Look, I’ve been following this AI gold rush pretty closely, and right now it feels like we’re watching the most expensive bet in tech history play out in real time.The numbers are honestly insane. The biggest cloud players — think Amazon, Microsoft, Google, Meta — are burning through something like half a trillion dollars in capital expenditures over just two years. We’re talking new data centers popping up faster than anyone can count, insane power deals, custom chips, whole new grids being planned… it’s like the entire industry decided to build the next decade of computing all at once.And the question everyone (including me) keeps asking is:Okay… but how are they actually going to make that money back?Here’s what I’m seeing in early 2026 — the realistic ways companies are starting to turn AI from a gigantic cost center into something that actually prints money.
1. The boring-but-huge winner: Just charge more for cloudThis is still the clearest, most proven path.Every time a company wants to run serious AI (training, fine-tuning, inference at scale), they almost always end up renting GPUs/TPUs from the big three (AWS, Azure, Google Cloud). And because demand is still outstripping supply in many regions, they’re quietly able to charge very healthy premiums.A lot of people underestimate this part because it’s not sexy. There’s no viral consumer app announcement. It’s just… higher cloud bills. But when your biggest customers are increasing their monthly spend by 3×–10× because of AI workloads, that compounds very quickly.I suspect this will remain the #1 way the capex gets paid back for at least the next 3–4 years.
2. Enterprise “platform” deals (the quiet nine-figure contracts)We’re now seeing more and more multi-year, nine-figure+ deals where large enterprises basically outsource their entire AI infrastructure + model access + fine-tuning + safety + compliance stack to one vendor.These deals are messy, take forever to close, and rarely get talked about publicly. But when they do land, they’re massive. Think hundreds of millions per year for one customer.The winners here are the companies that can offer the whole stack end-to-end and convince risk-averse enterprises they’re the safe choice.
. Vertical SaaS + AI agents (finally starting to work)This one took longer than most expected, but we’re now seeing real traction in certain verticals:Legal tech (contract review + redaction + regulatory monitoring)
Healthcare revenue cycle + clinical documentation
Financial services compliance & fraud
Customer support that actually resolves 60–80% of cases autonomously
The pattern seems to be: the companies that win are the ones that already had deep vertical data + workflows + customer relationships before layering AI on top. Pure AI-first startups are mostly still struggling to get past pilot stage at scale.
Consumer subscriptions… still mostly underwhelming (with one big exception)Outside of maybe one very obvious leader, consumer AI subscriptions are not moving the needle much yet for the giants.Most people still use the free tier, or they bounce between three or four different chatbots depending on the task. The “family plan” or “pro” tiers help, but they’re not yet the cash cow many hoped for.That said — when you look at the one consumer product that really cracked the code, you see what’s possible: extremely high retention, viral growth, and people happily paying $20–200/month. It’s just very hard to replicate.So… is the math starting to work?In early 2026 my honest read is:The cloud infrastructure bet still looks very likely to pay off (maybe not spectacularly, but solidly)
Enterprise platform + vertical SaaS is ramping faster than Wall Street expected six months ago
Pure consumer AI monetization remains the weakest link for most players
If you force me to summarize it in one sentence:The capex will probably be justified — but mostly through boring, enterprise, cloud-style revenue rather than through the sexy ChatGPT-for-consumers story that got everyone excited in the first place.And honestly? I’m okay with that. Sometimes the most valuable things in tech end up being the least glamorous ones.What do you think — are we underestimating the enterprise wave, or is the whole thing still way too expensive for whats coming back?
Comments