
Feb 13, 2026 · 2h 22m
Anthropic CEO forecasts human-level AI within years as development costs soar
Dario Amodei — The highest-stakes financial model in history
The race to build AGI is no longer just a software challenge, but a massive geopolitical and financial wager on computing scale and safety.
- 1Artificial general intelligence could arrive within the next few years, functioning like a collective country of geniuses.
- 2Sustaining AI advancement requires massive computing budgets and a transition toward systems capable of continual learning.
- 3Accelerating medical breakthroughs will depend on AI navigating real-world feedback loops and clinical trials.
Don't miss
Dario Amodei defines AGI not as a single brain, but as a highly coordinated country of digital geniuses capable of solving humanity's toughest problems.
The brief
Anthropic CEO Dario Amodei envisions artificial general intelligence as a country of geniuses in a data center, potentially arriving within the next few years to radically reshape global productivity.
Achieving this leap requires massive capital, turning AI development into the highest-stakes financial model in history. Labs must balance immense computing costs against the promise of exponential economic growth.
To truly revolutionize fields like medicine, AI must move beyond static training data. Amodei argues that systems need continual learning and real-world feedback loops to solve complex scientific challenges.
As capabilities accelerate, the industry faces a delicate tension between rapid deployment and safety regulations, forcing developers to build robust guardrails alongside powerful models.
What was said on this episode
87 statements · 57 positive · 16 negative · 4 mixed · 10 neutral
AI capability progress has broadly matched Dario Amodei’s expectations.
“the underlying technology, like the exponential of the technology has gone broadly speaking, I would say about as I expected it to go”
Listen at 0:10
AI progress is nearing the end of its current exponential phase.
“The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential.”
Listen at 0:53
Scaling outcomes depend less on clever techniques than on a few broad factors.
“all the cleverness, all the techniques, all the kind of we need a new method to do something like that doesn't very much”
Listen at 2:58
AI progress remains consistent with the Big Blob of Compute hypothesis.
“I don't think I've seen very much that is not in line with that hypothesis. And so the pre trained scaling laws were one example of kind of what we see there.”
Listen at 4:15
Reinforcement learning exhibits scaling similar to pretraining.
“And so we're seeing the same scaling in RL that we saw for pre training.”
Listen at 5:21
Reinforcement learning exhibits scaling similar to pretraining.
“we're seeing the same scaling in RL that we saw for pre training”
Listen at 5:21
The sample-efficiency puzzle may not materially matter for AI progress.
“In fact, I would guess it probably doesn't matter.”
Listen at 6:31
Broader RL training will increasingly produce generalization.
“And then I think we're going to increasingly get generalization so that that kind of takes out the RL versus the pre training side of it.”
Listen at 8:31
Long-context models can learn and adapt effectively within context.
“But if we give them a context length of a million, they're very good at learning and adapting within that context length.”
Listen at 9:13
Broad RL data is intended to produce generalization rather than teach isolated skills.
“the goal is very similar to what was done five or 10 years ago with pre training with we're trying to get a whole bunch of data not because we want to cover a specific document or a specific skill, but because we want to generalize”
Listen at 12:17
A country of geniuses in a data center will emerge within ten years.
“On the basic hypothesis of, as you put it, within 10 years we'll get to, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that.”
Listen at 13:51
AI will perform end-to-end coding within one or two years.
“I mean, I think we'll be there in one or two years. There's no way we will not be there in 10 years in terms of being able to do it end to end coding.”
Listen at 14:54
AI systems will perform end-to-end coding within one or two years and certainly within ten.
“I think we'll be there in one or two years. There's no way we will not be there in 10 years in terms of being able to do it end to end coding.”
Listen at 14:54
Unverifiable tasks remain a source of uncertainty for advanced AI capabilities.
“My one little bit, the one little bit of fundamental uncertainty, even on long timescales is this thing about tasks that aren't verifiable”
Listen at 15:00
AI has a reliable path to non-verifiable expert tasks.
“I am almost certain that we have a reliable path to get there.”
Listen at 15:23
AI may master verifiable tasks without fully mastering non-verifiable tasks.
“The world in which we don't make it or the world in which we don't get there is the world in which we do all the things that are verifiable.”
