New technology is arriving later than the announcements say. The hold-ups are practical: power connections, chip factories, mines, drug trials, and too few people to fix security flaws.
These are our predictions. The date after each one is when it will be settled.
The predictions
Power shortages will hold back AI data centres to 2028 (End of 2028)
Memory makers will earn higher margins than Nvidia through 2028 (End of 2028)
Anthropic and OpenAI will keep over 70% of US business spending on AI models to 2028 (End of 2028)
The most advanced chips will still be made by TSMC in Taiwan at the end of 2029 (End of 2029)
China will keep export licences on key minerals to 2029 (End of 2029)
Small reactors and fusion will sell under 100 MW of power in the West before 2030 (End of 2029)
Grid storage will use more battery cells than electric vehicles in 2028 (End of 2028)
Clean power keeps growing, but green fuels and carbon removal stop at legal quotas (End of 2029)
The US will buy masses of cheap drones but still spend more on missiles (End of March 2029)
SpaceX will still run most working satellites in 2029 (End of 2029)
AI will find software flaws faster than they get fixed (End of 2028)
Quantum computers will threaten encryption before they beat ordinary computers at paid work (End of 2030)
Humanoid robots will sell over 100,000 a year, but few will do paid factory work (End of 2028)
AI-designed drugs will fail mid-stage trials as often as others (End of 2029)
Risks
Of the risks we re-checked this month, nine are happening now and nine more are getting worse. Two have eased.
What changed this month
We re-checked our twenty open forecasts. We are more sure of eight and less sure of seven. Three are unchanged. We dropped two: we no longer expect 10,000 Western humanoid robots in paid industrial work by the end of 2027, and we no longer expect SpaceX to face real competition in reusable rockets by then.
Some of the twenty also contained factual errors. The correction is shown beside each one.
What subscribers get
Anyone can read the headline of each prediction, what most people expect, and the date it will be settled. Subscribers get our full view and reasoning, what we expect to see first and when, what would change our mind, and what it means for investors, defence, government and businesses.
A prediction that turns out wrong stays on the site, marked as wrong.