🟢 The state of AI in 2026
A rapid survey of where AI can deliver ROI looking forward and a discussion of the different levels of trust you can give to LLMs.
In 2022, I wrote about the metaverse's potential use cases and where I saw business scalability. Unlike most analysts these days, who mostly plot future developments from tech companies' enthusiastic press releases, I just considered one simple thing. Where were the problems to be solved by the metaverse? Inevitably concluding there was nothing much and foreseeing how it would fail to ramp up to Meta's expectations, I helped corporate customers (and a few startups) hedge their bets on this one and (mostly) steer away from the whole mess.
Fast forward to now, and Meta reportedly lost ~$70bn on this adventure, rapidly becoming a huge cautionary tale for the rest of the industry about going all in on an unproven technology.
Just kidding.
No one cared that much, and as you all know, we are now embarked even further on a huge bet on a largely unproven technology. So, let's discuss AI.
An unproven technology?
First, what do I call it, an unproven technology, while you might be using ChatGPT, Claude, or any other consumer-grade LLM every day to plan your next trip to Italy or get the bullet points from your last online team meeting?
On the surface, 'unproven' is a bit harsh in itself. I should explain:
For context:
The four main U.S. hyperscalers building and running LLM infrastructure (Microsoft, Amazon, Google/Alphabet, Meta) reported in 2025:
- Microsoft: about $80B for AI-enabled data centers.
- Meta: about $60–$65B in capex (AI infrastructure).
- Alphabet (Google): raised 2025 capex outlook to $91–$93B amid AI demand.
- Amazon: projected $125B capital spending.
