Dylan Patel: GPT-5, NVIDIA, Intel, Meta, Apple

The router points to OpenAI's future business strategy. Consider model companies: Anthropic focuses entirely on B2B, offering API access. While OpenAI has a Codex and API business, most of its revenue comes from consumer subscriptions. However, they lack a method to upsell or monetize free users. In other consumer apps, free users generate revenue through ads, but this isn't compatible with AI. An AI assistant can't degrade user results by injecting ads, and banner ads are ineffective. The challenge is monetization. I believe the router brings them closer to monetizing these users.

We've already seen AI's value creation. There's the famous $300 billion problem, now a $600 billion problem. Soon, it will be a $1 trillion problem. There's some reality to that, of course, but it ignores that infrastructure spending today accounts for five years of revenue, not just one, and revenue is growing, not flatlining. The main thing is that AI is already generating more value than the spend; it's just that the value capture is broken. I legitimately believe OpenAI isn't even capturing 10% of the value they've already created in the world just through ChatGPT usage.

The biggest factor is the orders from Google, Amazon, and Meta. Their custom silicon, unlike Microsoft's, isn't as good. However, these three companies have significantly increased their orders over the past year. Amazon is producing millions of Trainium chips, and Google is creating millions of TPUs. While TPUs are fully utilized, Trainium isn't yet, but I believe Amazon and Anthropic will resolve this. The main threat to Nvidia is the broader adoption of custom silicon. If AI becomes more concentrated, custom silicon will perform better.

A lot of capital is being invested to dislodge Nvidia from its top position, but it's challenging. How do you beat Nvidia? Hyperscalers are fortunate because they can largely replicate Nvidia's capabilities for their own captive customers. This is a huge advantage, allowing them to win on supply chain by using cheaper providers. It's essentially a margin compression exercise. For certain workloads, like Meta or recommendation systems, they might specialize more effectively. However, for the most part, they are targeting the same workloads as Nvidia.

Chips account for 60% to 80% of a cluster's cost, depending on the specific chips. Companies have already purchased these chips but can't deploy them because data centers aren't ready. This issue affects Google, Microsoft, Meta, and many others. Building power infrastructure in the US is incredibly difficult. This includes the power grid, interconnections, transmission substations, and even finding electrical contractors and electricians. In Texas, a traveling electrician can earn oil-field-level pay. It used to be that if you were physically able, you could make $100,000 in West Texas, but who wants to do that now?

The U.S. needs Intel, and I believe the world needs Intel. Samsung is performing worse than Intel in leading-edge process development. Based on various industry customers who have tested chips from both Intel and Samsung, the industry generally agrees that Intel is further along in 2-nanometer class process technology than Samsung. However, both are far behind TSMC, which is, to some extent, a monopoly. A common question is why TSMC isn't making more money or why they are only raising prices by 3-10% next year. As a monopoly, they could raise prices significantly more, but they are "good Taiwanese people" rather than "dirty American capitalists."

Jensen has a massive balance sheet and incredible cash flow. The new Trump tax bill allows companies to depreciate the entire cost of GPU clusters in year one. This has huge tax implications for companies like Meta, saving them billions annually. Nvidia could leverage this by investing in infrastructure. While this might seem like competing with their own customers, they are already doing so, as customers are developing their own chips. Nvidia should accelerate the data center ecosystem through strategic investments.