Ketu Desai
By Ketu Desai
This is not a stock market bubble; it is a historic value transfer from the losers of AI to winners. The forward 12-month P/E for the S&P 500 tech sector now sits at 23.6x, down from its peak above 30x last fall. It is difficult to argue for a bubble when the multiple for tech stocks is down for the year, back to where it was 6 years ago.
The tech multiple is lower, despite margins at a near 30% high. To put this in perspective, the forward multiple during the dot.com era was 60x, with many leading names north of 100x. Nvidia trades at 17x forward earnings, cheaper than Coca-Cola at 23x, which grows one-tenth as fast. The gains this year have not kept up with tech’s astounding 50.7% earnings growth. AI specific names are expected to grow 55% this year, compared to 22.1% for the broader S&P. The market is rewarding what is growing and expanding margins.
There are two transfers of value occurring.
The first is from the hyper-scalers that are spending money on CapEx to their beneficiaries. These hyper-scalers will spend $680bn on CapEx this year. AI semiconductors are expected to have $525bn in free cash flow. The combination of lower free cash flow, lower buybacks, and a shifting business model from capital light to capital intensive is driving this re-valuation. For instance, Microsoft’s multiple is down eight multiple turns this year, Meta is down six turns, and Amazon is down five.
The second transfer is from AI losers such as software companies, financial data providers, private equity companies, consulting companies, travel-booking companies to AI winners such as semiconductors and infrastructure names. Over the last year the former group of companies has been down anywhere from 20% to over 50%, while AI winners are up nearly 100%. Since the introduction of ChatGPT, software has gone from 14% of the S&P to under 9%, while semiconductors have gone from 7% to nearly 20%.
Other AI losers and hyper-scalers account for the additional donation to semiconductors. The correlation pre-ChatGPT between software and semiconductors was nearly one to one but has now fallen to just 0.30. The market is revaluing AI losers as their terminal value is increasingly uncertain and transferring that value to winners.
One of the key beneficiaries of this value transfer is memory stocks. Micron earned more in the first quarter of this year than in any single year prior to 2025 with nearly double the operating income of Walmart in the last quarter. Margins have expanded from 20% to 74% in the last two years. The multiple is 8.8x and the PEG has lost demand at well below 1.0x.
Memory has been historically commoditized, so the multiple deserves to be low for now. That said, we have a historic shortage, and it does take years to bring in new capacity. Plus, hyperscalers are signing 5-year agreements, compared to the industry standard of a year. Samsung said that “based solely on the demand currently received for 2027, the supply-to-demand gap for 2027 is set to widen even further than in 2026.” In a bubble market the terminal value should be expanding instead of barely keeping up with earnings growth and margin expansion. It is waiting for more confidence that a highly cyclical industry is transitioning to a less cyclical one before expanding the multiple.
The numbers AI-related companies are reporting are unprecedented. At an event in May, Anthropic’s CEO said that they saw 80x growth in the first quarter. Demand for agentic AI is driving unprecedented demand for a higher reliance on technology. Nvidia’s CEO Jensen Huang said, “In 10 years, we’ll have, hopefully, 75,000 employees – as small as possible and as big as necessary. Those [human] employees will be working with seven and a half million [AI] agents. The agents will be working around the clock… We are going to solve some really incredible problems.”
Citi estimates that the CPU total addressable market could expand from $29.3 billion in 2025 to $131.5 billion in 2030, or a 35% compound annual growth rate. The increased reliance on technology will continue to grow as physical AI, robotics, and self-driving cars become more prevalent. Less than 1% of the population currently uses AI to its capabilities and we are already falling behind. Google in the past year has gone from processing 480 trillion tokens to 3.2 quadrillion. Goldman estimates by 2030, consumption will rise to 120 quadrillion tokens per month.
For investors, expect much more dispersion as the market re-values both winners and losers. There is an 85% spread between what the top quintile stock is doing versus the bottom. There have been multiple days in recent weeks with more S&P stocks down than up than in overall 52-week lows & highs. Only 4% of S&P stocks have even managed to hit new highs.
The winners are winning big, and the losers are losing big, which is leading to offsetting results for many passive investors. Around 222 companies of the 500 S&P stocks are more than 20% off their high, and 109 are off more than 40%. The gains for the winners will not come easy, as periodical violent events can shake out weak hands. Using technicals and conviction in fundamentals will be critical during those times. The combination of value transfer, volatility, offsetting positions, and dispersion means that it is probably better to be a stock picker than an index investor.
Looking forward, the market will focus on the latest in AI, geopolitics, rates, and a new Fed Chair.
