LIVE: Microsoft CEO Satya Nadella in Conversation with Larry Fink at Davos WEF | AC1G
- By
- DRM News
- Length
- 35 min
- Language
- English
- Text from
- the video's captions
- Transcribed
- 13 October 2026
Summary
At Davos, BlackRock's Larry Fink interviews Microsoft CEO Satya Nadella about AI's place in the long arc of computing, its diffusion across firms and countries, and its organizational impact. Nadella argues AI is a platform shift at least as big as the web or mobile, that diffusion depends on token efficiency, capital, grids, and skilling, and that sovereignty now means firms controlling models that embed their tacit knowledge. He closes by predicting a multimodel world where orchestration and context engineering are a firm's real IP.
Key points
- Nadella frames AI as the latest stage in a 70-year arc of digitizing people, places, and things, comparable to or greater than the PC, web, mobile, and cloud.
- He traces AI's evolution in software engineering from GitHub Copilot code completions, to chat, to agent mode, to fully autonomous agents working 24/7.
- Fink says BlackRock now processes $14 trillion of other people's money, with computations that once took 12 hours now taking minutes thanks to AI.
- Nadella argues diffusion is everything: tokens per dollar per watt must improve, and countries need capital investment, modernized grids, and broad skilling.
- He warns AI will lose 'social permission' if tokens don't improve health, education, public sector efficiency, and private sector competitiveness.
- A telltale sign of a bubble would be if only tech firms benefit; real success means AI-accelerated outcomes like drugs, healthcare, and finance across the world.
- Nadella says firm-level sovereignty means embedding tacit knowledge in models you control, or you leak enterprise value to a model company.
- He predicts a multimodel world where firms orchestrate closed and open source models with their own context and data as their IP.
Questions it answers
00:00AI in the arc of computation
How does Satya Nadella contextualize AI historically?
He sees AI as the latest step in a 70-year arc of digitizing people, places, and things, of the same class as the internet, mobile, PC, or cloud, maybe greater.
- AI belongs to the long arc of computation.
- Software lets us reason about the world in malleable digital form.
01:50Copilot to autonomous agents
How did AI capability evolve in software engineering?
GitHub Copilot code completions first built his conviction, then chat, then agent mode, and now fully autonomous agents that work 24/7, though coherence over long tasks still needs improvement.
- Agents still sit within human agency.
- Code becomes an output like documents.
04:47BlackRock's scale
What has AI done for BlackRock's productivity?
Fink says computations that took 12 hours now take minutes while processing $14 trillion of other people's money, and BlackRock could not operate at its scale without AI.
- AI enables scale otherwise impossible.
- Bending the productivity curve creates surplus.
05:55Diffusion and skilling
How does AI diffuse across economies and people?
Supply-side efficiency (tokens per dollar per watt), capital investment, modern grids, and above all broad skilling; otherwise AI loses social permission to consume scarce energy.
- Token factories will spread like the electricity grid.
- Diffusion correlates strongly with skilling.
- A rural Indian farmer used an early GPT bot in a local language.
14:54Bubble or real economy
Is AI a bubble?
A bubble sign would be if only tech firms benefit; real diffusion means AI-accelerated outcomes like drugs and healthcare across sectors and countries.
- Demand must spread beyond the supply side.
- Countries that diffuse fastest win.
17:21Organizational change
How must organizations change with AI?
Copilot flattens information flow — Nadella gets a 360 brief on a meeting and shares it across functions; firms need mindset, skill set, and data set (context engineering), and must redesign structure.
- Information flow inverts and flattens.
- Immediate productivity gains require hard change work.
22:48Barbell adoption
Who is adopting AI — big companies, small, which countries?
It's a barbell: startups adopt easily, large firms have data and relationships but face bigger change management; globally, quality of talent is similar everywhere but US adoption energy is higher.
- Neither incumbents nor entrants can coast.
- Financial sector adopted AI faster than cloud.
27:30Europe and sovereignty
What does Nadella say about Europe and sovereignty?
Europe should focus on global competitiveness of its industrial and financial output; firm sovereignty means embedding tacit knowledge in models you control, or you leak enterprise value.
- Data center location is the least important issue.
- Comparative advantage must be preserved in the AI era.
Notes
The arc of computation
- Satya Nadella frames AI within a 30–70 year arc of digitizing people, places, and things to build analytical and predictive power — from mainframes to mobile cloud.
- He ranks AI as at least the class of the internet, mobile, PC, or cloud, and maybe greater.
From Copilot to agents
- His belief in this generation of AI built up when GitHub Copilot started doing code completions.
- The progression: completions → chat in the coding flow → agent mode for small tasks → fully autonomous agents given entire projects, working 24/7 (though long-term coherence still needs work).
- He compares to the early 80s: nobody predicted 4 billion people typing daily, which created knowledge work; AI will similarly amplify human agency.
- Bill Gates' old question — what's the difference between a document, a website, and an application? — is finally answered by AI transforming one into another via code.
BlackRock at scale
- Larry Fink: computations that took 12 hours now take minutes, processing $14 trillion of other people's money with hundreds of thousands of mandates.
- Fink: without AI, BlackRock could not function at its current scale.
Diffusion is everything
- Nadella: the zeitgeist admires AI in the abstract, but it must change outcomes for people, communities, countries, and industries — or it loses social permission to spend scarce energy on tokens.
- Supply side: tokens per dollar per watt must monotonically improve; token factories will diffuse like electricity grids.
- Demand side: every firm must use it; skilling broadly is the strongest correlate of diffusion.
- He cites a rural Indian farmer using an early GPT-3/3.5 bot in a local language to reason over farm subsidies and complete a form.
Bubble test
- A telltale sign of a bubble would be if only tech firms benefit; real diffusion means AI-accelerated drugs, healthcare, and sector-wide gains.
- Fink: countries and companies that diffuse technology fastest will be the ultimate winners.
Organizational change
- Copilot flattens information flow — Nadella gets a 360 brief on Larry Fink and shares it across functions, inverting the old trickle-up structure.
- Formula: mindset → skill set → data set (context engineering).
- Adoption is a barbell: small companies start fresh; large firms have data and relationships but face a bigger change-management challenge.
Europe and sovereignty
- Europe should worry less about data sovereignty and more about global competitiveness of its industrial and financial companies.
- Firm sovereignty means embedding tacit knowledge in weights you control — otherwise you leak enterprise value to a model company.
Multimodel future
- Not one dominant model: firms orchestrate closed and open source models, distill their own, and feed them context — that orchestration is the IP.
Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.