▶The primary competition in the AI hardware market is between NVIDIA's GPUs and Google's TPUs, with other chipmakers like AMD relegated to a secondary, second-source role.Apr 2026
▶The massive capital expenditure on AI compute by large tech companies is currently justified, as evidenced by a roughly 10-percentage-point increase in their Return on Invested Capital (ROIC) since ramping up spending.
▶Physical, real-world constraints, specifically shortages of electricity ('watts') and semiconductor manufacturing capacity ('wafers'), are significant bottlenecks for the AI buildout that may prevent a 2000-style oversupply bubble.Apr 2026
▶AI is forcing a fundamental shift in the business models of software companies, requiring them to abandon high-margin SaaS models and accept lower gross margins to remain competitive in an AI-native world.
▶Baker argues AI may be a 'sustaining innovation' that benefits large incumbents with data, capital, and distribution, which contrasts with the common narrative that AI will be a disruptive force that unseats them.
▶While acknowledging a 'rolling bubble' in niche sectors, Baker actively refutes the broader 'AI bubble' thesis by comparing current valuations to the dot-com peak and highlighting rising ROIC and physical supply constraints.
▶Baker's assertion that the operational value-add from many venture capital firms is 'wildly overstated' and unappreciated by founders directly challenges the VC industry's prevailing narrative and fee structures.Apr 2026
▶The leadership in frontier AI model development is highly contested. Baker notes Google's temporary cost and hardware advantage with TPUs, but also predicts xAI will be first with a Blackwell-trained model and that NVIDIA's future chips will re-establish a wider performance gap.Apr 2026
Not enough data for timeline
Sign up free to see the full intelligence report
Get started free