a16z Podcast: "The 2045 Superintelligence Timeline: Epoch AI’s Data-Driven Forecast"

Date: November 24, 2025

Episode Overview

This episode features a deep-dive conversation with David Owen and Jaffa Edelman from Epoch AI, discussing their rigorously data-driven approach to forecasting the trajectory of artificial intelligence through 2045. The discussion covers the economic realities of AI scaling, possible world-altering milestones on the road to superintelligence, challenges and breakthroughs in data center infrastructure, imminent disruptions in the labor market, and speculative but measured predictions around AI’s influence on the global economy and politics. Throughout, the conversation tests both alarmist and utopian narratives against hard evidence and uncertainty, resulting in a nuanced vision of what the next two decades may hold.

Key Discussion Points & Insights

1. Are We in an AI Bubble?

Timestamps: 00:00 – 06:30, 03:38 – 06:30

• David Owen and Jaffa Edelman challenge the notion of an AI investment bubble.
  • Owen: “People are spending a lot on these models. They're presumably doing this because they're getting value from them. ... I don't think it's a bubble because it's not burst yet.” (00:00, 05:29)
  • Profits from inference are already materializing; if development stopped today, margins would quickly pay off sunk R&D costs.
  • “A lot of people worrying about bubbles just aren't necessarily used to the level of spending and...success that sort of happened in, like, scaling. But if there is a bubble, it could happen very suddenly and be pretty bad.” — Jaffa Edelman (04:40)

Memorable Quote:

"I don't think it's a bubble because it's not burst yet. When it's burst yet, then you'll know it's a bubble." — David Owen (05:29)

2. AI Scaling, Plateau, and Self-Improvement

Timestamps: 06:30 – 12:47

• No clear evidence yet that model pre-training or capabilities are plateauing: innovation is still ongoing in both pre-training and post-training techniques.
• The much-hyped "software-only singularity" (an AI recursively improving AI) is not reflected in current industry patterns:
  • Current R&D still heavily relies on costly experimental compute, not just researcher labor.
  • “We don't really have good evidence that researchers and just researchers would be able to speed things up without doing more experiments.” — Jaffa Edelman (09:11)
• Analogies to human learning styles (imitation vs RL) are seen as speculative:
  • “As we've scaled up, we have managed to do quite well at like, having models that remember more and more things.” — David Owen (12:13)

3. AI & Automation: Hype vs Reality in Coding

Timestamps: 13:09 – 17:24

• While AI is producing a growing share of code, the much-cited prediction that “AI will write 90% of code” remains ambiguous.
  • “Far more than 90% of the code I write is written by AI these days, but I know I'm not like the average coder at all.... Probably not 90%, but it's hard to tell.” — Jaffa Edelman (15:34)
  • The uplift in productivity is real but hard to quantify; a lot of AI-written code may be for new tasks that wouldn't have existed otherwise.
  • Financial metrics, rather than code volume or replaced headcount, will best measure AI’s true impact.

4. Labor Market Impacts and Economic Disruption

Timestamps: 18:08 – 22:59

• AI’s labor threat is nuanced: many jobs might be partially automated (at the “task” level), and new jobs could emerge.
  • “It just does not seem at all implausible to me that [AI could] automate all of...existing jobs, with the exceptions of ones that require manual labor, that people actually care about being done by a human.” — Jaffa Edelman (18:36)
• Watch for sharp, AI-driven unemployment raises as major catalysts for public response.
  • “A 5% increase in unemployment over a very short period of time, like six months due to AI, is something that I think will have a very substantial impact on the world.” — Jaffa Edelman (19:20)
• Despite expected automation of ~10% of current jobs over the next decade, overall unemployment may not move drastically due to market adaptation and job churn.

5. Education & Career Advice in an AI World

Timestamps: 24:07 – 26:38

• Traditional advice stands: pursue what you’re passionate about, but acquire general, flexible skills (teamwork, communication).
  • “The skills that are going to be useful are not...knowing a programming language. It's going to be more general purpose skills...” — Jaffa Edelman (24:26)
  • “If you have a better time in college, that's like four years of your life, you've had a better time during.” — Jaffa Edelman (25:44)
• “Prompt engineer” as a career is already fading as LLMs improve; adaptability is key.

6. Frontiers: Computer Use, Benchmarks & Productivity

Timestamps: 26:38 – 38:23

• AI agents are now tangibly useful in automating digital tasks, but have not yet triggered the disruptive “Sonnet 3.5 moment” for computer use.
  • Vision-motor limitations and long-context challenges remain bottlenecks but are actively being attacked.
  • “This was the first year I found computer use actually useful....That’s because of improvements we’ve seen in computer use over the past year or so.” — Jaffa Edelman (28:32)
• Productivity and GDP growth: if AI continues its current trend, expect “a percent kind of GDP increase” by 2030 without full AGI.
  • If AI achieves full remote labor automation, "30% GDP growth seems like a lower bound on something that's reasonable." — Jaffa Edelman (32:46)

