Theme · Updated
What podcasters are saying about AI pricing & unit economics
Summary
In the past 30 days, 38 speakers on 25 finance podcasts discussed AI pricing & unit economics in 50 views. Chorus has followed AI pricing & unit economics since October 1, 2026.
Who talks about it most
- Steve Hou · 3
- George Noble · 3
- Jim Paulsen · 3
- Josh Brown · 2
- Jeetu Patel · 2
- Ethan Mollick · 2
- Citrini · 2
- Naveen Rao · 2
- Chris Whalen · 2
- Ed Elson · 1
The latest views
Newest first. Every quote is checked against the episode's transcript.
Josh Brown · Ritholtz Wealth Management
Per-token AI costs fall, but usage growth means total compute spending keeps rising, showing sustainable demand.
Josh agrees that per-token costs are plunging, as they have throughout the history of technology. But usage is growing so fast that total dollars spent on compute are not falling; they are rising. He argues this is a sign of how sustainable the AI demand boom is, because the services are so popular and in such demand that usage outpaces price declines.
“The thing is the usage is going up by so much that the overall dollar amount being spent on compute does not fall, it goes up.”at 32:31
Prof G Markets: Josh Brown: If This Is A Bubble, It’s Pathetic ·
Ed Elson · Prof G Media
Falling per-token spend hurts AI labs because their customers' revenue will slow.
Ed agrees that falling per-token compute costs are good for the labs, since they can get more compute for less. But he worries that per-token spend by the startups buying tokens or LLM usage is also falling, which is already happening. That would mean the labs' revenues start to slow, which is a problem for Anthropic and OpenAI. He says the labs are highly dependent on VC-funded startups.
“if that per token spend is also going down, which it is, then that is also gonna be a problem because it means that their revenues are gonna start to slow.”at 33:27
Prof G Markets: Josh Brown: If This Is A Bubble, It’s Pathetic ·
Jeetu Patel · Cisco
AI is being paid for: end-user revenue at Anthropic and cloud revenue at Azure and Google show it.
Patel answers the question of who pays for AI by pointing to disclosed revenue. Anthropic's revenue is end-user revenue that customers pay for, and Microsoft Azure and Google both report cloud revenue, which shows the infrastructure layer is also being paid for. He argues that once trust and security improve a notch or two, usage should rise sharply. The argument rests on the revenue evidence and an expected rise in usage once trust improves.
“If you look at just what [Anthropic's] revenues are, that's the end user revenue that's being paid for. And then if you look at what both Azure and Google [reported] as revenue on the cloud, that shows you that the infrastructure is also getting paid for.”at 6:29
Bloomberg Talks: Cisco Systems President & Chief Product Officer Jeetu Patel Talks AI Industry ·
Jeetu Patel · Cisco
AI monetisation is real: end-user and cloud revenues show someone is paying.
He answers who pays for AI by pointing to Anthropic's end-user revenue, which is being paid for directly. He adds that Azure and Google Cloud both report cloud revenue, which shows the infrastructure is also being paid for. Once trust and security improve, he expects a large jump in usage.
“If you look at just what Anthropic revenues are, that is the end user revenue that is being paid for. And then if you look at what both Azure and Google Reporter as revenue on the cloud, that shows you that the infrastructure is also getting paid for.”at 32:03
Bloomberg Surveillance: Outlook for Inflation, Oil, and Tech ·
Jack Altman · Benchmark
Frontier labs likely keep a per-inference cost edge over inference providers, though training amortization complicates it.
Frontier labs must recover training costs through their prices, while an open model is free to use, so it carries no amortization. Inference still has to be run either way. Jack thinks labs still have the advantage per unit of inference over inference companies, but that gap could be offset by other factors. He says he mostly buys Ev's economic case for closed labs, but he doesn't have the margin numbers to be sure.
“And it's probably right that on a per inference unit basis, the labs are still going to have the advantage over the inference companies.”at 32:46
Uncapped with Jack Altman: Benchmark Breakfast #1 | Jack and Ev ·
Steve Hou · Silicon Data
The falling token price index reflects cheaper models spreading as usage rises, not a deflationary race to the bottom.
