Too Much Compute: Why This Top Hedge Fund Manager Believes AI is Heading for a Massive Oversupply | David Orr

Alphabet

Across the shows: 68 speakers in other episodes, 62 bullish and 46 bearish calls, past 30 days

• David Orr · Militia CapitalBullish

  Hyperscaler Google has small downside from AI capex; contracts lock in returns and capacity.

  Orr names Google, Amazon and Microsoft as his favorite AI play. Their AI capex projects earn about a 35% return on capital and about half of the capacity being built is already locked in by contracts. Doing the math, their downside is small even in a nasty bust, unless software companies go bankrupt in large numbers very quickly. If compute stays tight for a decade, the 35% returns are incredible. He had sold these stocks ahead of a possible selloff and says he might rejoin them later.

  “they're like 35% return on capital projects and they've locked in around half of the capacity they're building. So if you do the math on it, their downside is pretty small.”at 11:08

• David Orr · Militia CapitalBearish

  AI is a headwind to Google's search ads, which will shrink over time, though YouTube benefits.

  Orr sees search as a headwind from AI. People are creatures of habit, so the decline will be gradual and not overnight, but the wall-of-ads search business will start shrinking. Google can integrate its own AI into search to slow the shift, but it won't be able to keep forcing users through ads, because AI assistants will answer simple local questions well and won't put ads ahead of the answer. He sees YouTube and Maps as beneficiaries: AI makes content creation cheaper, and Google has a scale advantage there.

  “But for search, specifically, I think they're... It's a headwind. It's not like they're gonna die overnight, but that business is just gonna start shrinking at some point.”at 19:23

• David Orr · Militia CapitalBearish

  Google's top talent is spent expanding ad-driven search moats, making search worse rather than better.

  Orr argues that Google's best and brightest are working on privacy-invasive ways to monetize and on expanding the moats of the largest tech businesses, rather than on useful work. He describes this as a search business that is being made worse and more annoying so people search more and see more ads. He treats this as a bad outcome for everyone, which he links to his broader view that AI talent is being pulled into moat-building. The argument is an opinion about Google's direction rather than backed by hard numbers.

  “So they're... You have all the smartest people right now figuring out how do we infringe on people's privacy to the maximum amount? ... How do we make search actually worse and kind of annoying to you so people have to search even more?”at 42:29

Amazon.com

Across the shows: 47 speakers in other episodes, 61 bullish and 10 bearish calls, past 30 days

• David Orr · Militia CapitalBullish

  Hyperscalers like Amazon have small downside from AI capex; contracts lock in returns and capacity.

  Orr names Google, Amazon and Microsoft as his favorite AI play. Their AI capex projects earn about a 35% return on capital and about half of the capacity being built is already locked in by contracts. Doing the math, their downside is small even in a nasty bust, unless software companies go bankrupt in large numbers very quickly. If compute stays tight for a decade, the 35% returns are incredible. He had sold these stocks ahead of a possible selloff and says he might rejoin them later.

  “they're like 35% return on capital projects and they've locked in around half of the capacity they're building. So if you do the math on it, their downside is pretty small.”at 11:08

Microsoft

Across the shows: 43 speakers in other episodes, 52 bullish and 7 bearish calls, past 30 days

• David Orr · Militia CapitalBullish

  Hyperscalers like Microsoft have small downside from AI capex; contracts lock in returns and capacity.

  Orr names Google, Amazon and Microsoft as his favorite AI play. Their AI capex projects earn about a 35% return on capital and about half of the capacity being built is already locked in by contracts. Doing the math, their downside is small even in a nasty bust, unless software companies go bankrupt in large numbers very quickly. If compute stays tight for a decade, the 35% returns are incredible. He had sold these stocks ahead of a possible selloff and says he might rejoin them later.

  “they're like 35% return on capital projects and they've locked in around half of the capacity they're building. So if you do the math on it, their downside is pretty small.”at 11:08

Bloomberg L.P.

• David Orr · Militia CapitalBearish

  AI will clean financial data cheaply, eroding Bloomberg's data-cleaning moat within a few years.

  Bloomberg's moat was years of staff cleaning and structuring financial data. Orr argues AI can now collect old annual reports and clean the data far faster and cheaper. He expects that in a few years he will trust AI's numbers more than Bloomberg's analysts' output. That makes a competing product that costs a tiny fraction of the price and is better, which erodes Bloomberg's pricing power.

  “Bloomberg's old moat was that, you know, years and years of guys, like, cleaning up this data... But now... we just have the AI cleaned it up. I trust the AI's numbers better than the Bloomberg guys' numbers in a few years.”at 36:46

Meta Platforms

Across the shows: 70 speakers in other episodes, 72 bullish and 36 bearish calls, past 30 days

• David Orr · Militia CapitalUncertain

  Meta's social moat is eroding as it becomes media, though AI-improved content could keep users engaged.