Listen at 16:23
Anthropic has experienced approximately tenfold annual revenue growth.
“within anthropic, there's this bizarre 10x per year growth in revenue that we've seen”
Listen at 21:27
AI capability growth and economic diffusion can both follow fast exponentials.
“And so I think everything we've seen so far is compatible with the idea that there's one fast exponential, that's the capability of the model, and then there's another fast exponential that's downstream of that, which is the diffusion of the model into the economy.”
Listen at 23:09
AI capability growth and economic diffusion can each follow fast exponential trajectories.
“there's one fast exponential, that's the capability of the model, and then there's another fast exponential that's downstream of that, which is the diffusion of the model into the economy”
Listen at 23:16
AI will diffuse faster than previous technologies, but economic adoption will not be instantaneous.
“I think AI will diffuse much faster than previous technologies have, but not infinitely fast”
Listen at 24:58
Even AGI will not be adopted infinitely quickly.
“And I don't think even AGI or powerful AI or country of geniuses in the data center will be an infinitely compelling product.”
Listen at 27:15
Powerful AI could sustain 3–10x annual growth at hundreds of billions in revenue.
“It will be a compelling product, enough maybe to get 3 or 5 or 10x a year growth, even when you're in the hundreds of billions of dollars, which is extremely hard to do and has never been done in history before, but not infinitely fast.”
Listen at 27:22
The world does not currently have a country of geniuses in a data center.
“We don't have that now. That is very clear.”
Listen at 28:10
A country of geniuses in a data center will perform professional video editing.
“I think the country of geniuses in a data center will be able to do that.”
Listen at 31:31
Computer-use AI must reach a reliability threshold before deployment.
“And you know, there may be harder measures as well, but I think computer use has to pass a point of reliability.”
Listen at 32:39
Reliable computer use is necessary for broad AI deployment.
“I think computer use has to pass a point of reliability.”
Listen at 32:42
Coding models currently provide roughly 15–20% total-factor productivity improvement.
“I would say right now the coding models give maybe, I don't know, a, like 15, maybe 20% total factor speed up.”
Listen at 37:44
Coding models currently provide roughly 15–20% total-factor productivity improvement.
“the coding models give maybe, I don't know, a, like 15, maybe 20% total factor speed up”
Listen at 37:47
AI coding productivity gains are increasing progressively over time.
“And so we're seeing this snowball gather momentum where it's like 10%, 20%, 25%, 40%.”
Listen at 39:03
Pretraining generalization and in-context learning may suffice for AGI-like capability.
“I think these two things within the existing paradigm may just be enough to get you the country of geniuses in the data center.”
Listen at 41:57
Current AI learning methods could generate trillions of dollars in revenue.
“I certainly think just as things are, this, I believe, is enough to generate trillions of dollars of revenue.”
Listen at 42:08
Existing pretraining and in-context learning could generate trillions in annual revenue without continual learning.
“this, I believe, is enough to generate trillions of dollars of revenue”
Listen at 42:11
AI systems may achieve continual learning within one or two years.
“And I think there's a good chance that in the next year or two we also solve that.”
Listen at 42:26
Continual learning may be solved within the next one or two years.
“I think there's a good chance that in the next year or two we also solve that”
Listen at 42:26
Longer AI context windows are technically feasible.
“There's nothing preventing longer contexts from working.”
Listen at 43:09
AI will match a six-month human collaborator within ten years.
“I have a strong view, 99, 95% that like all this will happen in 10 years.”
Listen at 45:51
AI may match a six-month human collaborator within one to three years.
“And then I have a hunch this is more like a 50, 50 thing, that it's going to be more like one to two, maybe more like one to three.”
Listen at 45:59
Country-of-geniuses AI systems could exist within one to two years.
“I really do believe that we could have models that are a country of geniuses. Country of geniuses in a data center in one to two years.”
Listen at 47:53
Trillion-dollar AI revenue could take as long as five years after powerful AI.
“I could even stretch it to five years, although I'm skeptical of that.”
Listen at 48:24
Powerful AI may arrive in 2026 or 2027.
“I said, we'll get that 2026, maybe 2027.”
