The Mag 7 Prints Are In. The 2027 Inference Bet Just Got Stronger.
The Mag 7 Prints Are In. The 2027 Inference Bet Just Got Stronger.
Yesterday I wrote a framework piece on the four numbers I was watching from the Mag 7 prints. The numbers landed last night. Alphabet was up 6% after-hours on a capex raise to $180–190B and a "significantly higher" 2027 capex tease. Meta beat earnings too but fell 7% after raising 2026 capex to $125–145B and citing "higher component pricing." Microsoft posted Azure +40% YoY and Q1 capex of $29.88B (up 89% YoY). Amazon reported but the AWS-margin story is still being parsed.
The pre-mortem framework had four scenarios. One specific scenario actually played out, and it has one specific implication for a sub-$10K MRR shop's 2027 inference bill. This post is the recap, the scenario that won, and the three-sentence honest takeaway before the takes drown out the signal.
The scenario that won
The bet I floated yesterday was: prints roughly in line, capex commentary defended-but-not-raised, stay month-to-month and bet on 2027 price compression. What actually happened was materially better on revenue at three of the four hyperscalers, capex raised (not just defended), and the market reaction was bifurcated based on how each company framed the raise.
Alphabet framed its raise as demand-driven. Cloud grew 63%, operating margin nearly doubled from 17.8% to 32.9%, and the capex commentary positioned 2027 increases as funded by Cloud revenue going on-line today. The market rewarded that with a 6% pop.
Meta framed its raise as input-cost-driven. The phrase "higher component pricing" did the heavy lifting. The same capex magnitude with the same AI-buildout story got punished 7% because the framing implied the spend wasn't yet matched by AI-revenue compounding. Same number, opposite reaction.
The signal here is that the market is now grading capex raises on revenue-coverage, not magnitude. Alphabet won that grading; Meta lost it. Microsoft was somewhere in between because Azure +40% gave them air cover, but the +89% capex YoY would have been punished without it.
What this actually means for your inference bill
The 2027 price-compression bet from yesterday strengthened, not weakened, after these prints. The mechanical reason is concrete enough to write down.
Alphabet raising capex to $190B with an explicit 2027 step-up means a lot of new capacity comes on-line in 2027. New capacity needs customers. New customers means competitive pricing — Cloud at 32.9% operating margin is now profitable enough that Google can afford to keep building, and the marginal capacity has to find revenue somewhere. Locking in a multi-year inference contract today, before that capacity exists, is the wrong move.
Meta's capex raise is a supply-side signal that feels bearish but is actually neutral for indie inference. Meta does not sell hosted inference to indie devs at any meaningful scale. The capex they're raising goes into Meta's own products — Llama serving, ads ranking, Ray-Ban glasses — none of which competes for your $50/month Anthropic spend. The component-pricing complaint is a Meta problem, not an indie-stack problem.
Microsoft's Azure +40% looks scary at the headline, but $10B of new annual cloud revenue in one quarter is dominated by enterprise migrations and enterprise OpenAI workloads. The competitive intensity for sub-$10K MRR cloud spend is therefore not actually rising as fast as the headline suggests. Microsoft does not need to price Azure aggressively for indie shops because indie shops are not the customer Microsoft is fighting for.
Net read: hosted-inference unit pricing keeps falling through 2027. Stay month-to-month. Don't pre-commit to multi-year inference contracts at 2026 prices.
The Cloud margin number nobody else will frame this way
Google Cloud went from 17.8% operating margin to 32.9% in one year. That's not a normal margin expansion. The mechanical interpretation is that Cloud finally crossed into "this business funds itself" territory after seven years of "we're in this for strategic reasons."
The implication for solo operators using GCP is awkward: the "GCP will get cheaper than AWS to win share" narrative that some indie devs have been waiting on for three years is now less likely going forward, not more. Google now has a real economic reason to hold pricing, not slash it. The strategic-loss-leader era is over.
If you've been holding off migrating off GCP because you thought a price cut was coming, stop waiting. The price cut is not the trade Alphabet is making with this margin. They're holding pricing and letting the volume compound.
The Meta read for solo operators specifically
The 7% punishment was about Meta raising capex while missing user growth. The implication is that Meta's AI infrastructure spend is not yet translating to user-facing product wins that compound. The market wants to see the revenue connection.
For an indie dev shipping AI features, the relevant lesson is the same at every scale: capex without a clear revenue compounding story gets punished. Your equivalent is "spending on better models without a corresponding plan for how the better outputs improve customer retention or expansion." If your roadmap has "upgrade to Opus 4.7 for our high-value workflows" without a corresponding "and here's how we monetize the better outputs," you're in the same trap Meta is in. Make the revenue connection explicit before you ship the upgrade. Write it in a sentence above the roadmap card.
The three-sentence indie read
Here's everything a sub-$10K MRR shop needs from these prints.
Stay month-to-month on inference. Don't migrate to Azure for indie pricing — that price cut is not coming. Tie any model upgrade to a written revenue-compounding story before you ship it.
That's it. The other 80% of Mag 7 earnings coverage on April 30 is for retail investors, options traders, and thinkpiece writers. The signal-to-noise ratio for solo operators on this kind of cycle is brutal, and the right response is to read the prints, update your priors once, and go back to whatever you were building.
I read four of the post-earnings analyses this morning before writing this. The best one was a Bloomberg piece on capex-revenue coverage as the new market grading rubric. The worst was a vibes essay about whether AI is "still real." Neither one changed what I do tomorrow. The framework piece I wrote yesterday already had the action items, and the actual prints sharpened them rather than overturning them. That's usually how good frameworks work.
The honest counter-take I keep coming back to: most indie operators reading Mag 7 earnings coverage are doing it because the news feels important, not because the news is actionable. If the read above doesn't change anything you're doing this week, that's the correct response, not a failure of the analysis. The prints reinforce the existing strategy. Keep building.