Five earnings reports, one AI trade hiding in plain sight

The Five Earnings Calls That Rewrote the Rules of Big Tech

Last week was the closest thing markets get to a State of the Union for the American tech economy. Tesla, NVIDIA, Alphabet, Amazon, and Microsoft all reported within days of each other. Taken one at a time, each was impressive. Taken together, they tell a single story, and it's not the one most headlines ran with.

The story is this: the AI infrastructure buildout is no longer speculative. It is the operating reality of five of the most powerful companies on earth, and all five of them are accelerating, not plateauing.

Start with NVIDIA, because everything else orbits it. Jensen Huang reported $68 billion in quarterly revenue (up 73% year-over-year) and guided to $78 billion next quarter. That implies another $10 billion in sequential data center growth. But the number that matters most isn't revenue. It's the Groq acquisition and the $10 billion Anthropic investment, which together mean Jensen is methodically locking the three most important frontier AI labs (Anthropic, Meta, and OpenAI) into NVIDIA silicon at the architectural level. That is not a vendor relationship. That is a chokehold, and it's being assembled in plain sight.

Then Microsoft. Azure grew 40%. AI annual recurring revenue hit $37 billion, growing 123% year-over-year. But Amy Hood said something in that call that deserves more attention than it got: the company is shifting from pure per-seat pricing to seat-plus-consumption across GitHub, Dynamics, and eventually M365. When you have a $600 billion-plus RPO book and you're repricing it upward via usage meters, the revenue ceiling just disappeared. The renegotiated OpenAI deal (royalty-free model IP through 2032, OpenAI now a paying Azure customer) is the most elegant restructuring of an expensive partnership anyone has pulled off in years.

Amazon's AWS printed 28% growth on a $150 billion annualized run rate. That is re-acceleration, not continuation. And Andy Jassy almost casually dropped a $50 billion standalone valuation for Trainium, Amazon's custom AI chip program. Almost nobody is pricing that into AMZN today. The Leo satellite broadband network, being built alongside GlobalStar and Apple, is a trillion-dollar adjacency assembled so deliberately that it barely registers as a story yet.

Alphabet's CFO Anat Ashkenazi said 2027 capital expenditure will "significantly increase" above an already-raised $180-190 billion range, and attributed it to "tangible demand signals." Google Cloud's AI solutions revenue is up 800% year-over-year. The backlog nearly doubled to $462 billion in a single quarter. And the TPU external sales program means Alphabet just entered the AI hardware business at enterprise scale.

Then Tesla, which is the complicated one. Elon Musk ran a call about a decade-long AI and robotics transformation while his CFO flagged negative free cash flow for the rest of 2026 and real tariff exposure in energy without much fanfare. Both things can be true. The 1.3 million paid FSD subscribers with falling churn and rising miles-per-customer is genuinely significant. Tesla reframing itself as a software company with cars as the distribution mechanism is not just a talking point; it shows up in the numbers. But the gap between Musk's timeline optimism and Taneja's near-term caution is wide enough to drive a Cybertruck through.

The thread connecting all five? Capital commitment. These companies are not hedging. They are spending at a scale that only makes sense if the returns are already visible to the people running the models. That's the signal the market keeps underweighting.

✦

Quick Takes

Three more stories worth knowing about

Google's TPU Move Is the Sleeper Story of Earnings Season

Alphabet opening up its TPU chips to external enterprise customers is reshaping the AI hardware competitive map faster than most appreciate. AWS has Trainium, Microsoft has Azure Maia, and now Google is in the market with silicon that has been battle-tested at the largest inference scale on earth. NVIDIA's dominance isn't threatened yet, but three hyperscalers building credible alternatives and now selling them externally changes the negotiating dynamic. Watch Cloud backlog growth for evidence that TPU access is becoming a deal-closer.

Microsoft's OpenAI Deal Is Better Than Anyone Reported

The headline was that Microsoft "renegotiated" its OpenAI relationship, which sounded like a retreat. It was almost the opposite. Microsoft secured royalty-free access to OpenAI's model IP through 2032 while OpenAI commits to Azure as its primary cloud provider. Microsoft turned its most expensive partnership into a long-duration cost advantage while adding OpenAI as a paying customer. The Street focused on the reduced equity stake. The CFO focused on the margins. Hood was right.

Amazon's Satellite Play Deserves a Separate Valuation

Andy Jassy compared the Leo satellite broadband network (built alongside GlobalStar and Apple) to early AWS in terms of capital intensity and eventual free cash flow profile. AWS today generates roughly $100 billion in annual revenue. If Jassy's analogy holds even loosely, and if the Apple partnership means Leo hardware ships embedded in the world's most popular devices, the addressable market is enormous. No analyst model currently assigns meaningful value to this. That gap tends to close eventually.

✦

Overlooked Stock

SKYT

SkyWater Technology, Inc.

Every conversation about AI chips eventually arrives at NVIDIA, TSMC, and a handful of hyperscalers. Almost nobody mentions SkyWater Technology, a pure-play US-based semiconductor foundry headquartered in Bloomington, Minnesota, and that gap is interesting.

SkyWater is not trying to compete with TSMC on leading-edge logic. It operates in a different part of the value chain: defense, aerospace, bio-health, and advanced computation applications where US-sourced, security-cleared fabrication is either required or strongly preferred. The CHIPS Act reshaping is tailored almost perfectly to SkyWater's positioning.

The risk is genuine. This is a small, capital-intensive business with lumpy government contract timing and execution risk on technology development programs. It is not a smooth compounder. But as Washington gets serious about onshoring sensitive semiconductor production, and as defense AI applications scale, SkyWater is one of very few companies that can actually fulfill that mandate. The market has not priced that scarcity in.

The Contrarian Take

The AI CapEx Supercycle Could Peak Before Most Investors Realize It Started

Five major tech companies just committed, collectively, to something approaching half a trillion dollars in capital expenditure over the next two years. The consensus reads this as confirmation that AI demand is limitless. The contrarian read is different: when the buyers of a single supplier's product are also the largest investors in the world, the risk of simultaneous overcapacity is real.

Historically, infrastructure booms funded by the same cohort of players tend to overshoot. Fiber in 2000 is the cautionary case. Cloud capacity in 2022 created a genuine digestion period that showed up in AWS growth rates.

This cycle has more demand-side evidence than any previous one, and it is substantive: enterprise adoption is real, AI revenue is measurable, and the use cases are multiplying. But "measurable" and "sufficient to justify the build" are different thresholds. The scenario where 2027 sees a CapEx pause is not the base case. It is, however, more probable than current valuations imply.