
Sep 30, 2026 · 2h 60m
Listen from 2:02:07
Listen at 2:02:07
AI’s boom tests infrastructure, regulation, and human judgment
White House AI Accord, Model Welfare, Ken Griffindor School of Market Wizardry | Ali Golshan, Mike Intrator, Vlad Tenev, Pari Singh, Minna Song, Dev Ittycheria, Daragh Murphy, Noah Friedman
The episode connects AI’s rapid expansion to the practical constraints of power, security, talent, markets, and public oversight.
- 1AI infrastructure is becoming a competition over power, heterogeneous compute, security, and the economics of capacity.
- 2Open-weight models and autonomous agents expose unresolved questions about liability, regulation, formal verification, and financial safeguards.
- 3AI is moving into hardware, housing, healthcare, payments, and customer intelligence, where adoption depends on workflow and trust.
Don't miss
Ali Golshan lays out why autonomous agents require security controls across software, infrastructure, formal verification, and hardware.
The brief
The hosts frame the AI boom through a White House superintelligence dinner, open-weight regulation, energy demand, talent competition, and the political reach of long-form media.
The central tension is whether AI’s economic promise can outrun its constraints: power generation, compute capacity, liability rules, recruiting norms, and safeguards for autonomous systems.
Ali Golshan explains NVIDIA’s OpenShell project as a full-stack answer to agent security, combining infrastructure controls, formal verification, and hardware protections.
CoreWeave’s Michael Intrator describes the shift from GPU provider to AI cloud, while Robinhood’s agentic-finance discussion raises questions about guardrails and market effects.
Later interviews trace AI’s expansion into hardware engineering, housing, healthcare, payments, and customer intelligence, where specialized workflows may matter more than slogans.
The episode’s broadest takeaway is that AI’s next phase will be decided not only by model capability, but by who can make deployment reliable, economical, and trusted.
What was said on this episode
74 statements · 40 positive · 17 negative · 17 neutral
A 24/7 livestream of the president may emerge soon.
“I wouldn't be surprised if there's a 24/7 livestream of the president or something close to it in the near future.”
Listen at 18:49
Advisors joining competitors should disclose and leave active roles promptly.
“if you are basically transitioning from one company to a competitor, you need to, as soon as you know that that's actually happening, you need to make it clear and step back from your active role.”
Listen at 25:43
Elite employees lacking organizational loyalty are legitimate poaching targets.
“anybody who is elite and not loyal is unfortunately fair game.”
Listen at 31:33
A 1% increase in power usage produces roughly a 1% increase in GDP.
“I would bet anyone that 1% increase in power usage corresponds to roughly 1% increase in GDP.”
Listen at 35:09
A one-percent power-use increase roughly corresponds to one-percent GDP growth.
“I would bet anyone that 1% increase in power usage corresponds to roughly 1% increase in GDP.”
Listen at 35:09
Adding 10 gigawatts of power would increase GDP by 2%.
“And so 10 gigawatts would be a 2% increase in GDP.”
Listen at 35:15
Ten additional gigawatts would increase GDP by approximately two percent.
“And so 10 gigawatts would be a 2% increase in GDP.”
Listen at 35:15
One continuously consumed gigawatt generates about $60–65 billion in GDP.
“A single gigawatt generates about $60 to $65 billion of GDP per continuously consumed gigawatt.”
Listen at 37:22
Anthropic and OpenAI each reportedly have $60–70 billion ARR and roughly one gigawatt live.
“Anthropic and OpenAI are reportedly sitting between $60 and $70 billion in terms of ARR, and both of them have around 1 gigawatt live.”
Listen at 37:40
AI compute has tripled annually while revenue has grown tenfold annually.
“compute has been tripling every year, but revenue has been 10x-ing every year.”
Listen at 39:09
Demand currently favors AI models that mimic human behavior.
“So far, there is way more demand for models that mimic human behaviour, period.”
Listen at 48:48
Putting an open-source model behind an API is presented as the filtering mechanism.
“How do you filter an open source model? You put it behind an API.”
Listen at 55:53
Self-hosted unrestricted AI models will face substantial opposition.
“The self-hosting unrestricted models gonna have a lot of pushback”
Listen at 56:06
The AI industry is broadly converging on restricting unrestricted models.
“everyone's sort of aligned on this.”
Listen at 56:33
The prediction that AI music will end streaming will not come true.
“I don't think anything he's talking about will come to fruition.”
Listen at 1:06:49
AI music will not make Spotify and streaming obsolete.
“I don't think anything he's talking about will come to fruition.”
Listen at 1:06:49
Spotify will survive competition from AI-generated music.
“I still remain optimistic that Spotify will survive the onslaught of AI-generated music”
Listen at 1:07:38
NVIDIA intends OpenShell to become a community-owned project.
“We want it to be a true community project.”
Listen at 1:10:30
OpenShell is intended to become a widely adopted standard.
“We intentionally built it open so it becomes a standard.”
