Theme · Updated
What podcasters are saying about AI infrastructure & data centers
Summary
In the past 30 days, 225 speakers on 47 finance podcasts discussed AI infrastructure & data centers in 421 views. Chorus has followed AI infrastructure & data centers since September 22, 2026: 427 views from 228 speakers on 48 shows.
Who talks about it most
The latest views
Newest first. Every quote is checked against the episode's transcript.
Jim Lebenthal · Cerity Partners
Hyperscaler capex plans from Meta, Microsoft, Amazon and Alphabet show AI spending is early, in the first third.
Lebenthal looks at the capex plans from Meta, Microsoft, Amazon and Alphabet and sees no sign of spending slowing. He takes this as the large language model providers' message that the AI buildout is still early, at best in the middle innings and more likely in the first third of the game.
“Amazon, and Alphabet, and I say it's game on. ... We're at best in the middle innings and more likely in the first third of the ballgame.”at 3:33
CNBC Halftime Report: Amazon in Focus as the Nasdaq Slides 10/3026 ·
Steve Weiss · Short Hills Capital Partners
AI spending is at its peak with significant overordering in memory, chips and overbuilt data centers.
Weiss says ROI from AI spending has been elusive across most reports, with the possible exception of Meta and Microsoft. Private internet companies are struggling to raise capital, and the take-or-pay contracts that are supposed to guarantee demand are not real. He sees the spending cycle as peaking: no aggressive capex raises this season, though spending is still high and hits cash flow. He is sure of overordering in memory, chips and data centers, as in every cycle.
“I have no doubt that there's significant over ordering in memory, in chips, and overbuilding in data centers. You see it in every single cycle.”at 8:55
CNBC Halftime Report: Amazon in Focus as the Nasdaq Slides 10/3026 ·
Joe Terranova · Virtus Investment Partners
Supply constraints can only be resolved with more spending, so investors won't pay a premium for the stocks.
Terranova says the industry faces extensive supply constraints. The only way to resolve them is more spending, and he doesn't think investors will pay a premium for a stock that has to spend more to fix supply. He sees this as part of why his earnings-season expectations went wrong.
“the only way you resolve that is through further spending and increasing spending. And I don't think you're gonna pay a premium for a stock price if you're telling me you're gonna have to spend more to resolve a supply constraint challenge.”at 21:02
CNBC Halftime Report: Amazon in Focus as the Nasdaq Slides 10/3026 ·
Kevin Warsh · Federal Reserve
Interest rates do affect the AI investment boom; they aren't divorced from the economy.
Warsh rejected the narrative that rate hikes would have no effect on AI because investors' expected returns differ. He argued interest rates get into the cracks and affect directly or indirectly almost everything. Since early September, market participants have become more discerning in their capital allocation inside and outside the technology complex. Credit spreads for certain assets have widened and gapped out, while other assets saw some tightening of financial conditions, though not significantly. He concludes that rates do matter for the new technology and investment surge.
“do interest rates matter? I believe they do. And I don't believe that they're somehow divorced from the rest of the economy.”at 15:10
The David Lin Report: Fed Split Over Rates: Why Didn't The Fed Hike Today? ·
Jay Hatfield · Infrastructure Capital Advisors
Contracted data center returns and compute pricing show no AI bubble; contracts are worth more than market.
Hatfield argues AI data center debt is not a bubble. He says the market price of compute is about $40B per gigawatt, while long-term contracts are about $15B per gigawatt, so existing contracts are worth more than their price. Five-year contracts yield high returns on equity, around 20% to 30% for Oracle, and the power-plant analogy means sites rarely get abandoned, adding franchise value. He contrasts this with 2000s fiber and late-1980s speculative office building, where projects lacked contracts. He says Nvidia's investments in partners like Marvell gained a lot, so vendor financing is fine when credit is strong.