The usage-weighted token index has trended down since late May. Steve argues the bears read the chart as deflation and commoditization, but the same data fit healthy AI proliferation. Within the index, the proprietary and open-model sub-indices have not fallen. The overall decline comes from substitution away from expensive proprietary models toward cheaper open models. Total token usage has risen sharply, and this is consistent with people using cheaper models for more tasks.
“You can tell a bearish story. I can also tell you that this is exactly what AI proliferation should look like, right? And prices of tokens coming down and people using the cheaper models to do a lot more stuff.”at 4:33
Monetary Matters: Why Compute Prices Are Still Rising | Steve Hou of Silicon Data on GPU Rental Indices, Fed Meeting, and Anthropic S-1 (LIVE) ·
Steve Hou · Silicon Data
Token serving margins are clearly positive, and newer chips widen them; durability of margins is the open question.
Hou says the chart being cited does not show token production costs above selling prices, since it uses dual axes with normalised figures. He argues newer chips are much better at producing tokens, so tokens are selling well above cost. He then says the real question is whether those margins persist. If the people earning large margins from running or owning GPUs are joined by more entrants, competition could compress the spread to zero, unless frontier models and existing operators have a durable moat. He calls that the trillion-dollar question.
“newer chips are so much better at producing tokens that you're certainly producing, you know, those tokens are certainly selling well above cost.”at 41:08
Monetary Matters: Why Compute Prices Are Still Rising | Steve Hou of Silicon Data on GPU Rental Indices, Fed Meeting, and Anthropic S-1 (LIVE) ·
Ethan Mollick · Wharton School, University of Pennsylvania
Rising token spend is a management problem, not evidence that AI is failing to pay off.
Mollick says token costs are a growing expense companies want to control. He argues this reflects a lack of decisions about who uses tokens, for which processes, and under what limits, so people use as many as they can when they are incentivized to. He dismisses the public spending pullbacks as coming mostly from firms with competing AI products (Microsoft, Meta, Palantir), whose revenues are still growing. He reads cutbacks as a sign of unmanaged use rather than uselessness.
“I think people are worried about token cost, but that's often a management problem. Like, if you don't make any decisions about how tokens should be used... they'll use as many tokens as possible.”at 10:43
Big Technology Podcast: This AI Expert Argues We're Underestimating AI's Productivity Impact — With Ethan Mollick ·
Ethan Mollick · Wharton School, University of Pennsylvania
Falling token prices are a deflationary push that will make token-cost worries fade, despite rising usage.
Mollick says he is not worried about long-run token costs. Usage is rising because agentic work consumes many more tokens, and Jevons' paradox applies. Organizations are also getting more token-efficient: a frontier model can orchestrate cheaper small models to do subtasks. The result is that AI does more work for less money, which he calls a really deflationary push. He expects the concern about token costs to fade the way concerns about vibe coding and RAG have faded, because costs keep falling even as more compute comes online.
“So that lower cost means AI does more stuff for less money in a very short period of time. Really, really deflationary push.”at 1:03:34
Big Technology Podcast: This AI Expert Argues We're Underestimating AI's Productivity Impact — With Ethan Mollick ·
Mark Cuban · Cost Plus Drugs
AI inference costs are high today but falling very quickly, making consumer AI subscriptions cheaper to serve.
Cuban says the inference and processing needed to respond to a request is expensive today but coming down very quickly. He expects the same to happen for personalized medicine. From this he argues that model companies' $20 or $200 consumer subscribers should add personalized health services once they are available, since the marginal cost of delivering them will be small.
“the inference and everything that has to happen just to respond to a request today. Expensive like this, right? Coming down very quickly, but the same thing's going to happen for this type of personalized medicine.”at 38:46
a16z Podcast: Mark Cuban’s Plan to Make Healthcare Affordable ·
Robert Kaplan · Goldman Sachs
Token and compute demand will ramp, but prices for certain token types will fall.
