HOW TO LOSE A QUARTER-TRILLION DOLLARS BY SPENDING $3 BILLION

(MSFT), (NVDA)

What kind of market punishes a company for spending money to deliver services that customers have already contracted to buy? Apparently, this one. 

Microsoft’s (MSFT) down 25% since October because its quarterly capex came in at $37.5 billion instead of the expected $34.3 billion. That $3 billion “miss” triggered a quarter-trillion-dollar market cap evaporation. 

I’ve seen more rational behavior at airport security.

Azure grew 39% year-over-year in the most recent quarter, and the growth is entirely capacity-constrained. 

Microsoft can’t build data centers fast enough to meet demand for services already under contract. They’re sitting on a $625 billion backlog of cloud revenue that’s been sold and signed. 

The complaint that they’re spending too much building infrastructure to actually deliver what they’ve already sold is like yelling at a restaurant for buying too many ovens when the reservations are booked solid for the next three years.

The capex breakdown tells you everything about why this panic is misplaced. 

A substantial chunk is going toward Microsoft’s Maia 200 AI accelerator and Cobalt 200 ARM-based CPU. These chips aren’t trying to beat Nvidia (NVDA) in a drag race. 

They’re designed to make AI workloads cheaper to run, delivering roughly 30% better performance per dollar for inference. It’s the difference between buying the flashiest sports car and buying the one that actually gets you to work without bankrupting you on gas.

Data center COGS breaks down to roughly 40% hardware, 25% electricity and cooling, 20% operations, 15% licensing. 

Custom chips attack the first two categories directly – 65% of total costs. A 30% efficiency gain translates to 20% reduction in total COGS. 

Run that through Microsoft’s 66% cloud margin, and you hit 69%. Push custom chips to full deployment, and you’re at 73%, above their historical peak of 72%. 

Microsoft won’t go full custom because that requires firing everyone who knows Nvidia hardware, but this math puts pricing pressure on chipmakers either way. Jensen Huang sees the trap but can’t do much except play along.

The return on incremental capex separates investors who can do math from those who just read headlines. 

Fiscal 2026 to 2027 EBIT growth hits $22.6 billion in incremental earnings against $38.25 billion in incremental capex from 2025 to 2026. That’s a 59% return in year one. Not bad for “reckless overspending.” 

Fiscal 2027 to 2028 shows $30 billion in incremental EBIT against $21.7 billion in capex – 138% returns. The returns accelerate, which is what happens when you’re building for an actual market instead of hoping one materializes.

The OpenAI concentration risk is the one legitimate concern, and it gets beaten to death in every research note. 

OpenAI represents 45% of that $625 billion backlog. If Sam Altman’s grand experiment implodes, Microsoft catches some shrapnel. But the structure here is smarter than people give it credit for. 

Microsoft owns 49% of OpenAI’s for-profit entity, has exclusive cloud hosting rights, and is weaving Copilot into every corner of its enterprise stack. 

If OpenAI goes sideways, Microsoft doesn’t lose the technology or the customer relationships. They just get them at a discount. 

It’s the corporate equivalent of lending your buddy money to start a restaurant, except you also own the building, control the kitchen equipment, and have first right of refusal if things go south.

The stock trades at 25x forward earnings at $413, near the bottom of its 10-year range. Apply that to fiscal 2027 EPS of $18.87, and you get $471. 

Bear case 20x puts you at $377. Revert to 30x – reasonable for Azure growing 39% with margin expansion ahead – and you hit $566. 

The risk-reward skews heavily upside unless Microsoft suddenly forgets how to build software.

The capex panic is quarterly noise masquerading as a structural concern. Microsoft is building infrastructure for a cloud market projected to grow from $1.04 trillion to $2.65 trillion over five years. 

When your returns on incremental capital are running north of 100% and accelerating, you’re not spending too much. You’re spending exactly what the opportunity demands. 

Wall Street will figure this out eventually, probably around the same time TSA starts making sense.