Keep pulling the thread on Dylan Patel.
Google's Gemini Pro is the largest production model currently deployed by any major AI lab, exceeding the parameter count of OpenAI's GPT-5.4 and Anthropic's Opus.
Apple's share of TSMC's revenue is expected to decline over time, making it a less relevant customer and reducing its ability to dictate terms.
The first customer for TSMC's A16 node is expected to be an AI company, not Apple.
The speaker argues that if Huawei had maintained access to TSMC's foundries, it would likely be a more capable company than NVIDIA.
Elon Musk has signed a deal with Samsung to manufacture chips for Tesla's humanoid robots at its facility in Texas.
In a scenario where Taiwan's semiconductor fabs are destroyed, China would possess a relatively stronger and more vertically integrated semiconductor supply chain than the rest of the world.
The loss of Taiwan's semiconductor fabrication capacity would cause a massive contraction in global GDP.
The combined forecasted capital expenditure for Amazon, Meta, Google, and Microsoft for the current year is approximately $600 billion.
The total capital expenditure across the entire AI infrastructure supply chain, including hyperscalers and their suppliers, is on the order of $1 trillion for the current year.
Approximately 20 gigawatts of incremental data center capacity will be added in the United States this year.
Based on its revenue growth trajectory, Anthropic will need to add 4 gigawatts of inference capacity to sustain its growth over the next 10 months.
The market rental cost for 1 gigawatt of AI data center capacity is approximately $10 billion per year.