[Daily Briefing] 💸 $1.5T committed to AI, 🔒 421 patches, ⚛️ quantum meets fusion
Aggregate AI purchase commitments across the largest technology companies are now approaching 1.5 trillion dollars. Not spent. Committed: chips, data centre capacity, power contracts, multi-year cloud deals signed years ahead of the demand they assume.
At the same time, SMIC is putting prices up because its fabs are running near capacity, and essentially every major foundry on Earth is booked. The physical buildout is real, it is happening, and it is consuming a genuinely startling share of the world's industrial output.
Set that against the number that keeps turning up in surveys: roughly 80 percent of companies deploying AI report no meaningful profit impact yet. Both things are true at once. Enormous, concrete infrastructure spending on one side, and a return that is still largely promissory on the other.
Nvidia reports Q2 FY2027 earnings on 26 August, which is the next real stress test of whether the demand behind those commitments is showing up on anyone's income statement.
Why it matters: This decides whether the next two years look like 1996 or 2000. The bear case is simple, that commitments this size assume revenue nobody has booked. The bull case is that infrastructure always arrives before the application layer, and the fibre glut of 2001 is precisely what made YouTube possible five years later. Worth watching without picking a side too early.
Quick hits
- Microsoft patched 421 flaws in one day. The largest single patch batch on record, with one zero-day already exploited and a DNS Server bug analysts have flagged as potentially wormable. Read more
- IBM just put OpenAI inside its consulting army. GPT-5.6 and Codex get folded into IBM Consulting Advantage, and OpenAI effectively rents tens of thousands of consultants who already sit inside banks and government departments. Read more
- Grok 4.6 matches GPT-5.6 on the intelligence index. Four labs now sit within noise of each other, so the real competition has moved to price, latency, and which model survives a four hour agent run without drifting. Read more
- One phone call reached 22 million patient records. No exploit, no malware. A vishing call talked its way past a help desk, and voice cloning has quietly removed the last verification signal most of us relied on. Read more
- A quantum computer worked out how a fusion plant refuels itself. IBM, Oak Ridge and the Cleveland Clinic calculated how tightly tritium binds to molten salt, which is the difference between a reactor that refuels and one that starves. Read more
Tool of the day
The Star Builders, by Arthur Turrell
If the fusion story above left you wanting the actual physics rather than the press release version, this is the book. Turrell is a plasma physicist and he writes like someone who has stood next to the machines, which means the chapters on tritium breeding and confinement land properly instead of hand-waving. Honest caveat: it was written before the recent ignition results, so the final chapters read as more speculative now than they did on publication. The explanations have aged fine. The timelines have not.
Quick question for you: of the 1.5 trillion committed to AI, which single piece do you think is most likely to be quietly written off in two years, the data centres, the chips, or the power contracts? Hit reply, I read every one.
Nath, Future Technology
Forwarded? Get your own free briefing at futuretechnologyhq.com/newsletter
Some links may be affiliate links.