- 59%
- Accuracy
- 14 of 41
- Analyst ranking
- 5
- Predictions scored
- 5
- Still to fall due
All categories ▾
Half right · 50%We think the three large AWS campuses will meaningfully contribute to AWS' top line by the end of 2025 and jack up growth above the 20% YoY threshold.
What happened. AWS segment sales were $35.6bn in Q4 2025, up 24% year on year (Q3 2025: $33.0bn, up 20%). 'Above the 20% threshold' is scored as a point forecast at 20% under the threshold ruling. Q4 2025 growth was 24%, 4 points out, inside the 5-point band. The direction was right; the overshoot costs points. Sources: ir.aboutamazon.com.
Source: Amazon's AI Resurgence: AWS & Anthropic's Multi-Gigawatt Trainium Expansion (SemiAnalysis)
Right · 100%A hundred something billion dollars of revenue for NVIDIA last year, 200 something billion this year, and if next year, it could nearly double again.
What happened. Nvidia reported fiscal 2026 revenue (February 2025 to January 2026, the nearest match to 'this year') of $215.9bn, up 65%, in its first full-year release on 25 February 2026. '200 something billion' is a range ($200-299bn). The actual $215.9bn is inside it, so Right; forecast_value is set to the actual, following the range convention. Sources: nvidianews.nvidia.com.
Wrong · 0%There's no way you can pay for the scale of clusters being planned to be built next year for OpenAI unless they raise like $50-100 billion, which I think they will raise late this year or early next year.
What happened. OpenAI announced a $40bn funding round at a $300bn valuation on 31 March 2025, led by SoftBank. Its previous round (October 2024) was $6.6bn. He gave a $50-100bn range; the raise was $40bn, outside it. Scored from the nearest edge ($50bn): 20% short, inside the 25% band. Timing (early 2025) was right. Sources: cnbc.com, techcrunch.com.
Source: Dylan Patel & Jon Y (Asianometry) – How the semiconductor industry actually works (Dwarkesh Podcast, transcript) · 'early next year' read generously as the first half of 2025
Right · 100%China will have multiple firms capable of training better than GPT-4 next year alone. It's not even a question.
What happened. In 2024 several Chinese labs released models that beat the original GPT-4 on standard benchmarks, including DeepSeek (DeepSeek-V3, December 2024) and Alibaba (Qwen2.5). Stanford's 2025 AI Index says the gap between Chinese and US models shrank from double digits in 2023 to near parity in 2024. More than one Chinese firm trained a model better than GPT-4 within 2024, as he said. Sources: hai.stanford.edu, github.com, qwenlm.github.io.
Source: China AI & Semiconductors Rise: US Sanctions Have Failed (SemiAnalysis)
Mostly right · 75%Going forward, we expect AMD's share of x86 server units to increase from 13.9% in Q2 2022 to 21.2% in Q4 2023.
What happened. Mercury Research put AMD's x86 server CPU unit share at 23.1% in Q4 2023 (published February 2024). 21.2% against an actual 23.1% is 1.9 points (8.2%) under: outside 5%, inside the 10% band. He called the share gain but under-shot it. Sources: wccftech.com, profesionalreview.com.
Source: 2023 Datacenter Outlook – AMD and Intel Revenue, ASP, and Units – Genoa Ramp Details (SemiAnalysis)
Still to fall due5 predictions
- Dec 2026
At the current trajectory, we believe that Claude Code will be 20%+ of all daily commits by the end of 2026.
Artificial Intelligence · made Feb 2026 · source - Dec 2026
But by next year, you'll have 500,000 to 700,000 GPU clusters.
Infrastructure & Telecoms · made Feb 2025 · source - Dec 2026
One gigawatt, one site in 2026. But then you have a number of others. You have five different locations, some with multiple sites, some with single sites.
Infrastructure & Telecoms · made Oct 2024 · source - Dec 2030
We believe AI will propel datacenters to use 4.5% of global energy generation by 2030.
Infrastructure & Telecoms · made Mar 2024 · source - Dec 2026
Global Datacenter Critical IT power demand will surge from 49 Gigawatts (GW) in 2023 to 96 GW by 2026, of which AI will consume ~40 GW.
Infrastructure & Telecoms · made Mar 2024 · source