7. Benchmarks & Measuring Progress

Timestamps: 35:57 – 38:23

• Main near-term AI benchmarks will be completed soon; future benchmarks must get harder, more realistic, and more resource-intensive.
  • Watch for “relatively small…very impressive” new capabilities: “When you see arms being able to do things like, oh, yeah, I just refactored this entire code base…that’s going to be useful.” — David Owen (36:41)

8. Timelines for Major Milestones

Timestamps: 38:23 – 47:35

• Solving a Major Math Problem (e.g., the Riemann Hypothesis):
  • “I would not be surprised if AI solves a major unsolved math problem...in the next five years. ... Math seems unusually easy for AI.” — Jaffa Edelman (39:28)
  • Caution, however: society may retroactively discount milestones once reached, as happened with chess.
• Breakthroughs in Biology/Medicine:
  • AlphaFold is an early indication, but true autonomous discovery in biology is harder: “You need to be able to actually do experiments and get data and interact with the real world…for a lot of [biology problems] in a way that does not need to happen…for math.” — Jaffa Edelman (42:52)
• Superintelligence:
  • Peak probability around 2045 for “everything breaking down and going bananas” (superintelligence).— Jaffa Edelman (45:34)
  • Once AIs can do every remote job, exponential scaling to superintelligence will be close behind.

9. Robotics: Hardware Bottlenecks and Scaling

Timestamps: 47:35 – 50:38

• Progress in robotics has lagged due to the relative lack of scaled-up compute and the economics of hardware (robots remain expensive).
  • “I think of robotics as mostly a hardware problem ... it's just not obvious to me that there is a software problem here.” — Jaffa Edelman (48:13)
• Many robotics challenges may be solved via scaled imitation learning if investment follows.

10. Data Center Infrastructure: The Hidden Heart of AI

Timestamps: 50:49 – 55:42

• Epoch AI’s new “Dentist project” reports on the scale and physical realities of the ongoing AI data center build-out.
  • Satellite and permitting data show sites rivaling cities in electricity use coming online within years.
  • “We learned that the most likely candidate to have the first gigawatt scale data center is Anthropic, which would not have been my pick...” — Jaffa Edelman (50:56)
• Scalability is chiefly limited by investment and willingness to pay—not by insurmountable engineering or energy bottlenecks.
  • “We're scaling up as fast as there is money to scale up approximately. ... All of these things pale in comparison to the cost of your GPUs.” — Jaffa Edelman (52:42)
  • Energy limitations can be overcome at increased, but still manageable, cost.

11. Political Response & Policy Shockwaves

Timestamps: 55:42 – 58:53

• Political system’s response will be sharp but lags until major labor or economic disruptions visibly materialize.
  • Rapid, consensus-driven action is possible, as seen with the COVID-19 stimulus: “I think it's like everything else in AI, it's exponential, which means it will pass the point of people sort of care about it to people really care about it. Quite fast.” — Jaffa Edelman (56:12)
  • Nationalization, unemployment benefits, pausing development—previously unthinkable actions may suddenly become consensus.
  • Attention by policymakers is doubling/tripling yearly, in line with AI revenue curves.

Notable Quotes & Timestamps

• “People are spending a lot on these models. They're presumably doing this because they're getting value from them... You can look at Nvidia and how much they're selling each year, and you can see whether it keeps on growing.” —David Owen (00:00, 03:49)
• “If there is a bubble, it could happen very suddenly and be pretty bad.” —Jaffa Edelman (04:40)
• “The most reliable indicator here is going to be how much money these people are making from programmers and from subscriptions in general. And it's a lot of money.” —Jaffa Edelman (17:24)
• “I would not be surprised if AI solves a major unsolved math problem like the Riemann Hypothesis or similar in the next five years.” —Jaffa Edelman (39:28)
• “Assuming in the next 10 years we get AI that is capable of doing any remote job as well as any human...30% GDP growth seems like a lower bound on something that's reasonable.” —Jaffa Edelman (32:46)
• “If you have AI that can do [anything humans can do]...then it doesn't lead to crazy things. ...the default there is you either go crazy up or crazy down.” —Jaffa Edelman (34:36)
• “I expect...there will be some sort of strong [political] response, and it's going to happen very fast.” —Jaffa Edelman (56:12)

Section Timestamps (Quick Reference)

• Bubble Discussion: 00:00–06:30
• Scaling & Self-Improvement: 06:30–12:47
• Coding Automation: 13:09–17:24
• Labor Market: 18:08–22:59
• Education/Careers: 24:07–26:38
• Benchmarks/Computer Use: 26:38–38:23
• Timelines (Math, Biology, Superintelligence): 38:23–47:35
• Robotics: 47:35–50:38
• Data Centers, Scaling: 50:49–55:42
• Political Response: 55:42–58:53

Overall Tone and Takeaways

Epoch AI’s analysis is frank, evidence-oriented, and often counter-intuitive—pushing back on both doomerism and hype. The future of AI, as described, is less about sudden, godlike recursive improvement and more about relentless, exponential scaling—with very real, deeply disruptive implications for economies, labor, infrastructure, and politics. Key uncertainties abound, but the breakneck pace is undeniable: Whatever happens next, it’s likely to be stranger and faster than most anticipate.