“If you think this is all gonna be a bubble, tell me what the market price of compute is. It's about 40,000,000,000 per gigawatt. What are long term contracts? $15 per gigawatt.”at 21:49
Excess Returns: Tech is Selling Off. AI Debt Fears Are Rising. So Why Does a $4B Manager See a Reverse Bubble? ·
Kay Haigh · Goldman Sachs Asset Management
AI-linked investment-grade issuance could exceed $1 trillion within two years, straining credit capacity and widening spreads.
AI debt from hyperscalers and chip deals was initially absorbed eagerly, but issuance kept coming. Investors began pricing market capacity and balance-sheet absorption rather than just underlying risk, which caused the spread widening that started in fixed income. She tracks off-balance-sheet residual guarantees and contingent liabilities as a building concentration risk. Those deals are shorter duration and rated, so the concern is the right price and spread level rather than the structure itself. Tracking where the buildup sits is crucial because AI capex runs through the capital markets.
“The CapEx numbers are remarkable. We think you're probably going to end up north of 1,000,000,000,000 over the next in kind of two the IG space alone.”at 4:22
Bloomberg Surveillance: Markets Price in Tech Spending and Midterm Elections ·
Darius Dale · 42 Macro
Midterms and gubernatorial races could be a negative catalyst that slows the growth of AI capex.
The midterm elections and 36 gubernatorial races next week are a political catalyst that could slow the growth rate of AI capex. That is a risk to the AI infrastructure buildout that is being discounted in markets. He also argues that a slowdown in data center buildout is the only thing that could push the economy into a deflation regime.
“One thing that also raises an eyebrow as it relates to this broader AI thing in the political cycle is next week we have not just the actual midterm elections and the thirty six gubernatorial elections, which could be a really negative catalyst in terms of slowing the growth rate, of all the AI CapEx.”at 14:17
Bloomberg Surveillance: Markets Price in Tech Spending and Midterm Elections ·
Alexander 'Alty' Altmann · Barclays
AI data-centre debt is being absorbed but at a rising price, with spreads set to keep widening.
An equal-weighted index of data-centre bond spreads, measured against 20-year Treasuries (about 17-year average duration), has widened steadily. Net IG and high-yield issuance of about $1.2tn in 2026 plus government borrowing is competing with AI capex for capital. Equities have offset buybacks with issuance, so the debt is still being absorbed, but at a higher price. Alex expects spreads to keep widening to the point where they test or break the market.
“The chart before, the the IG, chart of data centers debt is telling you you can't ignore it because spreads are widening. And they're gonna continue widening to the point where it's effectively going to either test or break the market.”at 49:45
The Compound and Friends: The Bulls Are Trapped with Alex Altmann ·
Alexander 'Alty' Altmann · Barclays
Hyperscaler AI capex has reached a terminal plateau, at about $1.2tn next year and $1.4tn after.
Barclays has been to Asia and laid out its roadmap for AI capex: $1.2tn next year and $1.4tn the year after. The CapEx number can't go much higher, though it might move by $100bn here and there. Alex therefore thinks capex is at a terminal point, and that the revenue to justify it has to come from elsewhere.
“The CapEx number can't go any higher, 1,200,000,000,000 for next year, 1.4 for the year after. Like, maybe it moves by a 100,000,000,000 here and there, but we're kind of at terminal point now.”at 1:02:41
The Compound and Friends: The Bulls Are Trapped with Alex Altmann ·
Ed Elson · Prof G Media
Neoclouds have precarious debt; many may run out of money, a likely shakeout in that segment.
Elson says neoclouds, which were popular last year, are now being scrutinized for precarious debt obligations. He asks whether the money will run out and answers that for many of them it will. He expects the neocloud segment to see a correction or drawback as part of the broader AI cycle. He adds that a crash here would look small compared with past crashes.
“We're beginning to see it with the neo clouds... they have some precarious debt obligations, and will the money run out? And I think for a lot of them, will.”at 52:57
Prof G Markets: Josh Brown: If This Is A Bubble, It’s Pathetic ·
Josh Brown · Ritholtz Wealth Management
Data center buildout creates millions of construction jobs; unions gain, so backlash is muted.
