Google DeepMind Executive: AI's Hundreds of Billions in Capital Expenditure Is the Biggest Scientific Bet in Human History, Core Wager on "Recursive Self-Improvement"

As tech giants burn cash to build AI infrastructure on an unprecedented scale, the underlying logic of this high-stakes gamble is being clearly articulated: betting on AI achieving "recursive self-improvement."

On August 3, Jasjeet Sekhon, Chief Strategy Officer at Google DeepMind, stated publicly at the Agentic AI Summit held at the University of California, Berkeley, over the weekend that recursive self-improvement (RSI) is the "core investment logic" behind the industry's capital expenditures.

Recursive self-improvement refers to AI's ability to automatically create superior versions of itself, regarded within the industry as the next major milestone following Artificial General Intelligence (AGI).

Sekhon also admitted that current AI revenue is "not yet sufficient to support" the current scale of capital expenditure, meaning the industry faces the risk of falling into a market vacuum period.

Google's spending on AI data centers and related equipment this year amounts to approximately $200 billion, with plans to further increase investment next year.

Meanwhile, DeepMind researcher Oriol Vinyals and OpenAI co-founder Wojiech Zaremba, who shared the stage, both indicated that recursive self-improvement could be realized between 2027 and 2028.

The Logic Behind Massive Bets: Recursive Self-Improvement Replaces AGI

Within the AI industry's narrative system, recursive self-improvement is rapidly replacing AGI as the hottest concept.

AGI is the abbreviation for Artificial General Intelligence. It is one of the ultimate goals of AI research, referring to an AI system capable of reaching or exceeding human levels in most tasks with economic value.

Sekhon characterized the AI industry's current capital expenditure as "the biggest scientific bet in the history of human civilization," surpassing in scale the U.S. government's Apollo moon landing program, the Manhattan Project, and investments in internet development.

True recursive self-improvement means that AI models can independently redesign their overall architecture and develop entirely new models, thereby forming a cycle of continuous self-enhancement.

Sekhon acknowledged that current technology is "far from reaching this stage," but he pointed out that AI companies can currently use models to assist in designing certain components of other models, which can be seen as a "precursor" to recursive self-improvement. He likened this phenomenon to using steam engines to manufacture the next generation of steam engines in history.

Nevertheless, Sekhon stated:

It seems unwise to bet against the realization of recursive self-improvement.

However, Sekhon expects recursive self-improvement to appear "with high probability in the next few years."

DeepMind's Oriol Vinyals and OpenAI co-founder Wojiech Zaremba provided a more specific timeline during a panel discussion at the same summit: 2027 or 2028.

Reality of the Revenue Gap: Concerns Among Google Shareholders

Despite the grand strategic narrative, Sekhon's description of current financial realities was exceptionally frank.

He explicitly stated that revenue generated by AI currently "cannot support the capital expenditures we are undertaking," which means the industry faces the risk of falling into an "AI air pocket"—where capital expenditures are implemented on a large scale, but corresponding revenue fails to materialize in a timely manner.

This statement aligns closely with current concerns among Google shareholders. Although major AI investors, including Google, are seeing accelerated growth in software or cloud service sales, shareholders have shown some unease regarding the company's continuously rising cash consumption.

For comparison, Amazon AWS recorded a 37% revenue growth this quarter, with quarterly revenue reaching $42.2 billion, and operating profit margin increasing from 37% in the previous quarter to 39%, demonstrating a path for the accelerated realization of AI cloud service demand.

This provides some support for the commercial logic behind the industry's huge investments, but it also highlights the structural gap between revenue and expenditure admitted by Google.

Risk Landscape: From Cyberattacks to Biological Threats

At the summit, Sekhon and Dawn Song, a Professor of Computer Science at the University of California, Berkeley, who recently joined Meta's "Superintelligence" division, spent considerable time discussing the security risks posed by AI, particularly its potential as a tool for cyber and biological attacks.

Song pointed out that AI will "benefit attackers more" in the near term due to an inherent asymmetry between offense and defense:

Attackers only need one successful exploit, while defenders must prevent all attacks.

Sekhon agreed with this view and warned that as attackers pollute open-source code repositories and exploit low-quality code written by humans, "the coming period will be quite difficult," with vulnerable systems such as the U.S. power grid and hospitals being particularly susceptible.

However, Sekhon pointed to the field of biosecurity as a more severe threat. He stated:

We are very close to a world where an individual can design viruses or proteins simply by conversing with a model in natural language. This is an extremely dangerous world that remains advantageous to attackers in the long term.

He stated that addressing biological risks requires stricter licensing, monitoring, and tracking mechanisms for "biology-related materials," and revealed that Google is extending its AI-generated content watermarking technology, SynthID, to the biological field to assist DNA synthesis companies in screening potentially risky AI-generated biological sequences.