Listen at 48:59
Buying excessive compute could bankrupt an AI company if revenue falls short.
“And if my revenue is not a trillion dollars, if it's even 800 billion, there's no force on earth, there's no hedge on earth that could stop me from going bankrupt if I buy that much compute.”
Listen at 51:48
AI will accelerate clinical trials without making them instantaneous.
“I say the clinical trials are going to go much faster than we're used to, but not instant, not infinitely fast.”
Listen at 54:27
“Next year, you know, it goes up by roughly 3x a year. So like next year's 30 or 40 gigawatts and 2028 might be 1002029 might be like 3, 300 gigawatts.”
Listen at 56:44
Frontier AI profitability depends on forecasting inference demand accurately.
“I actually think profitability happens when you underestimated the amount of demand you were going to get. And loss happens when you overestimated the amount of demand you were going to get because you're buying the data centers ahead of time.”
Listen at 59:36
Accurate annual demand forecasts would make AI labs profitable.
“If every year we predict exactly what the demand is going to be, we'll be profitable every year.”
Listen at 1:02:44
Log-linear returns imply AI labs should allocate a moderate share to training.
“What I'm saying is the log linear return, what it leads to is you spend of order one fraction of the business, right? Like not 5%, not 95%.”
Listen at 1:04:12
AI-generated revenue will likely reach trillions of dollars before 2030.
“it is hard for me to see that there won't be trillions of dollars in revenue before 2030”
Listen at 1:07:15
A concentrated AI industry will retain positive profit margins.
“The point is it doesn't equilibrate to perfect competition with zero margins.”
Listen at 1:09:05
AI could raise economic growth to 10–20% annually, not 300%.
“I think we may get 10 or 20% per year growth in the economy, but we're not going to get 300% growth in the economy.”
Listen at 1:12:05
The AI industry is unlikely to become a monopoly.
“I don't think this field's going to be a monopoly.”
Listen at 1:13:11
Frontier AI will likely have three or four major firms.
“I think that's the same for AI. Three, maybe four.”
Listen at 1:13:52
AI models building successors could accelerate the whole economy together.
“I think once the AI models are building the next AI models and building everything else, the whole economy will kind of go at the same pace.”
Listen at 1:17:13
AI-driven growth may be geographically concentrated.
“I am worried geographically though.”
Listen at 1:17:22
Robotics will be revolutionized once AI models acquire relevant control skills.
“But I do think when, for whatever reason the models have those skills, then robotics will be revolutionized.”
Listen at 1:19:10
Once AI models acquire relevant skills, robotics will be revolutionized.
“when, for whatever reason the models have those skills, then robotics will be revolutionized”
Listen at 1:19:11
Robotics transformation may follow powerful AI by another one or two years.
“So will robotics be revolutionized? Yeah, maybe tack on another year or two.”
Listen at 1:19:43
Robotics may be revolutionized one or two years after advanced AI capabilities arrive.
“Will robotics be revolutionized? Yeah, maybe tack on another year or two.”
Listen at 1:19:44
AI may perform end-to-end software engineering within one or two years.
“I think we may get to the point in like a year or two where the models can just do sui end.”
Listen at 1:21:46
API-based AI business models will remain durable.
“I actually do think that the API model is more durable than many people think.”
Listen at 1:23:54
AI APIs will persist because developers need flexible experimentation access.
“But we're always going to have the API business model because there's always going to be a need for a thousand different people to try experimenting with the model in a different way.”
Listen at 1:25:27
API-based AI businesses will persist alongside other business models.
“I basically think it's always going to exist at the same time. I'm sure there's going to be other models as well.”
Listen at 1:25:48
Some AI outputs could create tens of millions of dollars in value.
“Those tokens could be worth tens of millions of dollars.”
Listen at 1:26:46
AI services will eventually adopt outcome-based payment models.
“at some point we're going to see pay for results in some form”
Listen at 1:26:56
Powerful AI requires a governance architecture balancing human freedom and AI oversight.
“in the long run we need some architecture of governance”
Listen at 1:33:27
Tennessee’s emotional-support AI law is poorly designed.
“I think that particular law is, is dumb.”