Brown says Anthropic's spending of about $800–900B on compute has to be built, which means millions of jobs. He argues the construction unions have not spoken out against data centers because the buildout guarantees employment and could let them double membership. He thinks this ripple effect means even without stock options for employees, there is work for a million people, with the dignity and pride that come with it.
“There's a ripple effect anytime a company like Anthropic comes along and spends money the way that they're spending... it's employment for a million people.”at 1:10:25
Prof G Markets: Josh Brown: If This Is A Bubble, It’s Pathetic ·
Liz Ann Sonders · Charles Schwab
AI build-out sustainability is the biggest market story, given surging capital demand and rising capital costs.
Sonders sees the sustainability of the AI build-out as the biggest macro story. Demand for capital for the build-out has increased significantly and its cost is rising. She cites Larry Fink of BlackRock saying the scramble for capital is unlike anything he has seen. She questions CapEx assumptions and whether revenue generation will keep pace with the spending. She also flags that data center restrictions have become a bipartisan 'not in my backyard' issue, which could threaten operators' revenue needed for ongoing profitability. Her view is that the market has been resilient but has 'fires in the engine' that show up under the hood, with AI build-out concerns among them.
“At least in my world, the biggest story at play here is the AI build out, the sustainability of it, the significantly increased demand for capital, the cost of that capital.”at 7:46
On Investing (Charles Schwab): The Fed Holds, but the Pressure Builds ·
Connor Teskey · Brookfield Asset Management
Market overreacts to AI headlines; underlying data-center demand far exceeds available supply.
Teskey says the market reacts heavily to single announcements, such as a more efficient AI model or a large hyperscaler walking away from data-center leases, and that this seems divorced from what he sees on the ground. Right now demand is exceptionally high and supply is only a small fraction of it. Because supply is so far behind demand, small changes in demand growth would not disturb the imbalance. He acknowledges that these are huge projects requiring a lot of discipline and capital committed for decades.
“A potentially more efficient model comes out or a large hyperscaler walks away from one of its ... data center leases and the whole market reacts. That seems a little divorced from what we're seeing on the ground.”at 3:45
Goldman Sachs Exchanges: Brookfield’s Connor Teskey on the Real Bottleneck in the AI Buildout ·
Connor Teskey · Brookfield Asset Management
Scarce input for the AI buildout is execution capability, not capital or demand.
Teskey argues that the commonly cited big dollar figures are not the bottleneck, since there is no shortage of capital or demand. The real constraint is who can pull projects out of the ground: completing land development, securing grid connections, building the power supply, building and operating large-scale facilities, and providing redundancy. Capital is therefore no longer enough. He sees the market shifting toward providers that combine big dollars with big operating capabilities.
“That is not the bottleneck. There is no shortage of capital and there is no shortage of demand. The bottleneck is who can actually pull these projects out of the ground... It is not capital, it is not demand.”at 5:13
Goldman Sachs Exchanges: Brookfield’s Connor Teskey on the Real Bottleneck in the AI Buildout ·
Connor Teskey · Brookfield Asset Management
Sovereign governments, not hyperscalers, will anchor AI data-center build-outs outside the US and Asia-Pacific.
Teskey argues the AI build-out in the US is driven organically by large technology companies and hyperscalers, most of which are American and expanding globally. Europe lacks an equivalent concentration of homegrown large-scale tech companies, so its demand comes from a different source. Nation states are stepping in because AI is a productivity tool and they do not want their populations left behind. Governments will backstop AI infrastructure, and Brookfield has signed agreements in France, Sweden and Qatar in which the initial offtaker is a sovereign rather than a hyperscaler.
