Podcast Summary: "The demand for infinite compute"

Podcast: Azeem Azhar's Exponential View
Host: Azeem Azhar
Date: November 7, 2025

Overview

In this solo episode, Azeem Azhar explores the economic and societal implications of our ever-growing appetite for computational power. Moving beyond generative AI headlines, Azeem sets the stage for a broader discussion: the relentless, seemingly unbounded demand for compute in business, science, and daily life. Using rich historical context, economic analysis, and industry insider insights, he demonstrates why this "infinite demand" for compute isn't just hype, but a fundamental shift transforming the fabric of the global economy.

Key Discussion Points and Insights

The True Driver Behind Tech Investment

• Not Just AI: While generative AI garners headlines, the real story is the broader, accelerating shift to computational infrastructure across industries.
  • "We're missing something important. All of this matters, generative AI, but in a different way. The real story is a broader shift in the economy towards computation." [00:23]
• Staggering Growth: Major cloud platforms (Amazon, Google, Microsoft) have shown rapid revenue growth, a large portion driven by AI workloads for both big AI players (OpenAI, Anthropic) and mainstream enterprises.
  • "Amazon's AWS grew more than 20% to about $33 billion... Google cloud... at about $15 billion... Microsoft's Azure grew by some 40%." [01:19]

Supply, Demand, and Utilization of Compute

• Hyperscalers & Chipmakers: Tech giants invest collectively over $400 billion/year in AI hardware. Chipmakers like Nvidia are shipping millions of advanced GPUs to meet demand, with no sign of unused capacity.
  • "Anthropic... bought a million AI processing units called TPUs from Google." [02:16]
• No Excess Capacity: Contrary to past infrastructure booms that resulted in excess capacity (e.g., telecom "dark fiber"), current compute resources are heavily utilized.
  • "As Gavin Baker... says, there are no dark GPUs anywhere." [12:32]
• Not the 'Springfield Monorail': Refers humorously to building infrastructure ahead of need—not happening in compute.
  • "Is this the monorail for the global economy?... It's not like the dark fiber during the telecoms bubble that wasn't used for more than a decade." [03:18, 12:08]

Historical and Economic Foundation

• Exponential Growth: Azeem estimates computational capacity has increased by 11 orders of magnitude (factor of 100 billion) since 1972, compounding at 62% per year.
  • "The total stock of computing power... increased by 11 orders of magnitude, which is roughly a factor of 100 billion." [07:47]
  • "Every single year, we saw a 62% growth in the total computational stock on the planet." [08:57]
• Donuts vs. Compute Analogy: Unlike donuts (diminishing marginal utility), our appetite for compute grows as it becomes cheaper and more capable.
  • "There's a marginal, diminishing marginal utility with donuts... But that isn't true with computing." [05:44]

Transforming How We Work and Live

• The Computational Fabric: The economy increasingly runs on computational infrastructure, replacing tasks once done physically with digital equivalents (in silico).
  • "The economy is moving into a computational fabric alongside the physical elements of the real economy." [13:16]
• Agentic Workflows: AI shifts from query-response interactions to continuous agent workflows, further driving compute usage.
  • "In an agentic world, the tasks run the whole time because the agents run the whole time... the software robots in the data center continuing to work for us." [17:00]
• Corporate Adoption of AI Agents: Research from Wharton shows 80% of large US companies use genAI weekly; 58% use agentic workflows.
  • "Most interestingly, 58% of them said they were using some kind of agent or agent workflow in their business." [16:27]

Infinite Demand: Economic and Practical Perspective

• Jevons Paradox in Action: As compute becomes cheaper and more efficient due to optimizations, demand rises—new use cases become feasible.
  • "It's that positive elasticity of demand that I talked about, because as the outputs become cheaper, you widen up the range of use cases that were previously too expensive to attempt." [21:38]
• No Saturation Point: Compute does not reach a point of diminishing returns like physical goods—more compute always unlocks new insights, products, and industries.
  • "After four donuts you stop eating. But after 4 trillion trillion computations, you discover a new industry. And that means we're not yet full. And quite likely we'll never get full." [26:30]

Notable Quotes & Memorable Moments

• On the core theme:
  "What happens with computing is we have an insatiable demand for it, and it also has a positive price elasticity: if the price comes down, we buy more and more of it, we use more and more of it." [05:57]

• On infrastructure usage:
  "There are no dark GPUs anywhere." (Quoting Gavin Baker) [12:32]

• On tech investment skepticism:
  "Is this exuberance? Is this excitement running ahead of where we are? Are we building capacity that we will never use? ... It's not like the dark fiber during the telecoms bubble that wasn't used for more than a decade." [03:06, 12:08]

• On historical growth:
  "I estimate that the total stock of computing power available globally between 1972 and 2023 increased by 11 orders of magnitude." [07:47]

• On agentic AI:
  "In an agentic world, the tasks run the whole time because the agents run the whole time. They are working while we are sleeping." [17:00]

• On infinite demand:
  "What do I mean by infinite demand? I don't mean that every atom in the universe will be enlisted for computing. ... What I mean is practically unlimited... Better chips, more sensors, more data, improved algorithms, we find new things that we want to compute." [20:00]

• On compute vs. donuts:
  "After four donuts you stop eating. But after 4 trillion trillion computations, you discover a new industry." [26:30]

Timestamps for Key Segments

• 00:23 – Framing: The real story is computation itself, not just generative AI.
• 01:19 – Cloud business explosive growth statistics.
• 02:16 – Surge in chip demand; Anthropic and Google TPUs.
• 03:18 – Analogy to “Springfield monorail”; risk of overbuilding disputed.
• 05:44 – Donuts vs. compute analogy; positive price elasticity.
• 07:47 – Calculating historical compute stock since 1972.
• 12:32 – No idle capacity: “There are no dark GPUs anywhere.”
• 13:16 – The “computational fabric” of modern economies.
• 16:27 – Corporate AI usage stats: 58% of enterprises use AI agents.
• 17:00 – Agentic workflows and always-on compute demand.
• 21:38 – Jevons paradox: Optimization increases, not decreases, overall demand.
• 26:30 – Closing metaphor: Why compute is unlike donuts; no satiation point.

Conclusion

Azeem Azhar masterfully untangles the hype and anxiety around exponential increases in computational infrastructure—arguing that we are witnessing not a speculative bubble, but a deep, structural transformation. The demand for compute is not waning; it is constantly fueled by new capabilities, business models, and scientific frontiers that emerge whenever more compute becomes available. Rather than “eating our fill,” society only becomes hungrier for computation, promising continued growth, transformation, and innovation for decades to come.