
Nov 19, 2026 · 58 min
Levie bets AI demand will outrun Anthropic’s risks
Why The Anthropic IPO Will Succeed — With Aaron Levie
The debate tests whether expanding enterprise workloads can justify frontier AI’s valuations and infrastructure commitments before adoption catches up.
- 1Coding could spread AI demand across legal services, manufacturing, cybersecurity, life sciences, sales, and other knowledge work.
- 2Cheaper models may expand viable workloads while frontier systems retain value for orchestration, judgment, reasoning, and tool use.
- 3Anthropic’s thesis still faces reliability limits, diminishing returns, high infrastructure commitments, and a potentially dangerous adoption mismatch.
Don't miss
Levie argues that agent swarms could let one person commission work equivalent to thousands of knowledge workers, radically expanding demand for compute and tokens.
The brief
Box CEO Aaron Levie makes the bullish case for Anthropic’s IPO: frontier AI remains early, token consumption is rising, and only a small group of companies operates at the frontier.
Coding is the leading use case, but Levie sees it as a gateway to legal work, manufacturing, cybersecurity, life sciences, sales, and other large commercial domains.
The most expansive claim is that agent swarms could let one person commission work equivalent to thousands of knowledge workers, creating a new demand curve for compute and tokens.
Alex presses the bear case: unreliable outputs, expensive infrastructure, valuation, diminishing training returns, and a duration mismatch if bills arrive before slower industries adopt AI.
Levie’s answer is that cheaper models broaden viable workloads while frontier systems remain valuable for reasoning, orchestration, judgment, and tool use; Box Architect shows how that thesis reaches enterprise workflows.
What was said on this episode
40 statements · 35 positive · 2 negative · 1 mixed · 2 neutral
AI adoption remains in its early diffusion phase.
“I think we are still in the very early part of the kind of inflection point of AI adoption and diffusion.”
Listen at 3:38
Frontier-AI businesses will grow substantially as adoption diffuses.
“these businesses will... only be much larger from at least a top line revenue standpoint.”
Listen at 4:01
Anthropic will not achieve another twelvefold growth within twelve months.
“I don't think anybody should think that there's another 12X in the next 12 months.”
Listen at 5:05
Coding is currently the dominant applied-AI use case.
“coding is sort of the dominant. sort of use case and applied AI use case right now.”
Listen at 5:18
Computer-using agents make many knowledge-work domains code-adjacent.
“everything is becoming sort of code adjacent”
Listen at 6:04
Economy-wide AI adoption will take roughly five to ten years.
“That is just a multi-year kind of like, you know, half a decade, decade long kind of journey that you have to go through for technology adoption.”
Listen at 7:10
Frontier-token spending could increase tenfold.
“at least at the frontier, easily we could be spending 10 times more on frontier tokens.”
Listen at 7:22
Economy-wide token usage could increase one hundredfold.
“overall tokens in the economy, we could be, you know, we could be using a hundred times more”
Listen at 7:28
Broad enterprise AI spending demonstrates meaningful utility.
“when you have, you know, in the tens of thousands of companies spending six or seven figures on AI, like it's working.”
Listen at 8:17
AI tokens will generate physical-product designs and manufacturing artifacts.
“every single one of those things we will be, will be generated by tokens in the future.”
Listen at 10:35
Cybersecurity token usage could grow roughly one hundredfold.
“that's all of the cyber domain that now has, again, probably a hundred X token growth from here that's associated with it.”
Listen at 12:12
Sales agents could improve sales-process effectiveness by five to twenty percent.
“you'll be five or 10 or 20% more effective in your sales process because of what agents are going and doing to kind of prepare you for that.”
Listen at 13:35
AI could enable roughly one thousand Pixar-quality movies annually.
“You're going to have like a thousand movies a year that are all of Pixar quality today.”
Listen at 15:06
Agentic AI will become pervasive within approximately three years.
“It will be like totally pervasive, you know, in like three years from now.”
Listen at 15:57
AI-system usage will become effectively uncapped as compute replaces human-time limits.
