SoftBank CEO Masayoshi Son: AI Is Now Building the Next AI, Not Humans

OpenAI’s next frontier model won’t be designed by human engineers. It will be designed by AI itself. That’s the core claim SoftBank CEO Masayoshi Son made in a far-reaching CNBC interview in Paris, aired in early June 2026 .

Son argued that the most advanced AI systems have already crossed a critical threshold: they can now architect their own successors, recursively pushing intelligence beyond what any human team can build. The development, he said, marks an early sign of artificial superintelligence (ASI) and compresses the timeline for AI that far surpasses human cognition.

OpenAI’s next model will be designed by AI

Son stated plainly that humans are becoming the bottleneck. OpenAI’s next model will be built by AI rather than by human engineers because people “will no longer be smart enough” for the work . The claim signals what Son views as a landmark shift—AI moving into territory where it can recursively improve its own architecture without direct human design .

That framing is bold and carries an implicit prediction: once a system can improve itself, the intelligence curve steepens dramatically. Son says models built this way “are going to be exponentially smarter than all of us” .

A sharply revised superintelligence timeline

Son’s latest forecast represents a dramatic acceleration from his previous public roadmap. As recently as 2024 and 2025, he had told shareholders and conference audiences that ASI—defined as AI 10,000 times smarter than the human brain—would arrive in roughly 10 years, with AGI (human-level intelligence) arriving in three to five years .

In the CNBC interview, he collapsed that timeline. Son now says ASI could emerge within “the next couple of years” . The shift from a decade to roughly two years is one of the fastest public timeline accelerations from a major tech investor and marks a decisive break with his earlier, more gradual projections.

SoftBank’s $64.6 billion bet on OpenAI

Son isn’t just forecasting; he’s positioned SoftBank to ride the entire curve. The Japanese conglomerate now holds roughly 13% of OpenAI, valued at about $64.6 billion, making it one of the company’s largest external shareholders .

The OpenAI stake is the centerpiece of a broader AI strategy built on vertical integration. SoftBank is assembling the stack from chips to data centers to the leading model layer. Key pieces include:

• €75 billion ($82B+) in AI infrastructure investment in France, announced the day before the CNBC interview .
• The Stargate project, a massive U.S. data-center build-out, and the Izanagi chip initiative .
• Ownership of Arm, which provides foundational chip architecture for the AI industry .

Son framed this entire push as a bet on an epochal shift. He told CNBC the AI revolution is “50x bigger than the dot-com boom” and called it “the biggest revolution of technology and realization that mankind ever experienced” . He said he personally uses ChatGPT two to three hours a day and described his AI conviction as a 50-to-100-year story .

What Son says about AI vs. human expertise now

A defining theme of the interview was the claim that current frontier AI has already eclipsed human ability to design the next generation of models. Son’s stance is that humans are now the constraint—and that removing human designers from the loop is the natural, necessary next step .

He framed the transition not as a distant hypothetical but as a present reality. Once AI can improve itself, he argued, the path to superintelligence becomes a near-term certainty rather than a long-term speculation. That claim, if true, has profound implications: it would mean the industry has entered recursive self-improvement territory sooner than most forecasters expected .

The risks behind the rhetoric

Son’s vision carries meaningful financial and strategic risks. The OpenAI stake was funded in part by a $40 billion bridge loan secured in March 2026, maturing in 12 months, which shifted S&P’s outlook on SoftBank to negative and pushed its debt-to-equity to 137.6% . OpenAI itself isn’t projected to be profitable until 2030 .

The strategy amounts to a concentrated, leveraged bet on exponential AI adoption. Success hinges on OpenAI maintaining its lead and on 2027 refinancing outcomes, while the ASI timeline Son sketched in Paris remains an extraordinary prediction that will face deep scrutiny as the months unfold .