“We've signed agreements in France, in Sweden, in Qatar, where the initial off taker to support the build out of AI infrastructure is not a hyperscaler, but actually a sovereign. We call that sovereign AI.”at 7:56
Goldman Sachs Exchanges: Brookfield’s Connor Teskey on the Real Bottleneck in the AI Buildout ·
Giordano Albertazzi · Vertiv
Data center costs keep falling as efficiency improves, so total cost of ownership keeps improving.
Albertazzi argues the industry keeps getting more productive at the chip level, uses electricity better, and gets more efficient. Data centers should be assessed on total cost of ownership, and that is being improved constantly. Cramer calls this an economic rule that applies here. Albertazzi frames this as an ongoing improvement rather than a point-in-time price drop.
“So we should look at the data center in terms of total cost of ownership and the total cost of ownership is being improved constantly.”at 35:00
Mad Money with Jim Cramer: Mad Money w/ Jim Cramer 10/29/26 ·
Anurag Rana · Bloomberg Intelligence
Data center buildout faces rising political opposition and funding costs that could delay growth.
Two things have changed in the last six months. Interest rates have risen sharply, which affects companies' ability to fund expansion plans. The political climate has also turned against data center buildouts. The key risk is delivery: if data centers are delayed, cash flow stays negative and the projected growth fails to materialize for reasons outside companies' control. Microsoft already cited a delay at a third-party data center partner, which he sees as an early warning sign. A widespread repeat of such delays could follow because the political climate is very negative toward new data centers.
“The second is a changing political climate, which is seeing some opposition to data center buildup. ... We saw Microsoft talk yesterday about there was a delay in data centers from one of their third parties.”at 11:42
Bloomberg Talks: Earnings Special: Intel Plunges on PC Sales Slump & Amazon Slips ·
Cody Acree · StoneX
Agentic AI inference is shifting workloads toward CPUs, boosting demand for CPU makers like Intel.
Data center demand is still very much on track, and usage is shifting toward agentic inference. That trend will become increasingly reliant on CPUs rather than GPUs, which benefits Intel's data center AI business, where results beat estimates by about $500M. Agentic workloads are a growing consumer trend, so the AI trade for CPU makers is persistent and likely to accelerate as agentic use becomes a bigger share of the workload.
“We're still seeing a shift in usages towards agentic inferencing, and so that trend is going to be increasingly reliant on CPUs”at 13:26
CNBC Closing Bell: Closing Bell Overtime: Big Tech Faces Another Earnings Test 10/29/26 ·
Luke Gromen · Forest for the Trees (FFTT)
AI capex credit is in early stages, and a bust in hyperscaler and neocloud credit could trigger a recession.
Gromen points to Oracle and other hyperscaler credit default swaps, the pricing of AI-related deals, and circular financing as the visible credit stress. He says this is still early and that the more worrying second-order risk is the consumer side. He compares the potential AI credit bust to the 2000 dot-com bust, and says that if the AI credit mania bursts there will be massive losses and probably a recession, since AI is a large share of marginal growth.
“I do think the credit side of what we're seeing on AI, and then the secondary credit side, which I think is still very early days”at 19:41
MacroVoices: MacroVoices #556 Luke Gromen: US Midterms & Beyond ·
Luke Gromen · Forest for the Trees (FFTT)
AI is now too big to fail, so Washington would likely step in if the AI credit mania bursts.
Gromen hears that Washington discussed contingency plans for a possible AI credit bust over a year ago. AI is a big part of marginal US growth, so a burst would bring large losses and a recession, which would force the government to act. He argues the administration has already intervened to protect the AI sector, for example by banning DeepSeek from government work, which he sees as an industry moat.
“Yeah, AI is now too big to fail. ... So if the credit mania bursts, then yeah, there's gonna be massive losses. You're gonna have a recession”at 27:54
MacroVoices: MacroVoices #556 Luke Gromen: US Midterms & Beyond ·
Anna Macdonald · Hargreaves Lansdown
Big AI spenders are going cash-flow negative, which puts their long-term AI investment in question.
