Keep pulling the thread on Gavin Baker.
There are plans to invest an additional $3 to $4 trillion in data center capacity in the US over the next five years.
Over the past three years, the inflation-adjusted dollar amount spent on data center capacity has exceeded the total cost of the entire US interstate highway system, which was built over 40 years.
Google has reported a 150x increase in the volume of tokens processed over the last 17 months.
The largest public companies spending on GPUs have seen their Return on Invested Capital (ROIC) increase by approximately 10 percentage points since they began ramping up AI-related CapEx.
The major tech companies investing in AI capex collectively generate approximately $300 billion in free cash flow annually and hold around $500 billion in cash on their balance sheets.
Google's leadership, exemplified by an internal statement attributed to Larry Page, views winning the AI race as an existential priority, a mentality also shared by Meta.
Gavin Baker asserts that Google, not other chipmakers like AMD or Intel, is NVIDIA's most significant competitor in the AI hardware market.
Google's TPU chip is considered the only viable alternative to NVIDIA's GPUs for AI training and potentially the best alternative for inference currently available.
Google's Gemini has gained 15 to 20 percentage points of traffic share in the AI chatbot market over the last two to three months.
Gavin Baker suspects that when including search integrations, Google's total AI-related traffic is currently larger than that of OpenAI and Anthropic.
Application SaaS companies cannot succeed in the AI era without accepting pressure on their gross margins due to the high cost of compute.
The development of reasoning capabilities in LLMs has fundamentally changed their economics by enabling a user-driven data flywheel through reinforcement learning.