Listen at 1:10:34
- Ali Golshanon Formal verification for AI agentsPositive1:12:34
Formal verification can substantially improve agent safety when applied at appropriate stack layers.
“I'm incredibly optimistic about formal verification, but it has to be applied the right places in the stack for it to have the effect it needs.”
Listen at 1:12:34
- Ali Golshanon AI agents in legacy environmentsNegative1:20:17
Many current AI-agent problems result from unsuitable operating environments.
“Most of the problems we're seeing is we are running an advanced technology into an environment that is not meant for it or designed for it.”
Listen at 1:20:17
CoreWeave is transitioning from GPU provision to delivering an AI cloud.
“What we're really here to do now is to deliver an AI cloud”
Listen at 1:23:48
Compute-driven intelligence will permeate every part of civilization.
“compute is going to drive intelligence that is going to permeate every asset of civilization”
Listen at 1:27:34
Compute-driven intelligence will permeate every part of civilization.
“compute is going to drive intelligence that is going to permeate every asset of civilization”
Listen at 1:27:34
Capital-intensive cloud businesses tend toward consolidation.
“capital-intensive businesses tend towards a, uh, consolidation”
Listen at 1:32:44
Capital-intensive cloud businesses tend toward consolidation.
“capital-intensive businesses tend towards a, uh, consolidation”
Listen at 1:32:44
Robinhood aims to provide users with hedge-fund-like trading capabilities.
“we want to give you the power of a hedge fund in your pocket”
Listen at 1:40:40
Robinhood Agents will let ordinary users create sophisticated trading strategies.
“Robinhood agents will make it easy for anyone to be a financial trading quant.”
Listen at 1:41:06
Robinhood Agents require trade approval by default.
“Trade approvals are on by default.”
Listen at 1:45:21
Agentic trading currently has clear product-market fit and utility.
“Agentic Trading has like clear product market fit and utility now.”
Listen at 1:53:08
- Pari Singhon AI hardware engineeringPositive1:54:55
AI can reduce hardware design cycles from months to hours.
“when we see design cycles of rockets and airplanes and cars go down from 12 months to 12 weeks to 12 hours”
Listen at 1:54:55
- Pari Singhon AI-assisted hardware engineeringPositive1:54:55
AI can reduce rocket, airplane, and car design cycles from months to hours.
“when we see design cycles of rockets and airplanes and cars go down from 12 months to 12 weeks to 12 hours”
Listen at 1:54:55
- Pari Singhon AI-human hardware design collaborationPositive1:55:49
AI-human collaboration will produce new classes of hardware products.
“I do think we're going to get to the point where AIs and humans collaborate together to design a new class of products.”
Listen at 1:55:49
- Pari Singhon AI-human hardware collaborationPositive1:55:50
AI-human collaboration will produce new classes of products.
“we're going to get to the point where AIs and humans collaborate together to design a new class of products.”
Listen at 1:55:50
- Pari Singhon AI hardware verificationPositive1:59:52
AI can shift hardware verification from annual manual checks to continuous verification.
“what AI enables for us is this idea that we can move from manual verification which is once every year, to continuous verification”
Listen at 1:59:52
- Pari Singhon AI-assisted hardware verificationPositive1:59:52
AI can shift hardware verification from annual manual checks to continuous verification.
“what AI enables for us is this idea that we can move from manual verification which is once every year, to continuous verification”
Listen at 1:59:52
- Pari Singhon AI-assisted hardware supply-chain collaborationPositive2:02:07
AI can reduce supplier-change impact analysis from weeks or months to seconds.
“what used to take weeks or months before can now be done in seconds.”
Listen at 2:02:07
AI can process hardware-partner changes in seconds instead of weeks or months.
“what used to take weeks or months before can now be done in seconds.”
Listen at 2:02:07
Flow reports that 96% of customers arrive inbound and launch within 30 days.
“96% of our customers come to us inbound and they're able to get up and running within 30 days.”
Listen at 2:03:09
Elise AI products are used by 20% of the US apartment market.
“today, we serve 20% of the US apartment market already uses some of our products.”
Listen at 2:07:35
Housing and healthcare together account for about 40% of US GDP.
“housing and healthcare together are the 2 largest sectors there, about 40% of the US GDP.”
Listen at 2:08:47
Housing and healthcare together account for approximately 40% of US GDP.
“housing and healthcare together are the 2 largest sectors there, about 40% of the US GDP.”
Listen at 2:08:47
Meta lacks substantial enterprise expertise
“Meta really didn't have a lot of DNA around enterprise expertise”
Listen at 2:13:46
Meta would find it difficult to build enterprise expertise organically
“to try and organically build that would probably be challenging”
Listen at 2:13:57
Products loved by developers should not be bet against
“You don't bet against a product that people love.”
