A senior executive at Alphabet's DeepMind has cautioned that current revenue from artificial intelligence is insufficient to support the massive capital expenditures the tech industry is undertaking, though he believes the emergence of Recursive Self-Improvement (RSI) could eventually change the equation.
Jasjit Sehion, who joined Alphabet from Bridgewater Associates in April, stated that the enormous spending on AI-related infrastructure is a precursor to the next phase of AI development. This phase, known as Recursive Self-Improvement (RSI), refers to AI systems that can autonomously create better versions of themselves. While RSI has long been considered a component of Artificial General Intelligence (AGI), the tech industry, particularly the corporate sector, has only recently begun adopting this terminology in recent weeks.
Sehion noted that AI revenue "is not yet sustainable" for the level of capital expenditures currently underway. However, he added that betting against the present and future of the tech industry "seems unwise," pointing out that the early signs of RSI are already visible. He remarked that this should not surprise anyone, as "the steam engine was used to build the next generation of steam engines."
Despite this optimism, Sehion acknowledged the possibility of an "AI air pocket" 鈥?a period where spending has occurred but revenue has not yet materialized. He compared the scale of construction, which is expected to exceed $400 billion in spending by 2025 and could surpass $700 billion by 2026, to projects larger than the Apollo program, the Manhattan Project, or the internet buildout of the 1960s.
Major hyperscale companies, including Alphabet, Amazon, Microsoft, and Meta Platforms, have repeatedly indicated that their capital expenditures for 2027 will exceed those for 2026. In late last month, Alphabet raised its 2026 capital expenditure forecast again, now expecting spending between $195 billion and $205 billion, up from the previously projected $180 billion to $190 billion.
Comments