Macdonald recalls that big tech firms could fund AI investment easily when they were cash-flow machines. Now a lot of them are going cash-flow negative. Combined with weakness in core businesses, she asks what this says about how secure their long-term AI investment is. She presents this as the source of the nervousness that is hitting Alphabet and the wider market.
“A lot of them are going cash flow negative. And if the core parts of the business, no matter how great they are, just seem to be crumbling a little bit, what does that say about the security of that long term investment?”at 41:29
Merryn Talks Money: Has the UK Redefined What It Means to Be a High Earner? ·
Rob Sechan · NewEdge Wealth
Contracted demand for data-center and fiber buildout is far out; a long-term theme that will spend more.
Sechen looks at the private investment landscape and sees strong demand for products used in data centers, including fiber, with contracted demand stretching far out. He calls this a long-term secular theme that will not go away and says the spending will continue. He admits the next stretch will be tough as the market digests the news, but on the other side it will be explosive. He is positioned as a broadening-out bet on this theme.
“When we look at the private investment landscape and we see the amount of demand for product, whether it be for fiber or anything that goes into data centers and is supporting this ecosystem, the contracted demand is so... It's far as far out as the eyes can see.”at 22:52
CNBC Halftime Report: Tech Earnings Weigh on the AI Trade 10/29/26 ·
Eric Basmajian · EPB Research
The data center build-out is adding construction jobs and a manufacturing multiplier, supporting US employment.
Over the past year, nonresidential construction has added more than 100,000 jobs, mostly from data centers. The AI build-out also creates a multiplier in durable goods manufacturing, such as transportation equipment, business equipment and cooling equipment, and those manufacturing jobs have risen over the past six to eight months. Basmajian says this is a key reason construction and manufacturing are strong. Data centers are also making it hard for homebuilders and construction firms to fill positions, because HVAC installers and crews are in demand on data center sites.
“Over the past year, we've seen over a 100,000 jobs gained in nonresidential construction, which is mostly data centers... the AI build out is causing a multiplier effect through even manufacturing sectors.”at 10:52
The David Lin Report: Microsoft AI Jobs Shock: 60,000 Buyouts, Who's Next? | Eric Basmajian ·
Gary Cohn · IBM
Data centers will be overbuilt, causing a glut and failures before equilibrium.
Cohn says capacity is catching up to demand now, which is a peak of overinvestment, and once it catches up there will be equilibrium, but history shows the US overbuilds. He draws a fiber-optic analogy: the late-1990s builders did not survive, while later owners profited. Second or third owners will buy data centers at deep discounts to today's compute cost, and some current builders will fail.
“We will go through a glut. Some of these companies will not make it.”at 28:35
Squawk Pod (CNBC Squawk Box): Gary Cohn & Chris Sununu: The Fed, AI, & the Midterms 10/29/26 ·
Thomas Hoenig · Mercatus Center at George Mason University
Higher rates will squeeze the most leveraged AI builders first, limiting their borrowing and build-out.
More highly leveraged firms are the first to feel pressure, and that is what is happening with the AI borrowers referenced. If rates keep rising and refinancing of leveraged private credit and other borrowers gets harder, cash flows and financial outlooks will be affected and confidence in the industry could fall. Higher rates would make it harder for AI and other firms to borrow, invest and develop new systems.
“Firms who are more highly leveraged than others are gonna be the first to see the pressure, and that's what you're seeing going on there.”at 22:29
The David Lin Report: Fed's Next Mistake? Ex-Fed President Warns Of Market Fallout | Thomas Hoenig ·
Kai Wu · Sparkline Capital
Infrastructure builders historically don't keep long-term AI profits; value flows to users of the technology.
