Keep pulling the thread on Benedict Evans.
The mobile network industry, despite being a trillion-dollar market and spending $200 billion annually on CapEx, has low profitability because most of the value was captured by application-layer companies like Uber.
Large language models (LLMs) do not have network effects in the same way that operating systems like Windows do.
AI tools have achieved unquestionable product-market fit within the software development industry.
The CEO of SK Hynix predicts that the supply and demand imbalance for high-end memory chips used in AI will persist until 2030.
The rapid success of xAI's model in reaching top leaderboard positions indicates that building a competitive foundation model is achievable with sufficient capital and talent, suggesting a lack of deep, fundamental barriers to entry.
TSMC's net income was half of Apple's last year, illustrating that even a company with a near-monopoly at a critical infrastructure layer does not capture the majority of the value in the ecosystem.
The market for AI-powered coding tools has been successfully captured by the foundation model providers themselves, rather than by third-party application companies.
A key difference between the current AI platform shift and previous ones is that the underlying technology has a high marginal cost, making it economically unfeasible to support a consumer application with a billion users before establishing a revenue model.
There is currently no clear path to sustainable competitive differentiation for foundation model companies, as a new model tops performance leaderboards every 4 to 6 weeks.
OpenAI's strategy of hiring late-stage Meta executives led them to apply a "scaling" playbook prematurely, before achieving foundational product-market fit, a phenomenon described as a "cargo cult" approach.
Geoff Hinton's prediction 10 years ago to stop training radiologists was flawed because he did not understand the full scope of a radiologist's job.
The global installed base of smartphones is currently 5 billion.