Kaplan expects a big continued ramp in token demand and compute demand. He expects prices for certain types of tokens to decline, and he sees that as a healthy natural development. He argues that firms should be judged by whether they are merchants selling compute, or by whether they use it to monetize their own customer revenue. Model makers should also be assessed on how dependent they are on compute pricing and how speculative that is.
“We think there's going to be a big ramp up continued in token demand and demand for compute, but the pricing for certain types of tokens will probably decline.”at 26:52
Bloomberg Surveillance: Bloomberg Surveillance TV: October 27th, 2026 ·
Cliff Sosin · CAS Investment Partners
The price of intelligence should fall about 100x over the next five years.
Sosin says the price of intelligence is falling by a factor of about 100 over the next five years. He treats that as a first-principles input: if intelligence gets that cheap, the cost base of companies that buy it, such as Meta with its roughly $250B of spending, largely goes away. He cites the broader trend of rapidly falling model costs as evidence. He describes the 100x figure as a rough, approximate estimate.
“the price of intelligence is falling by a factor of a 100 or something over the next five years, first principles,”at 10:37
Invest Like The Best: Cliff Sosin - Owning the Complements to Intelligence - [Invest Like the Best, EP.497] ·
Dan Nathan · RiskReversal Media
AI model price war will push costs toward zero; value will accrue to applications, not model labs.
Each new Anthropic model is marketed as cheaper, and Nathan says OpenAI and Anthropic are in a real price war. The hope is that cheaper tokens drive more usage, and so far the data suggests usage is rising enough to offset lower prices. He is not sure OpenAI and Anthropic will capture the value. He thinks value will be created in applications built on top of the models, and that businesses deploying AI are happy to see costs fall.
“As the cost of the models come down and they are in a price war, make no mistake about it, between OpenAI... I'm not so sure it's OpenAI and Anthropic.”at 13:33
CNBC Fast Money: Anthropic’s long-awaited IPO filing... And next Fed decision on deck 10/26/26 ·
Dan Niles · Niles Investment Management
Open-weight models are cutting AI average selling prices sharply, squeezing margins for closed-model providers.
Niles names open-weight models as a headwind alongside the slowdown in Anthropic's revenue and the data center moratorium news. Open-weight models have cut average selling prices by more than 50% since May, a concrete figure that feeds into how the AI trade's margins will evolve.
“You're struggling with that and the fact that open weight is in fact dropping average selling prices by over 50% since May.”at 16:15
CNBC Closing Bell: Closing Bell Overtime: Markets Face a Big Week as AI Takes Center Stage 10/26/26 ·
Mikkel Rosenvold · Steno Research (Real Vision)
Falling compute costs could accelerate AI adoption through the Jevons paradox.
Rosenvold argues that if compute prices drop, it is an issue for the providers, but it would trigger the Jevons paradox discussed last year. Cheaper compute could then speed up AI adoption, so falling costs would be net positive for the broader AI theme.
“If compute prices drop, that's obviously an issue for their providers, but that would only trigger the Jevons paradox that we talked a lot about last year. Cost of compute falls, that could even speed up the AI adoption.”at 14:10
Real Vision: Why AI Is NOT the Next Dot-Com Bubble with Andreas Steno and Mikkel Rosenvold | Macro Mondays ·
Andreas Steno · Steno Research
Quality-adjusted token prices have fallen about 95% over two years, so AI deflation is real.
Steno argues that official inflation measures adjust prices for quality. A token is far more valuable now than a year ago because models are much better, so people pay roughly the same nominal price for much more output. On that basis he estimates the price of a token is down about 95% over the past two years.
“One thing is the nominal price you pay for an iPhone or pay for a token, but it will always be adjusted by the power of that iPhone and the power of that token... I'd argue that the price of a token is down 95% or so over the past two years.”at 14:46
Real Vision: Why AI Is NOT the Next Dot-Com Bubble with Andreas Steno and Mikkel Rosenvold | Macro Mondays ·
Paul Kedrosky · SK Ventures
Frontier token prices won't fall to marginal cost; new models are a halo-pricing defence.