Listen at 1:37:28
Tennessee’s AI emotional-support law is misguided.
“that particular law is, is dumb”
Listen at 1:37:29
A ten-year moratorium on state AI regulation would be dangerously long.
“Given the serious dangers that I lay out in adolescence of technology around things like the kind of biological weapons and bioterrorism autonomy risk and the timelines we've been talking about, 10 years is an eternity, I think that's a crazy thing to do.”
Listen at 1:38:17
Federal AI standards should preempt conflicting state rules rather than suspend state regulation entirely.
“the federal government should step in, not saying, states, you can't regulate, but here's what we're going to do”
Listen at 1:38:55
Federal AI standards should preempt conflicting state regulations.
“I think preemption is fine in the sense of saying that federal government says, here's our standard. This applies to everyone. States can't do something different.”
Listen at 1:39:06
Health-related AI benefits should face less regulation.
“I would deregulate a lot of the stuff around the health benefits of AI”
Listen at 1:42:22
AI-accelerated drug discovery could overwhelm the drug-approval pipeline.
“I actually worry more about the drug approval process where I think AI models are going to greatly accelerate the rate at which we discover drugs. And just the pipeline will get jammed up.”
Listen at 1:42:34
AI drug discovery will accelerate faster than existing regulatory pipelines can process.
“AI models are going to greatly accelerate the rate at which we discover drugs. And just the pipeline will get jammed up”
Listen at 1:42:38
Policymakers should oppose individual chatbot restriction bills.
“I do think we should fight against these chatbot bills that you're describing described individually. I'm against them. I think they're stupid.”
Listen at 1:46:39
US-China AI parity could create unstable deterrence.
“We could also have a world where it's unstable, like the nuclear equilibrium is stable.”
Listen at 1:48:13
Powerful AI could entrench authoritarian governments.
“My worry is if the world gets carved up into two pieces, one of those two pieces could be, be authoritarian or totalitarian in a way that's very difficult to displace.”
Listen at 1:49:30
AI’s underlying technological progress will continue exponentially beyond current advanced capabilities.
“the exponential of the underlying technology will continue as it has before”
Listen at 1:51:42
AI may create a temporary national-security advantage for an early country or coalition.
“I think there will be either a critical moment, a small number of critical moments, or some critical window where it's like AI is. AI confers some large advantage from the perspective of national security and one country or coalition has reached it before others.”
Listen at 1:53:11
Post-AI negotiations should preserve strong leverage for liberal democracies.
“And I think my interest is in making that negotiation be one in which classical liberal democracy has a strong hand.”
Listen at 1:54:07
Powerful AI could make dictatorships morally obsolete.
“I actually believe it could be the case is that dictatorships become morally obsolete.”
Listen at 1:58:45
AI-derived medical cures could be shared without sharing advanced data centers.
“You could imagine a world where we produce all these cures to diseases and the cures to diseases are fine to sell to authoritarian countries. The data centers just aren't.”
Listen at 2:00:27
AI benefits, wealth, and political freedom will be difficult to distribute.
“What will not come easily is distribution of benefits, distribution of wealth, political freedom.”
Listen at 2:03:32
AI data centers should be built in Africa, provided they are not Chinese-owned.
“there's no reason we shouldn't build data centers in Africa”
Listen at 2:05:00
Developing countries should build AI-driven pharmaceutical industries.
“We should also build. There's no reason we can't build a pharmaceutical industry that's AI driven.”
Listen at 2:05:14
Principle-based AI training produces more consistent behavior and generalization.
“It's kind of purely a practical and empirical thing that we've observed that by teaching the model principles, getting it to learn from principles, its behavior is more consistent, it's easier to cover edge cases, and the model is more likely to do what people want it to do.”
Listen at 2:06:43
Principle-based training makes AI behavior more consistent and generalizable than rule lists.
“by teaching the model principles, getting it to learn from principles, its behavior is more consistent, it's easier to cover edge cases”
Listen at 2:06:48
AI companies will develop differing model constitutions.
“Different companies will have different constitutions.”
Listen at 2:10:15
The AI exponential was not historically inevitable.
“That wasn't inevitable.”
Listen at 2:14:33
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Books & mentions
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.