All-Powerful AI is Here. We Just Need to Unlock It.
Summary
Josh and Ejaaz discuss the release of Claude Mythos, powered by NVIDIA's Blackwell chip, and the hardware roadmap leading to AGI. They highlight that future chips like Vera Rubin and Feynman will deliver massive compute multiples, while older GPUs remain valuable for inference, driving demand for NVIDIA and neocloud companies.
- Claude Mythos, a powerful AI model, was trained on NVIDIA's Blackwell chip from 2024.
- NVIDIA's chip roadmap includes Blackwell, Vera Rubin, Rubin Ultra, and Feynman, each offering large compute gains.
- Hardware improvements alone could lead to AGI within a few years without new breakthroughs.
- Older GPUs like the H100 hold their value due to high demand for inference compute.
- Neoclouds such as CoreWeave are thriving by renting out older GPUs for inference tasks.
- Software efficiency gains compound hardware improvements, increasing the value of existing GPUs.
- Demand for compute outstrips supply, leading to sustained pricing power for GPU owners.
- Michael Burry's short thesis on GPU depreciation has been invalidated by market dynamics.
Trade Ideas
NVIDIA hardware ensures AGI and demand growth
NVIDIA's chip roadmap (Blackwell, Vera Rubin, Rubin Ultra, Feynman) provides massive compute multiples without requiring new breakthroughs, making AGI inevitable and driving sustained demand for NVIDIA's hardware. The trajectory of hardware improvements combined with software efficiency gains means NVIDIA is positioned to dominate the AI compute market for years.
