Thanks, Michael. The AI market may see a correction (most new, rapidly hyped markets do), but the long-term fundamentals remain unchanged. The rapid drop in the cost of intelligence is too compelling for capitalism. Companies in competitive markets will be compelled to keep up, with much greater long-term costs to our economic system than a short-term bubble. I'd love to see you discuss this in a future article.
The fundamentals look solid in memory, storage, and silicon. Those are the areas with the highest margin in the tech stack because they are the biggest bottlenecks. Photonics might be promising on a significant retrace. I don't think the value is there in the models themselves. The problem is that there is nothing proprietary and the pre-training overhead is extremely expensive... especially to create a product whose model weights are seemingly distilled in 4-8 weeks... If you concede that margin compression is likely in the Silicon Valley models due to cheap open source AI that is "good enough" then eventually the most logical conclusion is that the TAM on the whole thing is much smaller than what has been fiscally modeled. I will concede that the earnings are there now in the first half of the infrastructure buildout. But there are a ton of ways this buildout can go sideways and the whole thing is priced to absolute perfection. Hence, increasing caution is strongly warranted as prices and valuations continue to melt up. Personally speaking memory is the most interesting of all the layers in the tech stack. I saw this early on last year and bougth a lot of Micron in the $100-125 range. Memory, storage, and silicon seem fairly priced to me. Hyperscalers are about right but maybe a hair expensive. The neoclouds are the wild west since they aren't well capitalized. Photonics are really tough to separate hype from losers. TSM and MU feel like the best two single megacap chokepoints but I am not a buyer at these prices.
Thanks, Michael. The AI market may see a correction (most new, rapidly hyped markets do), but the long-term fundamentals remain unchanged. The rapid drop in the cost of intelligence is too compelling for capitalism. Companies in competitive markets will be compelled to keep up, with much greater long-term costs to our economic system than a short-term bubble. I'd love to see you discuss this in a future article.
The fundamentals look solid in memory, storage, and silicon. Those are the areas with the highest margin in the tech stack because they are the biggest bottlenecks. Photonics might be promising on a significant retrace. I don't think the value is there in the models themselves. The problem is that there is nothing proprietary and the pre-training overhead is extremely expensive... especially to create a product whose model weights are seemingly distilled in 4-8 weeks... If you concede that margin compression is likely in the Silicon Valley models due to cheap open source AI that is "good enough" then eventually the most logical conclusion is that the TAM on the whole thing is much smaller than what has been fiscally modeled. I will concede that the earnings are there now in the first half of the infrastructure buildout. But there are a ton of ways this buildout can go sideways and the whole thing is priced to absolute perfection. Hence, increasing caution is strongly warranted as prices and valuations continue to melt up. Personally speaking memory is the most interesting of all the layers in the tech stack. I saw this early on last year and bougth a lot of Micron in the $100-125 range. Memory, storage, and silicon seem fairly priced to me. Hyperscalers are about right but maybe a hair expensive. The neoclouds are the wild west since they aren't well capitalized. Photonics are really tough to separate hype from losers. TSM and MU feel like the best two single megacap chokepoints but I am not a buyer at these prices.