.6% year-on-year growth. Amazon's AI services revenue run rate already exceeded $15 billion annualized by Q4 2025. The signal here: Amazon has been more aggressive than any hyperscaler in betting on its own silicon (Trainium 2 for training, Inferentia 3 for inference). If AWS AI revenue is accelerating, Trainium adoption commentary will tell you whether developers are shifting workloads off Nvidia H100/H200 infrastructure onto Amazon-native chip stacks.

Apple (Q1 2026 — fiscal Q2, reporting April 30):

Consensus expects $88 billion revenue. Apple's AI story is different from the other four — it runs inference on-device (Apple Silicon M4 Ultra, A18 Pro) rather than in the cloud, which insulates it from GPU supply constraints. The signal for developers is Apple Intelligence adoption metrics and whether the M4 Ultra's neural engine performance is driving Mac Pro developer adoption.

Azure Demand Ahead of Supply: The Constraint Developers Feel

The most actionable signal in this earnings cycle is Microsoft's acknowledgment that Azure is supply-constrained. Management has said explicitly: there is more demand for Azure AI capacity than there is capacity to serve it.

What this means in practice for developers:

API rate limits are hardware-limited, not policy-limited. GPT-5.5 and GPT-5 access through the Azure OpenAI Service is constrained by actual GPU availability in Azure data centers, not by arbitrary throttling decisions. The queue for Provisioned Throughput Units (PTUs) for enterprise customers has extended to 6–9 months in some regions.

Azure regions are not equal. AI compute is concentrated in East US, West Europe, and Australia East. Developers hitting rate limits in secondary regions cannot always solve the problem by switching regions — the capacity simply is not there.

The Q3 guidance beat or miss matters for 2027 planning. If Azure beats 37–38% guidance, Microsoft will accelerate data center construction. If it misses, construction timelines stretch. The Q3 call result directly affects when the supply constraint eases.

Meta's $135B Bet: What MTIA Actually Is

Meta's $135 billion capex figure is so large it requires unpacking. The majority is not going to Nvidia GPU clusters — it is going to Meta's own MTIA (Meta Training and Inference Accelerator) infrastructure, custom optical networking, and the data center footprint to house it.

MTIA Gen 2 handles recommendation model inference at scale. Every time someone opens Instagram Reels, Facebook Feed, or gets a Threads algorithmic push, MTIA chips are doing the computation. Meta processes more than 5 trillion inference operations per day across its platforms.

Why this matters for the broader market: if Meta's $135 billion is going to proprietary chips, Nvidia does not see most of that revenue. Meta is simultaneously the world's largest AI inference operation and the company most aggressively building hardware to avoid paying Nvidia's margins. The earnings call commentary on MTIA Gen 2 yield and deployment scale will tell you how far along this transition is.

Amazon AWS: Trainium 2 and the $15B Run Rate

Amazon's AI services revenue at $15 billion annualized is a relatively recent acceleration — it was less than $5 billion annualized entering 2025. The growth rate is compounding, and the question for Q1 2026 is whether the rate continues to accelerate or normalizes.

Amazon's bet is that Trainium 2 (custom silicon optimized for transformer model training) can eventually offer a cost advantage over H100/H200 clusters for specific workloads. Anthropic's Claude models are trained on Trainium 2 infrastructure as part of the Amazon-Anthropic $25 billion partnership. If Anthropic's Trainium utilization is growing, that is the proof-of-concept for Trainium's competitive case.

Inferentia 3 (inference-focused) is deployed across AWS SageMaker and Bedrock. If Bedrock AI services revenue is growing faster than AWS overall, it signals that developers are choosing managed inference over raw GPU instances — a preference for simplicity over control that changes how AI infrastructure gets sold.

Alphabet: TPU Self-Sufficiency and the Search Defense

Google's earnings have a dual character in 2026. On the offense: Google Cloud and TPU v6 (Trillium) represent the strongest vertically integrated AI stack outside of Apple. Google designs its own training chips, builds its own data centers, runs its own models, and sells inference through its own cloud. No other hyperscaler has this full stack.

On the defense: AI Overviews in Google Search is cannibalizing standard search result clicks. If the Q1 call reveals Search revenue growing at only 17–18% while AI API revenue grows at 40%+, it confirms that Google is succeeding in the transition but at the cost of its highest-margin legacy product.

For developers building on Google Cloud, the signal is whether Gemini 3.1 Flash-Lite's $0.25/M token pricing is attracting high-volume API workloads away from cheaper but less capable open-source alternatives.

What Developers Should Watch in Each Call

Microsoft (April 29): Listen for "demand versus supply" framing in Azure commentary. Any mention of PTU waitlists, region-specific constraints, or data center acceleration timelines is signal for when API access improves. Watch OpenAI segment revenue — Microsoft consolidates some OpenAI revenue through its Azure partnership.

Meta (April 29): Watch capex guidance update and MTIA deployment metrics. If Meta says MTIA Gen 2 handles X% of total inference operations, that quantifies how far along the Nvidia replacement is.

Alphabet (April 29): Watch Google Cloud growth rate acceleration or deceleration. TPU v6 capacity commentary. Any mention of per-query cost reduction for Gemini inference = pricing power signal.

Amazon (April 29): Watch Bedrock and SageMaker growth. Trainium 2 adoption metrics from Anthropic workloads. Any mention of Inferentia 3 deployment scale.

Apple (April 30): Watch Apple Intelligence active device metrics. M4 Ultra Mac Pro sales data. On-device inference usage = developer platform signal.

The Supply Constraint Is the Real Story

The collective $630 billion capex commitment for 2026 is often reported as a sign of AI boom. It is also a sign of how badly supply has lagged demand. Every hyperscaler is building faster than ever because they were caught under-building in 2024–2025. The Azure supply crunch is not an anomaly — it is the clearest symptom of an entire industry that spent two years under-estimating how fast demand would grow.

For developers, the practical implication is simple: if you need reliable, low-latency access to GPT-5.5-class or Claude Opus 4.7-class models at volume, the capacity constraint is real and the queue is long. Building on open-source alternatives running on your own infrastructure is not always a philosophical choice — sometimes it is the only way to get consistent throughput today.

Key Takeaways

• $630B combined AI capex from 5 companies in 2026: Microsoft ($80B+), Meta ($135B), Amazon ($105B), Alphabet ($85B), Apple ($65B) — infrastructure investment at unprecedented scale
• Azure is supply-constrained, not demand-constrained: management confirmed demand exceeds capacity; PTU waitlists 6–9 months in some regions; API rate limits are hardware limits
• Meta's $135B mostly goes to MTIA, not Nvidia: Meta is the most advanced Nvidia-replacement program in the industry; earnings commentary will quantify MTIA deployment scale
• AWS AI revenue $15B run rate: growing from $5B annualized entering 2025; Trainium 2 and Inferentia 3 are the product bets
• Google's vertically integrated stack: TPU v6 Trillium + Gemini + GCP is the only full-stack AI platform outside Apple — watch Cloud growth rate for competitive signal
• Post today, not tomorrow: earnings commentary will be analyzed everywhere on April 29 — publishing the developer-angle preview before results drop maximizes indexing and traffic window

For the Anthropic investment context behind Amazon's AI strategy, read Amazon $25 Billion Anthropic Investment — Trainium, AWS $100B Deal. For the chip supply constraints behind these compute buildouts, read SK Hynix Q1 2026: 71.8% Margin, HBM Orders Eclipse 3-Year Supply. For the open-source alternative to managed inference, read DeepSeek V4 Pro: 1.6T Parameters, Beats Claude on Coding.