If AI is a scale business that consolidates, providers compete down to marginal cost, which is mostly power and cooling for inference. Frontier labs resist this by launching constant new models to justify a premium, a halo price above production cost. Pricing pressure is worse than it looks, because most mainstream users (not the software engineers who dominated the first four years) don't care about frontier performance and can't tell models apart in blind tests. China already sells tokens near marginal cost, roughly eight months behind US models, so the premium is hard to defend.
“And way to think about that is it's a defensive tactic to try and make sure that they can charge what the marketers call, you know, have a halo price above the cost of actually producing tokens.”at 23:19
Hidden Forces: The Most Dangerous Investment Cycle In History | Paul Kedrosky ·
Steve Hou · Silicon Data
Token prices falling while GPU rental rates rise, squeezing AI model economics.
Silicon Data's token index, the expenditure-weighted effective price paid for models, has fallen since about the end of May to roughly half its mid-year level, while GPU rental rates have risen rapidly and continuously. Usage is shifting to cheaper models that handle everyday tasks, so volume grows while per-token prices fall. Steve also questions how AI will be monetised: metered usage is unproven, Meta plans free or low-cost access, and distilled cheaper models may let others capture the value that frontier labs created.
“since about the end of May, the token index or the expenditure weighted effective price paid for the models has been coming down pretty significantly, almost, you know, half of where it was back in the middle of the year. In the meantime, you know, rental rates for these chips have been going up, you know, rapidly continuously.”at 9:36
Merryn Talks Money: Can the AI Boom Survive a Margin Squeeze? ·
Citrini
Cheaper tokens won't drive proportional usage growth near term, so Jevons-style demand is doubtful.
Citrini raises questions about Jevons paradox, which says cheaper resources raise total consumption, not really applying to tokens in the short term. He cites an Anthropic Economic Index finding, which he extrapolates, that a 10% reduction in the cost of a task correlates with only about a 3% increase in usage. Demand for compute depends on how quickly new demand arrives and how much computation each task requires. As enterprises build efficient AI systems that work well, they may need fewer inference calls, since earlier token demand was partly driven by fixing things that did not work. He calls these short-term concerns but says they need to be taken seriously given the buildout's financing.
“questions about Jevan's [Jevons] paradox, at least in the short term, not really applying to tokens.”at 11:59
Monetary Matters: Citrini: Monkeys Are The Next Bottleneck | Citrini on AI Drug Discovery Experiment & Reflexive Financing Risk to AI Buildout & Hardware Trade ·
Citrini
Token and AI-experiment costs are falling fast, roughly 20x to 25x cheaper in months, showing deflation.
Citrini's own drug-discovery experiment, with about 10,000 GPU hours, cost about $50,000 at the time. Redoing it now would cost roughly $2,000 to $4,000, a steep fall in months. A Vega lab release showed an 86% reduction in tokens needed to design a cortisol binder. Citrini calls this commoditization or deflation, which is already happening in AI pricing.
“and that cost about $50,000 with the drug discovery thing. By the way, crazy statistic. From the time we did that to the time we were able to publish because we had the data, the cost, like if we were to have done that now, it would have cost somewhere between 2,000 to $4,000”at 1:14:41
Monetary Matters: Citrini: Monkeys Are The Next Bottleneck | Citrini on AI Drug Discovery Experiment & Reflexive Financing Risk to AI Buildout & Hardware Trade ·
Naveen Rao · Unconventional AI
Inference providers' margins will shrink as the price per token falls, since they mainly map models onto hardware.
Inference providers such as Fireworks take an open model and make it reliable and cost-efficient on infrastructure. Their value is mapping the model onto physical hardware. Rao expects the price customers pay per token to fall. Providers will then have to squeeze more margin out of the hardware, and their margins will fall. He sees value in that layer in the future but a much smaller share of the profit.
