DeepMind Essay: $1B of AI Work Could Cost $1,000 in 10 Years

A new essay from the DeepMind Institute argues that AI work costing $1 billion in compute today could cost about $1,000 in ten years, if efficiency trends hold (essay).

Google DeepMind research scientist Tim Genewein wrote it; co-founder Shane Legg shared it on X on 19 November. The Institute says its pieces reflect the author’s ideas and “should not be read as Google’s official view”.

Business Pill · GPUs Explained

An evergreen Business Engineer explainer on GPUs and compute. It teaches the general idea only and does not discuss the DeepMind Institute essay.

The key insight: A DeepMind Institute essay by Google DeepMind’s Tim Genewein says effective compute is growing at least 10x a year, and that efficiency gains of about 4x a year could make $1 billion of today’s AI work cost $1,000 in ten years. It names four routes to superintelligence.

The Compute Math

The essay multiplies three trends. Compute per dollar has risen roughly 1.5x a year for six decades, it says.

Training compute needed for a fixed performance level has fallen roughly 3x to 6x a year over the last decade. Spending on AI hardware has grown roughly 2.5x a year.

Together, it says, global effective compute stock grows “at least 10x per year”, or “a factor of 10 billion over ten years”.

Separately, hardware and software efficiency alone would cut the compute needed for a given AI workload by about 4x a year. On that basis, “A cognitive workload that can be performed by today’s AI systems for $1 billion in compute tokens would cost just $1000 in ten years”.

The essay cites compute per dollar up about 1.5x a year, AI hardware spending up about 2.5x and training effic
The essay cites compute per dollar up about 1.5x a year, AI hardware spending up about 2.5x and training efficiency up about 3x to 6x, for effective compute growth of at least 10x a year.

What Was Posted on X

Shane Legg, a Google DeepMind co-founder, shared the DeepMind Institute essay, saying it examines the path from AGI to ASI, which he called artificial general super intelligence.

What It Calls Cognitive Industrialization

The essay calls the result “cognitive industrialization”: routine cognitive work standardized and automated, much as mechanical industrialization did to physical labor.

It says the effective population of today’s AI agents might scale up by more than a thousand-fold by 2030, and more than a billion-fold within ten years. It also says this is “certainly not a given”.

How It Defines ASI

The essay places AI on a continuum. Human-level AGI is one point; ASI sits further along, as a system that “consistently outperforms large groups of human experts on tasks that could take them years to complete”.

It names four possible routes from AGI to ASI: scaling models and data, technological paradigm shifts, better coordination among many agents, and recursive self-improvement, where AI improves its own successors.

Why One Brake May Not Work

Because the routes compound, the essay argues that restricting only one input, such as training compute, “won’t be enough to stop overall progress”.

It warns that holding back one dimension while others advance risks a capability “overhang”, with sudden jumps when the constraint is lifted.

Its recommendation: “model, measure, forecast, and repeat”, with frontier labs and national and global institutions tracking progress on each route.

Where It Sits in the Debate

The essay follows a DeepMind paper we covered on how AGI could change wages and firms (DeepMind’s four AGI regimes).

It also lands the same day Anthropic’s Tao Lin predicted superhuman AI in about 1.3 years (our piece). The DeepMind essay gives no date for ASI; it calls fast progress over the next decade “a serious possibility”.

The Structural Read

As we read the essay, its central claim is about cost, not a date: it argues AI work gets cheap enough to scale by orders of magnitude even without new breakthroughs.

As we read it, the policy point is that capping one input, such as training compute, would not stop progress along the other routes.

Tim Genewein, DeepMind Institute essay, 19 November 2026

“Human-level AGI has long been the horizon for AI research, but it is far from the upper limit of intelligence.”

Three Implications

FOR AI BUYERS The essay says a workload costing $1 billion in compute today could cost about $1,000 in ten years if efficiency trends hold.

FOR POLICYMAKERS It argues restricting a single input such as training compute won’t be enough to stop overall progress.

FOR RESEARCHERS It calls for multi-agent scaling laws and new benchmarks to track each route to superintelligence.

What Is Not Established

The essay gives no timeline for AGI or ASI and calls none of its predictions certain. Its compute figures are trend extrapolations, and it notes practical ceilings on compute build-out.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The Map of AI

A Business Engineer map of 200+ AI companies across 9 layers.

Read the Map of AI →

The Bottom Line

A DeepMind Institute essay argues effective compute is growing at least 10x a year and that efficiency gains could make $1 billion of today’s AI work cost about $1,000 in ten years. It says the views are the author’s, not Google’s.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. All figures and quotes are from the DeepMind Institute essay by Tim Genewein dated 19 November 2026, read in full; the Institute says its pieces are not Google’s official view. Shane Legg’s post is cited for its sharing of the essay.

Sources: DeepMind Institute, ‘What comes after AGI? From AGI to Superintelligence’, Tim Genewein, 19 Nov 2026 · Shane Legg on X, 19 Nov 2026

Scroll to Top
FourWeekMBA