PlainAI AI explained plainly

Stanford HAI · 2025 · 04 · 07 · Impact · ~2 min read

Stanford's 2025 AI Index Report — the year in numbers

Stanford's annual AI Index — the most thorough numbers on the AI year — landed with hard data. AI is now in 78% of organisations (up from 55%), inference costs fell 280x in 18 months, FDA-approved AI medical devices jumped from 221 to 950, and AI safety incidents rose 56%. The report nobody loves but everyone cites.

What's actually new

  • FDA-approved AI medical devices: 221 → 950 in roughly 18 months. AI in healthcare is no longer a slide-deck slogan.
  • AI inference cost dropped 280x in 18 months. GPT-3.5-quality output went from a luxury to about $0.07 per million tokens.
  • 78% of organisations now use AI in some form, up from 55% the year before.
  • AI safety incidents up 56% in a year — 233 reported incidents in 2024 alone.
  • U.S. private investment in AI: $109 billion in 2024. 12x China, 24x the UK.

If you want more

Worth knowing~30s
  • 'Used AI' is a fuzzy measure. It includes companies that turned on Microsoft Copilot once. The real-impact share is much smaller.
  • Investment numbers skew toward OpenAI / Anthropic / xAI rounds. Most companies in the 'AI investment' bucket are infrastructure or applied tools, not foundation labs.
  • The 280x cost drop applies to GPT-3.5-quality. Frontier-model costs (GPT-5, Opus 4.5) haven't dropped at the same pace.
Who should care~20s

Anyone trying to make sense of the AI year. Policymakers needing data to cite. Educators teaching AI literacy. Investors and journalists. Anyone who'd rather read 400 pages of charts than another 'top 10 AI predictions' list.

What to do about it~20s

Bookmark the report. Skim the executive summary every April when the new edition lands. Use it as the empirical anchor when someone in your office says 'AI is taking over' or 'AI is overhyped' — both claims usually fall apart against the actual numbers.

Honest take~45s

Stanford's AI Index does the unglamorous work that nobody else does — measuring AI's actual progress year over year with primary sources. The 2025 edition's most underrated finding wasn't the cost drop or the investment surge, but the FDA-approved-medical-devices number going from 221 to 950. That's the AI story that doesn't trend on Twitter — quiet regulatory approvals turning into routine clinical use. Most of AI's actual impact looks like that. The screaming-headline AI is a small fraction.

Other recent impact updates

Sources

Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com