“Now I think there will be value there in the future, but the price of inference, like what people are willing to pay per token, will drop.”at 40:39
20VC: 20VC: Why Robotics Will be 100x Bigger than Coding Agents | Inference Will Be a $10TRN Market: What No One Sees | Why Software Investors are F***** Today with Naveen Rao, Unconventional ·
Naveen Rao · Unconventional AI
The inference market becomes a $10 trillion market within ten years as token costs fall.
Rao says AI models alone will be a trillion-dollar market within a few years. Labor and value generation are shifting into technology, which creates very large addressable markets. He argues that cutting the cost of delivering a token by one order of magnitude effectively multiplies the market more than tenfold, which he calls Jevons paradox. Cheaper intelligence unlocks many new uses.
“And the inference market is going to be a $10,000,000,000,000 market within ten years.”at 52:14
20VC: 20VC: Why Robotics Will be 100x Bigger than Coding Agents | Inference Will Be a $10TRN Market: What No One Sees | Why Software Investors are F***** Today with Naveen Rao, Unconventional ·
Brad Conger · Hirtle Callaghan & Co.
Token delivery's 95% margin means OpenAI and Anthropic's API businesses are more challenged than thought.
Conger says DeepSeek released open-source infrastructure code, DeepInfer, showing that delivering tokens is a roughly 95% gross margin business. That shocked the market because, with a margin that high and no differentiation, the API business models of OpenAI and Anthropic look far more challenged than expected. Melissa Lee notes this leads to price wars down to marginal cost, favoring open-source or Chinese competitors.
“DeepSeek released its open source infrastructure code called DeepInfer, and the upshot of that was that delivering tokens is a 95% margin business. And I think that shocked the market because it means that the business models of the API links like OpenAI and Anthropic are much more challenged than we thought.”at 32:54
CNBC Closing Bell: Closing Bell Overtime: Markets Search for Opportunity After a Wild Week 10/23/26 ·
George Noble · Noble Capital Advisors
AI compute and token prices are collapsing while AI company balance sheets deteriorate.
George says the industry's costs are collapsing and compute and token pricing is falling at a breathtaking rate. Every week a model comes out that is half as expensive as the prior one. At the same time, AI companies' balance sheets are deteriorating just as fast, with debt rising. Falling prices combined with rising debt is, in his view, the core problem with the AI business.
“costs are collapsing and the pricing on compute and tokens is collapsing at a breathtaking rate.”at 4:24
Monetary Matters: "Coreweave's a Zero" | George Noble on Why He's Short Neoclouds, Why AI is a Shale 2.0 Bubble Headed for Debt/Depreciation Brick Wall, and Why He's "Coming For The Momentum Bros" ·
George Noble · Noble Capital Advisors
AI subscriptions are underpriced; at true unit cost, the $200 tier's addressable market is a fraction of $20's.
George says AI labs sell consumer subscriptions at about $20 a month, but the cost of compute and energy implies a price closer to $200. Few people would pay $200 a month, so the addressable market at the real price is a small fraction of the market at $20. He concludes that the revenue is being sold below cost, like 'anyone can sell dollar bills for 20 cents.' He also says marginal costs eventually have to be charged, as with ride-hailing and delivery apps after their IPOs.
“sources I've read, they're selling stuff for $20... And you do the unit economics, should be like $200.”at 15:24
Monetary Matters: "Coreweave's a Zero" | George Noble on Why He's Short Neoclouds, Why AI is a Shale 2.0 Bubble Headed for Debt/Depreciation Brick Wall, and Why He's "Coming For The Momentum Bros" ·
Andy Jassy · Amazon
AI inference is still an order of magnitude too expensive; cutting its cost is needed for pervasive enterprise use.
Inference costs have fallen considerably over the last few years but are still, in Jassy's view, an order of magnitude too high for AI to be used as pervasively as possible. Rewriting enterprise applications that run on CPUs to use AI accelerators is a big undertaking, so the cost gap has to be large enough to justify it. If inference cost drops, that bar falls and applications become better and more capable with AI accelerators and inference. He argues Amazon's chip business is a big part of driving that cost down.
