What AMD Just Did Is Insane (Investors Aren't Ready)
Summary
Alex argues that the bear case for AI—the idea that spending will eventually collapse—is being refuted by the actions of major technology companies like Alphabet, Amazon, Meta, and Microsoft. Alex points out that these firms are increasing their capital expenditure budgets to a projected $700 billion this year, despite significant hits to their free cash flows. Alex describes the race for data center infrastructure as a zero-sum game for limited land and power, ensuring that spending remains high for the foreseeable future.
Alex breaks down how the industry is evolving through three different corporate strategies:
Mentioned Stocks
Reasoning: Alex explains that Nvidia is turning GPUs into a financial asset class to ensure continued spending from conservative institutional money. Alex mentions that Nvidia's software, like TensorRT-LLM, effectively doubles hardware performance for free, extending the value of the chips. Alex views Nvidia as a strong play but notes the risk of chip depreciation on the secondary market.
Reasoning: Alex highlights AMD's acquisition of Talus as a game-changer for the 'memory wall' bottleneck. Alex states that etching model weights directly into the chip allows for 120x speed increases and costs less than a penny per million tokens to run. Alex views AMD's pivot to hyper-optimized, swappable, low-cost chips as a winning strategy as the market shifts toward inference efficiency.
Reasoning: Alex points to Cerebras' massive wafer-scale engine which avoids the latency of traditional networking cables. Alex notes that their backlog is worth roughly 29 times their expected 2026 revenue, and their cloud services revenue is exploding. Alex believes their unique hardware architecture is superior for moving massive amounts of data compared to Nvidia's smaller chips.