Just been looking at how the big tech companies are positioning themselves for the AI infrastructure race, and there's actually a pretty interesting investment angle here that goes beyond just buying Nvidia.

So we're talking about $650 billion-plus in AI spending this year from the major hyperscalers. That's an absurd amount of capital, which tells you they're betting big on AI returns. But here's the thing - the real opportunity isn't just in the companies doing the spending. It's in the supply chain feeding all that infrastructure buildout.

Start with the chip makers and foundries. Nvidia's obviously the dominant player with their GPUs and CUDA ecosystem - that software moat is real and keeps getting wider. But I've been paying more attention to some of the less obvious winners. Broadcom's been quietly making moves with custom AI ASICs. They helped Alphabet build those TPUs, and now they're working with OpenAI and others on proprietary chips. That's recurring revenue potential right there. TSMC is another no-brainer - they basically have a monopoly on manufacturing advanced AI chips, which gives them serious pricing power.

The memory story is where it gets interesting though. High bandwidth memory is becoming critical for AI workloads, and it requires way more wafer capacity than regular DRAM. Micron's locked in long-term HBM contracts with the big tech companies, which is huge because it makes their business less cyclical. That's the kind of structural shift that drives stock returns over years.

Now, the big tech companies themselves - Alphabet, Amazon, Microsoft, Meta - they're not just spending recklessly. They're embedding AI into their actual business models. Alphabet's using Gemini for search, Microsoft's got Copilot driving enterprise software adoption, Amazon's using AI and robotics for operational efficiency. Meta's recommendation algorithm is printing money in ad revenue. These aren't vanity projects. When companies invest this much capital, they expect returns, and the early signs suggest they're getting them.

Here's the weird one that caught my attention: Energy Transfer. Nobody talks about it in AI conversations, but data centers need massive amounts of power. Energy Transfer has natural gas assets in the Permian Basin - some of the cheapest in the country - and they're positioned to benefit from all the AI data center builds. Plus it trades cheap with over 7% yield. It's the kind of boring infrastructure play that actually makes sense when you think about what's required to power all this AI infrastructure.

The big tech companies spending this kind of money on AI infrastructure tells you they see real ROI potential. Following that capital into the supply chain - chips, memory, foundries, infrastructure - is probably where the next wave of gains comes from.