The episode opens with a focus on NVIDIA's networking division, which achieved $31 billion in annual revenue, a figure many are unaware of. Last quarter alone, this division generated $11 billion, up 267% year-over-year, surpassing Cisco's entire annual networking business. This growth stems from NVIDIA's 2020 acquisition of Mellanox for $7 billion, positioning them to sell components for building AI factories. NVIDIA now provides NVLink for GPU communication, InfiniBand switches, and SpectrumX Ethernet platforms, essentially controlling both the chips and the connections between them. As hyperscalers develop training facilities, NVIDIA is selling the entire nervous system of data centers, not just processors. The discussion shifts to Micron, which reported quarterly revenue that nearly tripled year-over-year, beating estimates due to demand for memory chips in AI servers. Micron cannot scale production fast enough to meet this demand, indicating supply constraints on the memory side. Cloud providers and AI companies are competing for the same specialized high-bandwidth memory needed for training large language models. Samsung and SK Hynix face similar capacity issues, leading to pricing power that is expected to persist through at least mid-2026. This situation confirms that AI infrastructure spending remains robust despite concerns about return on investment timelines. Companies continue to invest in AI infrastructure even as returns are uncertain, reflecting ongoing market confidence. The episode highlights how NVIDIA's strategic moves have given it a tight grip on AI infrastructure. It also underscores the broader supply chain challenges in the AI sector, from chips to memory. Overall, the episode provides insights into the financial and strategic dynamics shaping AI infrastructure development.
