Just caught up on the latest capex plans from the big tech players, and the numbers are honestly staggering. We're talking about the largest tech companies planning to drop over $650 billion this year alone on AI infrastructure. That's basically the GDP of a mid-sized country going straight into data centers and chips. Here's what I'm actually watching in this space.

First, the obvious play: the chip makers. Nvidia's still the king here with its GPUs powering most AI workloads, and that CUDA moat isn't going anywhere. But the real opportunity I'm seeing is broader than just GPU makers. AMD's carved out a solid niche in inference, especially after some big wins with OpenAI and Meta. Then there's Broadcom, which most people sleep on. The company's basically the architect behind custom AI chips—they helped build Alphabet's TPUs and are now helping other players develop their own proprietary silicon. That's a multi-year revenue stream.

Taiwan Semiconductor Manufacturing is another angle. They've got the foundry monopoly on advanced logic chips for AI, which means serious pricing power. And here's something I've been thinking about: high bandwidth memory. AI chips need HBM to run efficiently, and we're talking about 3x the wafer capacity compared to regular DRAM. Micron's positioned well here as one of the big three DRAM makers, and they're locking in long-term commitments. That's huge because it makes their business less cyclical than it used to be.

Now, the large tech companies spending all this money—they're not doing it for fun. Alphabet, Amazon, and Microsoft are expecting serious returns on their cloud infrastructure plays. I like all three. Alphabet's using Gemini to juice search, Microsoft's AI copilots are driving enterprise adoption, and Amazon's using AI and robotics to squeeze out efficiencies in e-commerce. Meta's also in this conversation as a big spender, embedding AI into their recommendation engine and ad tools. The engagement and pricing power they're getting is real.

One thing people don't talk about enough: energy. AI infrastructure is power-hungry, and that's where I'm looking at Energy Transfer. They've got cheap natural gas assets in the Permian, which is basically the perfect setup for data center power needs. Over 7% yield on a beaten-down stock price? That's a sneaky way to play the AI buildout without directly touching chip stocks.

The way I see it, this AI capex cycle is just getting started. The large tech companies are clearly betting big, and the supply chain that feeds them—from chip designers to memory makers to infrastructure providers—should see sustained demand for years. Worth keeping an eye on how this unfolds.