Everyone’s using it. The value hasn’t caught up.
Welcome to The State of Applied AI — a monthly read on where the technology has actually landed in real work, as opposed to where the press releases say it has.
Start with the gap that defines the month. 81 per cent of US physicians now use AI in their work; 56 per cent of chief executives say they have seen no real benefit from it yet. Almost everyone has picked up the tool; far fewer have worked out what it is for. That gap is the story of applied AI right now, and it is widening.
The money has no such doubts. Corporate AI investment reached $581.7bn last year, more than half of all venture capital now goes to AI, and the spending is turning physical — a $500bn US data-centre pledge answered by a €200bn European one.
The evidence has changed category, too. The strongest result to date is not a demo but a randomised trial of 106,000 women, in which AI-supported screening caught 29 per cent more breast cancers than two radiologists reading alone; behind it, the first AI-designed drug to reach a patient trial. Alongside them, the first documented failures — a tool that quietly deskilled the doctors using it, and a medical scribe that invented things patients never said.
Below: first what changed this month, then the standing picture, industry by industry. Open what matters to you.
The 10 findings that entered or changed in the evidence base since the last edition.
In a randomised trial of nearly 106,000 women, AI-supported mammography found 29 per cent more cancers and cut radiologists’ reading workload by 44 per cent, with no rise in false positives. The first RCT-grade proof that AI can help screeners rather than replace them.
Rentosertib — its target and its molecule both AI-designed — reached a Phase 2 trial for pulmonary fibrosis. Small, early and China-only, but a genuine first.
Units running AI triage saw the share of patients reaching functional independence rise from 16 to 48 per cent. Observational rather than randomised, but hard to ignore.
Up 130 per cent in a single year. The signal is not just the size but the pace — and how much of it is now flowing into physical infrastructure.
More than half of global VC deal value went to AI in 2025 — a bet on a single sector the market has never placed before.
The OpenAI-led consortium’s pledge, with roughly $400bn and 7 gigawatts committed by September. Read it as an announced pledge, not cash deployed — but the direction is unmistakable.
The European Commission moved to mobilise €200bn, including a €20bn fund for AI ‘gigafactories’ — a direct response to US infrastructure spending.
AI companies took 54 per cent of US digital-health funding in 2025, up from 37 per cent a year earlier. Fewer rounds, bigger cheques.
After endoscopists routinely used an AI aid, their own unassisted cancer-detection rate fell from 28 to 22 per cent. The first documented clinical-AI downside — and a warning for every profession automating its judgement.
A transcription tool built on OpenAI’s Whisper, used across an estimated seven million visits, hallucinated text no one spoke — nearly 40 per cent of it judged harmful. Convenience without a human check, in miniature.
Industrial Research Unit, Straits Institute for Applied AI. (2026). The State of Applied AI — August 2026 Dispatch. https://www.straitsai.institute/state-of-applied-ai/august-2026