Listen at 2:17:29
Raising startup capital was very expensive after the dot-com crash
“raising capital was super expensive”
Listen at 2:18:34
BladeLogic had to build its business without favorable market tailwinds
“we had to build the business the hard way”
Listen at 2:19:14
BladeLogic raised $29 million in total before its S-1
“I only raised $29 million in total”
Listen at 2:19:55
Open-source-as-a-service was MongoDB's monetization breakthrough
“the breakthrough was, was really open source as a service”
Listen at 2:21:59
Open-source-as-a-service enables usage-based charging
“when you offer open source as a service, you charge for any— every little bit of usage”
Listen at 2:22:03
Customers will not permit providers to audit on-premise model users
“There's no way that's gonna happen.”
Listen at 2:23:11
Management should control performing companies; boards should control underperforming ones
“If the company's performing, management's in charge. If the company's not performing, the board's in charge.”
Listen at 2:24:23
Investors generally do not want to replace founders
“no investor wants to go in and replace the founder”
Listen at 2:25:35
A CEO's core responsibilities include clear strategy, team building, and culture creation.
“the job of a CEO comes down to 3 things. Being very, very clear in your strategy.”
Listen at 2:28:38
A CEO's job consists of strategy, team-building, and culture
“the job of a CEO comes down to 3 things”
Listen at 2:28:38
- Daragh Murphyon Bank and credit-card infrastructureNegative2:33:55
Banks and credit cards generally rely on legacy mainframe infrastructure.
“All the banks, every credit card in your wallet, they rely on legacy infrastructure, mainframes in the Midwest.”
Listen at 2:33:55
- Daragh Murphyon AI agents and bank-account paymentsPositive2:38:09
AI agents can optimize bank-account payments.
“In a world of agents, you can optimize that.”
Listen at 2:38:09
- Daragh Murphyon card interchangeNegative2:38:30
Agentic bank-account payments threaten card interchange
“this is actually the first time I think interchange is in danger”
Listen at 2:38:30
- Daragh Murphyon American commerce on card railsNeutral2:38:34
About 2% of American commerce flows across card rails.
“it's 2% of American commerce goes across these rails.”
Listen at 2:38:34
- Daragh Murphyon Mastercard and AmexNegative2:39:11
Agentic payments will be bearish for Mastercard and Amex
“it's actually very bearish, I think, for those like MasterCard, Amex”
Listen at 2:39:11
- Daragh Murphyon agentic paymentsPositive2:39:17
Agentic payments will benefit merchants and consumers long term
“longer term, I think it's going to be great for merchants and consumers”
Listen at 2:39:17
- Minna Songon Context-aware marketingPositive2:43:19
Context-aware marketing performs better than irrelevant personalized messaging.
“a brand that hits me with an email about performance in the gym or when I'm up late working on my business, like that's gonna slap, that's gonna hit a lot better than if they hit my mother with that”
Listen at 2:43:19
- Minna Songon Consumer customer demographicsNeutral2:46:00
Some brands’ actual highest-value customers are affluent Midwest mothers, not urban professionals.
“it turns out when you activate, it's like, no, it's actually Midwest moms with high disposable income. They're the ones paying your bills.”
Listen at 2:46:00
- Minna Songon consumer customer demographicsPositive2:46:01
Some brands' most valuable customers are affluent Midwest mothers
“it's actually Midwest moms with high disposable income. They're the ones paying your bills.”
Listen at 2:46:01
- Minna Songon marketing personalizationNegative2:48:53
Most current personalization relies primarily on behavioral signals
“that personalization was really just behavioral”
Listen at 2:48:53
- Minna Songon Behavioral personalizationNegative2:48:53
Behavioral personalization lacks meaningful context about consumers as people.
“that personalization was really just behavioral, right? You bought this, you might like that, right? It has no context on who the hell you actually are as a person.”
Listen at 2:48:53
- Minna Songon behavioral personalizationNegative2:49:23
Behavioral-only personalization produces ineffective recommendations
“it just doesn't work”
Listen at 2:49:23
Agentic commerce remains at an early development stage
“agentic commerce as a concept is in its very early days”
Listen at 2:49:57
Agentic commerce remains at an early stage of development.
“agentic commerce as a concept is in its very early days”
Listen at 2:49:57
- Minna Songon GraziPositive2:55:40
Grazi has become one of America's largest wine businesses
“Grazi, which has become one of the largest wine businesses in America”
Listen at 2:55:40
Young people continue to drink despite contrary claims
“the notion that young people don't drink anymore is ridiculous”
Listen at 2:56:00
- Minna Songon Young people’s alcohol consumptionNeutral2:56:03
Young people’s drinking was delayed by COVID rather than permanently reduced.
“I think that they were just delayed because of COVID”
Listen at 2:56:03
- Minna Songon Young people’s alcohol consumptionPositive2:56:10
Young people’s drinking returned to normal after COVID-related delays.
“Returned to normal. They were just delayed.”
Listen at 2:56:10
- Minna Songon loneliness epidemicNegative2:56:14
The loneliness epidemic remains real
“I still think the loneliness epidemic is real”
Listen at 2:56:14
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.