Current AI market valuations are contradictory. If high multiples for power, cooling, and optical companies are justified, then the lower multiples for chip makers like NVIDIA are wrong, and their stocks have significant room to grow. Both sets of valuations cannot be correct simultaneously.

When a company is growing 10x or 50x year-over-year, obsessing over the entry multiple is a mistake. An initially 'insane' valuation can look cheap in retrospect. The primary focus should be on determining if the company is on an exponential curve; price is the least important factor in that equation.

AI software is improving so rapidly that older hardware, like a three-year-old NVIDIA inference chip, is now more profitable than it was when new. This phenomenon, where software advancements outpace hardware depreciation, is unprecedented and makes existing infrastructure increasingly valuable.

Despite massive growth, Nvidia's stock trades at a modest 24x earnings multiple, implying the market is pricing in a 'peak year' scenario. In contrast, AI ecosystem partners like AMD and Broadcom have higher multiples, suggesting greater investor confidence in the long-term AI cycle itself.

NVIDIA's revenue growth is speeding up even as its revenue base expands massively, a rare feat that defies the "law of large numbers." This suggests strong network effects and a dominant market position are creating a self-reinforcing cycle of demand for its AI hardware.

Despite its massive market cap, NVIDIA trades at a forward P/E of 19, below the S&P 500 average. Tae Kim argues this is due to misplaced skepticism about a "peak," comparing it to Apple's single-digit P/E during its iPhone growth era, suggesting a major stock re-rate is possible through buybacks.

Andreessen highlights a unique economic phenomenon: the pace of AI software improvement outstrips hardware depreciation. This means a three-year-old NVIDIA inference chip can generate more revenue today than when it was new, a complete reversal of typical tech hardware value cycles.

A first-principles analysis shows that for NVIDIA's stock price to be justified, the company would need to pay out 100% of its revenue as dividends for 10 years, with zero costs, R&D, or taxes. This highlights how detached hype-driven valuations can be from fundamental business reality.

The Holt valuation framework, which prioritizes cash-based returns, indicates Nvidia could be worth 400% more. Unlike most high-growth companies whose projected earnings are 'faded' over time in the model, Nvidia's performance is so strong that even after applying a significant fade, it still appears dramatically undervalued.