Benedict Evans argues that foundation models lack sustainable competitive moats like the network effects seen in operating systems. He predicts they will become low-margin, commodity infrastructure, with economic value accruing to the applications and services built on top of them.
The AI sector is experiencing an unprecedented wave of capital expenditure, with major tech companies projected to spend hundreds of billions on infrastructure. This is driven by a competitive necessity to avoid being disrupted, creating a dynamic analogous to the mobile network buildout, which saw massive investment but flat returns for operators.
While generative AI has many potential uses, agentic coding is the first application to achieve undeniable product-market fit, with customers actively pulling the technology from developers. This success has narrowed the industry's immediate focus and serves as the primary driver for the current surge in demand for AI capacity.
The current AI moment is frequently compared to past shifts like PCs, the internet, and mobile, which were also transformative but had uncertain initial applications. However, a key difference is that the physical limits and rate of improvement for AI are unknown, making its trajectory less predictable than prior technological waves.
Enterprises are adopting AI not as a general-purpose chatbot but for specific, point solutions that automate back-office processes and solve tangible business problems like cash flow forecasting. This shift threatens to disrupt the existing SaaS landscape, as AI-native solutions may render many current applications obsolete.
Keep pulling the thread on Benedict Evans.