DeepMind Exec Reveals True Purpose Behind Trillion-Dollar AI Spending
Google DeepMind's strategy chief explains the real goal behind massive AI investments.
1Answer to the Trillion-Dollar Question
Jasjeet Sekhon, Chief Strategy Officer at Google DeepMind, revealed the true purpose behind the industry's trillion-dollar AI investments. Speaking at a UC Berkeley summit, he stated these expenditures are bets on 'recursive self-improvement' (RSI) technology. RSI refers to AI's ability to rewrite and upgrade its own code. Sekhon emphasized this capability doesn't exist yet, but investments are moving in that direction.
2Steam Engine Analogy
Sekhon explained RSI with a historical analogy: 'Steam engines built the next steam engine.' This comparison points to AI's potential to continuously produce improved versions of itself. The DeepMind executive admitted current AI revenues don't cover capital expenditures. However, he expressed confidence these investments will yield significant returns in the future.
3The New AGI: RSI
The industry long justified investments with the goal of Artificial General Intelligence (AGI). Sekhon's statements indicate the focus has shifted to RSI. This new target promises to transform today's data centers from cost centers into 'the most valuable machines ever built.' If successful, RSI could create systems that improve autonomously without human intervention.
4Massive Investments
Alphabet (Google's parent company) spent $44.9 billion on capital projects in a single quarter - nearly double the previous year's amount. The company raised its 2026 guidance to $205 billion and promised another 'significant' increase in 2027. Amazon, Microsoft and Meta announced similar investment levels. Sekhon compared this effort to something larger than the Apollo program or Manhattan Project.
5Revenue-Spending Gap
Google Cloud revenue jumped 82% last quarter with over $500 billion in backlog. But spending is growing much faster: Alphabet reported its first-ever negative free cash flow (-$5.9 billion). Sekhon highlighted the 'AI air pocket' risk: Expenditures might occur without corresponding revenues. This concern underlies every hyperscaler's earnings call.
6RSI Uncertainties
Recursive self-improvement isn't yet a shippable product. Significant questions remain about safety, control and technical feasibility. The 2027-2028 timeline Sekhon implied isn't guaranteed either. Competitors question whether DeepMind has the know-how to achieve this before OpenAI or Anthropic.
7Partially Realized Vision
A modest version of RSI already works: AI models can generate code and improve themselves in narrow ways. But there's a vast gap between this and the fully autonomous self-enhancement Sekhon describes. Numerous technical challenges must be overcome for this leap to occur.
8Explicitly Stated Risk
The most striking aspect of Sekhon's comments is his admission that the industry is making Apollo-sized bets on a capability that doesn't yet exist. While refreshingly honest, this is also somewhat terrifying for investors. Successful RSI could bring massive returns, but failure might waste trillions in investments. This represents one of technology's biggest gambles.