NVDABullunverifiable
Randy Kirk“I don't think like suddenly demand's going to go away for those things.”
Thesis at the time
BullishThe speakers argue that AI inference compute demand—now over 90% of total AI compute usage—is growing so fast that even a slowdown in training compute would not meaningfully reduce demand for Nvidia chips or related data center and memory investments. They point to rising usage of agentic AI tools like Grok and OpenAI's computer-use models as evidence that inference demand keeps accelerating.
Key arguments
- Inference compute is over 90% of AI compute demand, while training is less than 10%.
- Tools like Grokbot and OpenAI's computer-use models make agentic AI easier to use, increasing demand.
- Anthropic (Dario/Anthropic) got 10x more compute last year but still needed more, showing inference demand growth.
- Even if training slows down due to regulatory or 'doomer' pressure, investment in data centers, memory companies, and Nvidia should remain strong.
Counter-arguments acknowledged
- Risk exists that training could slow down due to regulatory or safety concerns ('doomers').
Hedges and caveats (from the video)
- Acknowledges risk that training could slow down due to regulatory concerns
- Speculative discussion about hypothetical 'doomer' scenarios
- No specific price targets or directional calls on assets
The call
- Date said
- Sep 20, 2026
- Price at prediction
- $222.27
- Confidence
- low
- Specificity
- vague
How it resolved
- Status
- unverifiable
Why this resolved this way(resolution audit)
Full rules: docs/resolution-spec.md.
Confidence Reasoning
The claim is framed within a hypothetical ('let's say the doomers win and they slow us down') and hedged with 'I don't think,' reducing conviction despite a clear bullish direction on continued demand for Nvidia and related infrastructure.