Reproducing GPT-2 now costs $48, putting superhuman AI under $500
jennyzhangzt · x · 2026-08-26
- Cost Comparison: Training GPT-2 in 2019 required 22 PF-days of compute and $43k. Today, nanochat achieves GPT-2 level capability for just $48 on 8×H100s.
- Projection: AlphaZero needed 100–150 PF-days to reach superhuman Go in 2017. Applying the same cost curve implies a cost of $220–330 today.
- Conclusion: The cost barrier to achieve superhuman AI capabilities may have dropped below $500.
More from Infra
- AI Capex Boom May Trigger Sovereign Debt Crisis in Late 2020s: SemiAnalysis — Traditional-Chip8339 · 2026-08-26
- Tune AMD V620 power/clock on Windows without VBIOS flashing — Brave_Load7620 · 2026-08-26
- Akta launches Company Data API for AI Agents with entity resolution — AppropriateAnt7344 · 2026-08-26
- Alibaba's RecGPT-Mobile-V2: On-Device Behavior Prediction with RL — _reachsumit · 2026-08-26
- AMD MI350X Runs Qwen3.6-35B: Open Source Kernel Achieves 78.5k tok/s on 8 GPUs — SmilingGen · 2026-08-26
- MetricFire releases MCP server to query monitoring data with AI tools — PKMNPinBoard · 2026-08-26