The Real AI Question Isn't Bubble Risk. It's Whether Profits Keep Up With Capex

Why Stocks Aren't Cracking With Rates Near 5%

The most interesting part of the latest market commentary isn't the bullish calls themselves. It's how durable the market has been. The 10-year Treasury yield has climbed back near 5%, and the Fed raised rates by 25 basis points in September, yet the S&P 500 sits only about 2% below its 2026 high. In past cycles, a 10-year yield at 5% would have hammered growth-stock valuations.

This time, corporate earnings are acting as a cushion. As of mid-September, the S&P 500's forward 12-month P/E had come down to roughly 19 to 20 times. That compression didn't come from a stock selloff. It came from rising earnings estimates. This is the core of Tom Lee's market thesis: the bad news hasn't disappeared, but companies are still growing profits faster than the bad news can bite.

That said, it would be too simple to call the rate hike itself good news. Higher rates raise the discount rate and the cost of capital, full stop. The market can only take comfort if this particular hike reduces uncertainty and makes the future rate path more predictable. If the 10-year holds well above 5% for an extended stretch, or oil spikes again, that comfort could evaporate quickly.

The Number That Matters Most in AI: 13 to 1

Dan Ives, citing a recent check of the Asian supply chain, put the demand-to-supply ratio for AI chips at roughly 13 to 1. He argues real supply-demand balance may not arrive until 2028, or early 2029 at the latest. It's worth being clear that this figure is not an industry-wide official statistic. It's an analyst's field estimate.

Still, the direction matters. Look at Nvidia's recent results, the scale of hyperscaler capex, and the bottlenecks in HBM memory and data center power together, and it's hard to argue the AI infrastructure buildout is close to finished. Ives frames it as every dollar of AI capex generating $5 to $6 in spillover spending across software, networking, security and power infrastructure. That multiplier isn't an economically validated fixed constant either. It's his framework for describing how the spending spreads.

If that framework holds, the next wave of winners isn't limited to GPU makers. The center of gravity in AI investment is widening from accelerators to memory, networking, power, cooling, data centers and software. That's also why software stocks are being re-rated, from "companies AI might replace" to "companies that sell AI to their customers."

Nvidia's 16x P/E Is Both Right and Wrong

Calling Nvidia "a 16-times-earnings stock" oversimplifies things. As of the September 17 close, its trailing-twelve-month P/E was about 28 times, and its forward 12-month P/E ranges roughly 18 to 24 times depending on the data provider. The 16-times figure mostly shows up when you apply projected earnings for fiscal 2028, a longer-dated forward multiple.

That distinction matters. The case that Nvidia is cheap rests not on current earnings but on the assumption that the market's elevated earnings estimates for the next one to two years actually materialize. Rubin-generation shipments, HBM supply, hyperscaler capex and gross margins all have to deliver on that assumption. If growth slows even slightly, that 16-times number can disappear fast.

Apple, Palantir and Tesla Aren't the Same AI Bet

Ives frames Apple as the distribution channel for consumer AI, Palantir as the proof case for monetizing enterprise AI software, and Tesla as the flagship for physical AI and robotics. All three are tied to AI, but the investment logic behind each is entirely different.

For Apple, what matters is not the AI model itself but its distribution power through more than 2 billion active devices and its services ecosystem. Palantir has proven strong revenue growth and cash flow, but its valuation is elevated too, making a growth slowdown its biggest risk. For Tesla, the key question is how fast autonomous driving and Optimus convert into actual revenue and profit.

Ives has indeed floated roughly an 80% probability of a Tesla-SpaceX merger, but the timing he points to is closer to around 2027, not "late 2025." Elon Musk has mentioned overlap between the two companies' businesses, but he has never confirmed a merger. This is better treated as an option value on top of the core investment case, not as part of the base case itself.

Insight Times Editorial Desk