“The cost of inference has come down quite a bit in the last few years, but it's still an order of magnitude too expensive in my opinion for it to be used as pervasively as possible.”at 32:39
a16z Podcast: Amazon CEO Andy Jassy on Why Amazon Is Investing $220 Billion in AI ·
Josh Brown · Ritholtz Wealth Management
AI competition is shifting from model quality to efficiency per dollar as cheaper models launch.
Brown cites OpenAI's lower-cost midrange GPT models launched in September and Anthropic's Haiku 5.5, which offers better performance at lower cost. He also notes IBM's CEO saying large enterprises are scrutinizing AI ROI and worrying that some AI spending has no real return. Brown predicts that efficiency per dollar spent, not raw model superiority, will define 2027 after three years of a model race.
“This is to me gonna be the story of 2027, efficiencies in AI per dollar spent. We spent so much for the last three years in a model race. Whose is better? It's about to tip over.”at 1:02:50
The Compound and Friends: An Economy So Good No One Can Live In It with Jill Schlesinger ·
Jill Schlesinger · CBS News
Buyers will pay for top AI only where needed; most users want good-enough, cheaper AI.
Schlesinger says there is a trend around when AI is good enough and for which users. Large companies that need specialised compute and software expertise will pay for the best. Many other buyers, such as a media company that only needs AI to run scripts and fact-check them, need far less. She compares this to a fintech firm's overbuilt systems, which customers do not need, and suggests companies may pull back after overselling capability.
“I think there's an interesting trend of when it is good enough and for which people... if you're a big media company and all you want is AI that can run scripts through the machine and fact check it.”at 1:03:15
The Compound and Friends: An Economy So Good No One Can Live In It with Jill Schlesinger ·
Jason Lemkin · SaaStr
AI product buyers are highly cost-sensitive, pushing routing toward cheaper models over top-tier intelligence.
Jason says Replit users are cost sensitive, and enterprise customers are too, because they do not want massive variable bills from product teams building prototypes. Replit has a model selector, and the routing choice shows that customers push to cheaper models when quality is similar. He argues this makes the market a super cost-sensitive model. Replit, with its large user base, is a useful lab for watching how quality and cost get optimized.
“But the enterprise customers are very cost sensitive because they don't want these massive variable bills for the product teams going off and building prototypes. It's a super cost sensitive model, right?”at 25:21
20VC: 20VC: Trillion or Bust: The New VC Game | Coreweave Sell-Off Creates Panic | Town at $1B, Anduril at $90B: Revenue Multiples Out of Control | General Atlantic Eyes IPO & Modal Raises $750M at $15.75B ·
Rory O'Driscoll · Scale Venture Partners
Model labs earn about 70% gross margins while inference hosts earn about 30%, shaping where developers spend.
Rory argues that if a team like Replit can buy tokens on Modal at a 30% gross margin rather than from Claude Code at a 70% margin, and the tokens are roughly the same, it will buy on Modal. He sees the inference market as expanding, but the big labs still take more than 95% of the revenue, since OpenAI and Anthropic together exceed $100B while open-weight inference providers are below 5%. He warns that if demand slows, price competition could get tougher, since everyone wants to hold market share.
“Do I want to give money to [Claude] Code and give them a 70% gross margin for this token? Or do I wanna buy it on Modal and give them a 30% gross margin for this token?”at 33:40
20VC: 20VC: Trillion or Bust: The New VC Game | Coreweave Sell-Off Creates Panic | Town at $1B, Anduril at $90B: Revenue Multiples Out of Control | General Atlantic Eyes IPO & Modal Raises $750M at $15.75B ·
20 older views on AI pricing & unit economics are on the episode pages, and all of them are a question away in your AI with Chorus.
Ask your AI about AI pricing & unit economics
Chorus gives Claude, ChatGPT and other assistants every view on AI pricing & unit economics from the market's podcasts. Ask who's worried, who isn't, who changed their mind, and why.
Chorus reports what speakers said. It isn't investment advice.