“That's why you are going to see basically uncapped level of usage of these systems”
Listen at 18:28
AI labs will eventually solve niche tasks such as podcast-breath removal.
“as versions of that, you know, become more valuable to the AI, you know, labs, I think it will solve those kinds of problems.”
Listen at 21:28
AI models may match programming-language performance on precise CAD and geometry tasks within years.
“we're probably just a couple years away from the models, you know, being able to do that as well as they can do, you know, javascript or python”
Listen at 22:07
AI systems will continue improving at computer use.
“these things will just get better and better at computer use.”
Listen at 23:48
Frontier-model inference is currently very profitable.
“inference, very profitable”
Listen at 25:12
Frontier-model businesses can generate cash and appear financially attractive.
“these businesses actually look pretty attractive. They generate cash. it works.”
Listen at 25:18
Frontier AI remains financially attractive despite commoditization risk.
“the frontier still remains very attractive at the moment.”
Listen at 25:33
Demand is rising across frontier and open-weight AI models.
“the data is showing that kind of everything is rising.”
Listen at 26:40
Cheaper models expand economically viable AI automation.
“by making the kind of efficient models affordable. And you can now actually successfully automate that task.”
Listen at 27:50
Frontier models remain valuable for reasoning, judgment, and orchestration.
“frontier model for decision making, reasoning, like the orchestration”
Listen at 29:39
Google’s Gemini models are improving substantially.
“the models are getting way better from Gemini.”
Listen at 31:11
Frontier models can design software experiences beyond conventional product-manager capability.
“It'll come back and it'll absolutely do something that no product manager could ever have done”
Listen at 33:11
Frontier AI can conduct effectively unlimited research for product design.
“you do have effectively a super intelligence now that just will go and do an infinite set of research”
Listen at 33:53
Anthropic’s infrastructure commitments are manageable if revenue reaches hundreds of billions annually.
“If you believe that the thing is going to generate hundreds of billions of dollars of, you know, a year in revenue, then those commitments are fine.”
Listen at 41:40
Token maxing is not the main reason for Anthropic’s revenue scale.
“I don't think that is like, like the core thing that is responsible for why Anthropic is, you know, an 80 plus billion dollar, you know, kind of revenue run rate company.”
Listen at 42:22
Billions in domain-specific post-training could materially improve models across commercial fields.
“you could spend billions of dollars on getting all of this RL data from every single one of those spaces. And I think the models would improve just meaningfully on every one of those dimensions”
Listen at 44:59
Agentic cybersecurity software could make the cybersecurity category two to three times larger.
“there's just no reason why that category isn't. two to three times larger as an end result of just the software part of that category.”
Listen at 46:53
AI cybersecurity inference could generate hundreds of billions in revenue.
“you could easily, you know, sort of underwrite that as hundreds of billions of dollars in revenue just in cyber inference.”
Listen at 47:12
AI could capture ten to twenty percent of major work markets within five to ten years.
“I think it's not unreasonable to think that would happen.”
Listen at 47:38
AI infrastructure funding could run out before expected returns materialize.
“It is absolutely possible that that could occur.”
Listen at 48:55
AI compute capacity is not yet substantially underutilized.
“I'm not yet seeing the part where we have so much compute capacity that's underutilized.”
Listen at 50:02
AI demand is currently increasing across Levie’s observed market.
“we're only seeing things that are up and to the right in terms of the overall demand for AI.”
Listen at 50:28
Box Architect will eventually run and optimize enterprise workflows autonomously.
“eventually we'll kind of run and optimize those solutions for you as well.”
Listen at 54:49
Box Architect provides autonomous management and automation of enterprise-content workflows.
“you can effectively have an autonomous platform that manages, automates, administers the workflow around enterprise content.”
Listen at 54:54
Box Architect can build contract-management systems that autonomously improve.
“it can design a contract management system for you that will now effectively autonomously improve in the future.”
Listen at 55:22
Aaron Levie remains satisfied with holding SpaceX shares.
“I still hold them. I'm happy about it.”
Listen at 55:55
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.
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.