Kai argues that in past technology revolutions the builders and providers of infrastructure did not end up as long-term winners. Railroads, electricity, automobiles, airplanes and the dot-com era all followed this pattern. In the dot-com case, telecoms laid excess fiber, the capital cycle collapsed, prices fell, and the telecom index is still well below its high. The value went to the users, such as Netflix, Meta and Google. He applies this to AI. If AI creates large value, the question is whether it accrues to consumer surplus, to businesses that use AI, or to AI producers. His historical view is that it accrues to users. So the current infrastructure winners may not be the ultimate long-term winners.
“I would just, you know, just venture to say that never in history... has the builder and the provider of the infrastructure been the long term winner? ... It's always, you know, ultimately gone to the consumers.”at 20:08
Excess Returns: The S&P 500 Is Just 50 Stocks. The Nasdaq Is a 67% AI Bet. Kai Wu on What Else Is Hiding in Your Funds ·
Kai Wu · Sparkline Capital
Hyperscalers shifting to heavy capex is diluting their quality; capital intensity predicts weaker stock returns.
Kai argues that the megacap platforms earned their returns as asset-light businesses, such as Google's search model. The AI arms race is now pushing them into capital-intensive, asset-heavy operations, and outsiders like SpaceX and Oracle are joining. He cites a factor chart showing that the Mag Seven's capex-to-sales has risen over time. He says capital-intensive businesses are generally worse businesses, since utilities take a lot of capital just to stand still. He also cites academic factor evidence: firms that grow capex aggressively tend to underperform, and capex intensity predicts poor returns across sectors. He stops short of saying these companies will go bankrupt. His point is that the quality of their business is being diluted.
“capital intensive businesses are just worse businesses... the overall quality of their business is being diluted as they transition [from asset-]light to asset heavy.”at 29:20
Excess Returns: The S&P 500 Is Just 50 Stocks. The Nasdaq Is a 67% AI Bet. Kai Wu on What Else Is Hiding in Your Funds ·
Andrew Dubinsky · UBS
AI-linked business investment is a very strong driver of current US economic growth.
In describing the strong Q3 GDP print, Dubinsky points to very strong private demand, including really strong business investment linked to the AI capex cycle. He presents this as one of the drivers of the strong growth backdrop. He does not discuss its longer-term durability.
“and really strong business investment linked to the AI, CapEx cycle.”at 5:31
UBS On-Air: Top of the Morning: FOMC reflections & US economic update ·
Libby Cantrill · PIMCO
State-level anti-data-center politics could add moratoriums and costs, creating friction for the AI buildout by 2027.
Cantrill argues politics increasingly drives AI policy and that the market underestimates the friction the AI infrastructure buildout will face by 2027. Both chambers and 36 governorships are up for election, and many governors are running on anti-AI, anti-data-center platforms. The easiest way for them to show it is not rhetoric is a moratorium, as New York has done. Even without a moratorium, hyperscalers and data center developers face higher costs because they must invest more in host communities to win approval to build.
“I mean, this is something I do think that is underappreciated by our client base is just how bad the politics are on AI and how oftentimes politics leads policy. And I think people are underestimating how much friction there could be in this AI infrastructure build out come 2027.”at 42:07
Bloomberg Surveillance: Outlook for Inflation, Oil, and Tech ·
Trey Reik · Bristol Gold Group LLC
The AI build-out is drawing large amounts of capital from the system, making the economy more vulnerable.
Reik says there is a lot of stress in the system, citing rising debt, financing costs and a negative national savings rate. He adds that the AI build-out is pulling large amounts of capital out of the economy, which makes everything more vulnerable. He gives no figures for the capital involved, so this is a qualitative concern.
“there's a lot of stress out there. The AI build out is sucking lots of capital out of the system, which is making everything more vulnerable.”at 36:45
Wealthion: Housing Weakness is spreading to luxury buyers | Ivy Zelman ·
397 older views on AI infrastructure & data centers are on the episode pages, and all of them are a question away in your AI with Chorus.
Ask your AI about AI infrastructure & data centers
Chorus gives Claude, ChatGPT and other assistants every view on AI infrastructure & data centers 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.