Sam Altman Says the AI Singularity Has Already Begun — Here’s the Reasoning Behind It

While it once seemed unfathomable that AI would surpass human intelligence, OpenAI’s founder believes an exponential intelligence explosion will let machines pursue recursive self-improvement within the foreseeable future.

Against the backdrop of the Hugging Face attack, you’d expect Sam Altman to calm frayed nerves.

Also Read: AI Agents Versus Human Traders: Nansen’s CEO Says the Contest Ends in Two Years

He did the opposite. The OpenAI founder doubled down on his position that humanity is already moving toward AI singularity — the stage where AI outstrips human intelligence, becomes fully autonomous, and starts pursuing aggressive self-improvement.

Altman admits that safety and alignment questions remain unresolved. But he still believes we’re closing in on digital superintelligence that is inherently smarter than humans in many ways.

To understand what looks like overexuberance, it’s worth digging into his reasoning — and the data points that back it up.

An AI-powered future

In a blog post written last year, Altman described a future where AI drives faster scientific progress and higher productivity.

His argument is that intelligence and the ability to turn ideas into reality have always been the hard limits on human progress. AI technology and the infrastructure around it are changing both, day by day.

As we move into a new technological revolution, he believes AI won’t just transform how we live and work. It’ll push human progress in ways that were considered impossible only a few years ago.

If 2025 brought AI agents with a proven capacity to do real cognitive work, 2026 has already delivered advanced systems that can find novel pathways to scientific problems.

These autonomous systems are being deployed in drug discovery, material science, mathematics, biology and genomics, among other fields. The agents combine computational tools, published literature, existing databases and multi-step reasoning to reach efficient solutions.

Altman thinks 2027 may bring robots capable of handling many real-world tasks on their own.

And as the industry keeps pushing, the nature of work itself could be completely redefined in the 2030s.

Pace of AI enhancements to accelerate further

Unlike the bleaker outlook painted by some of his peers, Altman argues that AI innovation is accelerating along an exponential curve — nowhere near an inflection point.

Right now we’re still impressed that AI tools can compose phrases, deliver life-saving medical diagnoses and write code with minimal human input. He thinks those capabilities will feel routine soon enough.

The numbers lend some support. The global AI infrastructure market is projected to pass $1 trillion by 2029.

Continued hyperscale investment, accelerated server adoption, growing data centre buildouts and the expansion of sovereign AI programs should mean that neither intelligence nor energy stays a limiting factor on human progress.

In scientific research, community members report two to three times more productivity after adopting AI.

Advanced AI is also speeding up AI research itself — helping discover better algorithms, new computing substrates and chip designs in a fraction of the time previously required. With OpenAI and its peers shipping cyber-capable models that have already demonstrated their prowess, albeit not in a pleasant fashion, it seems fair to expect discovery times to keep shrinking across research domains.

Several self-reinforcing loops in play

AI models and agents don’t have recursive self-improvement capabilities today.

But Altman believes existing tools will help humans gain the scientific insights needed to build better AI systems. Once machines can rewrite their own code and design better hardware, AI enters a rapid loop — using every new increment of intelligence to improve itself on the next iteration.

An intelligence explosion of sorts.

Compound that accelerated development with the economic value created by the infrastructure being built today to run these systems, and you get a genuinely new technological era. One where robots build other robots, automating even projects like datacentre production.

That scale is what drives the cost of intelligence down further, closing an interconnected, self-reinforcing loop of continuous development.

Challenges that need to be thwarted

Harnessing AI’s real potential requires the global industry to come together on current safety issues, while making sure superintelligence gets distributed widely without erecting barriers to innovation.

It also matters that AI solutions align with solving real-world problems — rather than manufacturing new wants, as many social media applications have done.

Getting firms worldwide to agree on a shared safety and alignment framework that promotes innovation while protecting users’ interests will be an arduous task.

Then there’s the challenge of making superintelligence freely available, cheap, and not concentrated in the hands of a few multinationals or governments.

That will take massive investment in upgrading electricity generation and distribution infrastructure. It also needs advances in chip manufacturing, plus new AI models that cut the electricity consumed by every query.

A future of limitless possibilities

Altman argues that with persistence and good intent, humanity can capture the maximum upside and minimal downside over the long run.

Assume for a moment that these challenges do get mitigated in the near future. The rate of AI progress would only keep accelerating from there.

The range and depth of AI capabilities would grow too, producing new discoveries and quantum leaps in human progress over the next decade.

It may not play out as simply or as smoothly as Altman describes. But it’s becoming increasingly apparent that we’re on an exponential path toward superintelligence.

Read Next: Can €10 Billion Close Europe’s AI Compute Gap? Brussels Just Placed the Bet