The exponential leap: Understanding AI's real impact

Matt Clifford CBE explores AI's rapid evolution and its profound implications for markets. With computational power doubling every seven months, knowledge work faces unprecedented disruption—but significant challenges remain ahead.

By Lisa Tugwell
Published | 2 min read

Key points

  • The computational power required to train AI models has increased a billion-fold in the past decade, representing an unprecedented acceleration unlike any technology in history.
  • AI's capability to delegate complex human tasks is doubling every seven months, with models expected to handle month-long work assignments by decade's end, fundamentally disrupting knowledge work.
  • Five critical constraints—technical barriers, business models, supply chains, power infrastructure, and political backlash—may slow AI progress, with political resistance currently underpriced by industry.

The AI inflection point: Why 2026 may be the last normal year

The financial markets are gripped by AI fervor, yet according to Matt Clifford, former advisor to the UK government and chair of the Advanced Research and Invention Agency (ARIA), the real disruption hasn't truly begun. In a compelling keynote at RBC’s recent UK Focus Conference, Clifford presented a case that challenges both the sceptics and the believers in the AI revolution.

"The increase in computational power over that period is about a billion fold, which is really striking. I think it is worth dwelling on that. I think there are very, very few, I'd say probably no technologies in the history of the world where we've seen a billion fold increase in the key input in a very short period of time."

Matt Clifford CBE, Vice Chair, AI Safety Institute, Co-Founder of Entrepreneur First

This is not a bubble—it's a structural shift

Clifford reframes the AI debate entirely. Unlike previous technological hypes, today's AI advancement is grounded in a measurable, sustained fourfold annual increase in computational power for over a decade. This isn't speculation; it's an empirical trend underpinned by real investment from tech giants committing to half-trillion-dollar capital expenditure. The comparison is instructive: Moore's Law itself had no theoretical justification, yet became a self-fulfilling prophecy through market forces. AI appears to be following the same trajectory.

The complexity of tasks AI can handle is accelerating

Perhaps the most actionable metric Clifford introduced is the length of complex work that can reliably be delegated to AI models. At the start of 2023, this was measured in minutes. By summer 2025, it had stretched to hours. As of December 2025, frontier models achieved four-and-a-half-month task capability. If the doubling trend continues—and Clifford believes it will—by 2029 we may see AI capable of handling month-long assignments. This trajectory suggests that by the end of this decade, the majority of knowledge work will be susceptible to AI automation.

"If you double an hour seven times, you go from an hour to a working month. A world where you can delegate an hour of work to an AI is not actually that different from one where you can delegate a minute. But one where you can delegate a month is very different from one where you can delegate an hour."

Matt Clifford CBE, Vice Chair, AI Safety Institute, Co-Founder of Entrepreneur First

Five barriers may—or may not—stop the train

Clifford identified five potential constraints worth monitoring: technical breakthroughs, business model viability, supply chain capacity, power infrastructure, and political backlash. While each presents genuine challenges, he makes a compelling argument that none is insurmountable given sufficient capital and determination. The exception is politics. Public skepticism about AI—driven by job losses, environmental concerns, child safety, and copyright issues—represents the most underpriced risk in his assessment.

Investors must plan for disruption, not absence of it

The financial sector faces an uncomfortable truth: the base case for AI's impact may be more transformative than most investors are currently pricing in. This suggests that holding portfolios as though AI disruption won't happen represents a riskier bet than assuming it will. Companies positioned to benefit from automation in knowledge-intensive sectors warrant particular attention.

2026 may be the last "normal" year before widespread disruption

For financial institutions, regulators, and corporate strategists, the timeline for AI-driven transformation is compressed. Industry experts and researchers suggest that 2026 represents a pivotal threshold, beyond which significant disruption from AI becomes unavoidable. This compressed window means businesses must accelerate adaptation of their models, workforce strategies, and risk frameworks now. Those who continue planning assuming business-as-usual beyond 2026 may find themselves fundamentally unprepared for the acceleration that follows. The time for proactive strategic positioning is effectively now.

Our expert

Lisa Tugwell
Lisa Tugwell
Head, Pan-European Equity SMID Sales, RBC Capital Markets

 

Stay informed

Get the latest insights and news from RBC Capital Markets delivered to your inbox.