Investors should prioritize NVIDIA (NVDA) and AMD as the industry shifts from training models to "inference-time compute," where hardware optimized for real-time reasoning will become the most valuable asset. The emergence of "chain of thought" reasoning suggests a massive, underestimated demand for energy and data center infrastructure to support AI that "thinks" constantly. Look for "AI-first" software companies that integrate reasoning-based models like OpenAI’s o1 to solve complex problems, as traditional SaaS providers face obsolescence from hyper-deflation. Expect a significant market disconnect in 2025, as technological capabilities are projected to grow at an exponential 1,000x pace while most investors only anticipate linear growth. Focus on companies capable of achieving massive scale by leveraging the 40x year-over-year drop in the cost of compute to embed AI into every product.

Artificial Intelligence (AI) Sector

The discussion highlights a fundamental shift in how AI models are being developed and scaled. The focus has moved from "training time compute" (the initial building of the model) to "inference time compute" (the power used when the AI is actually thinking and generating an answer).

• Hyper-Deflation of Costs: There is a projected 1,000x drop in cost for AI capabilities. This is driven by a consistent 40x year-over-year deflation in the cost of compute.
• Chain of Thought Reasoning: This is identified as the "biggest breakthrough ever" in neural network research. It allows models to use more processing power during the "thinking" phase to produce significantly smarter results.
• Market Underestimation: The speakers believe the general public and investors are vastly underestimating the growth for the coming year. While many expect linear growth (e.g., 2x), the technological trajectory suggests another exponential leap.

Takeaways

• Focus on Inference: Investors should look toward companies providing the infrastructure for inference (running AI models) rather than just those focused on training. As models become more reasoning-heavy, the demand for "action-time compute" will skyrocket.
• Margin Compression vs. Volume: While the cost per unit of AI capability is dropping by 1,000x, this "hyper-deflation" suggests that AI will become embedded in every software product. Look for companies that can leverage these lower costs to reach massive scale.
• Expect Volatility in Expectations: Because the market is currently expecting "2x" growth while the technology is moving at a "1,000x" pace, there is a significant gap between current valuations and the potential reality of the next 12 months.

Compute Infrastructure & Hardware

The shift toward inference-time compute and reasoning models changes the requirements for the hardware and energy sectors supporting AI.

• Shift in Bottlenecks: For 40 years, the bottleneck was training. Now, the bottleneck is the ability to perform "chain of thought" reasoning quickly and at scale.
• Increased Demand for "Action Time" Power: As AI models are required to "think" more before they speak, the hardware required to support these long chains of reasoning will become the most valuable asset in the ecosystem.

Takeaways

• Hardware Providers: Companies like NVIDIA (NVDA), AMD, and specialized chipmakers are the primary beneficiaries, but the focus will shift toward chips optimized for high-speed inference and reasoning rather than just raw training power.
• Energy and Data Centers: The massive increase in "inference time compute" implies that AI will be "thinking" constantly. This suggests a sustained, long-term demand for energy and data center capacity that exceeds previous estimates based solely on model training.

Software and AI Applications

The "1,000x increase in capability per unit price" makes previously impossible software applications suddenly viable and affordable.

• Reasoning Capabilities: The transition to reasoning models means AI is moving from simple pattern matching to complex problem-solving.
• Massive Gains in 2025: The speakers suggest that the next year will see gains that society is not prepared for, specifically because we have only been focused on this new "reasoning" approach for less than two years.

Takeaways

• Disruption of Traditional SaaS: Software companies that do not integrate "chain of thought" reasoning may quickly become obsolete as the cost of much smarter AI competitors drops toward zero.
• Investment Timeline: The "next year" is cited as a critical window. Investors should look for "AI-first" companies that are already deploying reasoning-based models (like OpenAI's o1 or similar architectures) to solve complex enterprise problems.