  Orr says Meta's ad business has done incredibly well over the last few years, but the platform has become less social and more media, which he sees as weakening its moat. He compares it to Facebook, where friends and family were the reason to stay, and argues a niche upstart could now draw users away, so the platform could become a commodity in about five years. He then presents the bull case: AI could make content better and more addictive, keeping users engaged. He calls it a bear case he is laying out and says it could be a good business, depending on how this plays out.

  “It could be a good one. It depends because it used to be called social media, and I feel like every year it becomes less social. It's just now media. ... And so if it's just only media, then maybe they don't have any moat at all.”at 45:43

Market sentiment, positioning & volatility

• David Orr · Militia Capital

  The semiconductor and AI selloff was driven by crowded positioning and forced leverage unwinds, not fundamentals.

  Orr says the AI trade peaked around June when too many investors locked in high prices, leaving no new marginal buyers. Hedge funds run 6 to 10 times leverage. Strange moves, such as Japanese small-cap value rising while the Nikkei fell 5%, suggested a leveraged fund blowing up, triggering a chain reaction. Levered long semiconductor trades then broke as a group. The price action drove the narrative, and nothing fundamental changed.

  “I just think it was positioning, and then the price drove the narrative. You know? That's all that happened.”at 5:46

• David Orr · Militia Capital

  Mega-cap share and bond issuance plus a possible Anthropic IPO drain market liquidity, a bearish signal.

  Orr says there are already several large liquidity sinks. The Magnificent Seven are issuing shares heavily and also issuing bonds, taking as much liquidity from the market as they can. An Anthropic IPO could be another one, though it may not happen. He sees liquidity as another reason to be bearish.

  “The magnificent seven, they're just issuing shares like crazy right now or in the bond market especially too. They're just taking everything out of the liquidity they possibly can.”at 14:12

AI & jobs

• David Orr · Militia Capital

  Most white-collar jobs will go away as AI makes routine knowledge work nearly free within five years.

  Orr says white-collar tasks such as accounting and routine administrative work are simple. As AI improves, he expects all of that work to go to zero in about five years, at almost no cost and using very little compute. He uses personal tax returns as an example: a machine that handles a complex return once can be copied many times, so the price falls to nearly free. He thinks most white-collar jobs will go away, and the companies that replace them with AI could become cash printers if they have a moat.

  “No. That, I mean, that stuff's so simple. And just the way AI is improving, you can assume that all of that stuff is gonna go to zero in five years.”at 27:24

• David Orr · Militia Capital

  AI will shrink the number of people doing physical work too, pushing talent toward hands-on trades.

  Orr argues that as AI absorbs digital work, the smartest and hardest-working people will be freed up to do physical things, such as designing and building better homes. He sees a shift back toward the physical world. He also says the number of people doing the physical part of building will fall too, because automation will take over jobs like driving. He gives the homebuilding example and a limited partner's hotel project as illustrations. This is a long-run thesis about labor rather than a near-term forecast, and he admits it is speculative.

  “I think the number of people even doing the physical part of it is gonna go down too. Like, all the drivers are gonna go away.”at 41:35

AI infrastructure & data centers

• David Orr · Militia Capital

  AI compute is heading for oversupply because the physical world will take years to catch up.

  Orr expects AI to be in oversupply: there will be too much intelligence, and the physical world (power, labs, buildouts) will take years to catch up. Short term, bottlenecks will limit deployment while compute is already in excess. Hyperscaler capex through 2028 is enormous, and besides software companies almost no one is buying this compute. Software companies currently need more compute but will eventually say they don't need it and resell capacity, pushing prices down sharply. Evidence cited: spare-capacity offers fell from about $12 per compute hour in September to about $4 an hour now, though he calls this a noisy signal because most big contracts are locked in. He says a year or year-and-a-half timeline would mean June was the top, but if it takes about four years, AI stocks could still rise first. Long term, he thinks intelligence cannot be oversupplied.

  “AI is gonna be in an oversupply. So we're gonna have too much intelligence, and it's gonna take years for the physical world to catch up.”at 5:02

Software & SaaS

• David Orr · Militia Capital

  Most software companies look doomed over a few years under AI unless they have wide moats.

  Orr is bearish on most software companies over a few years. Unless they have a good moat, it is hard to see how they are not 'totally screwed', because AI makes their products easy to replicate. Wide-moat companies should benefit. He notes that software companies are currently the biggest buyers of AI compute but will eventually stop needing it and resell their capacity.

  “in my opinion, being bearish most software companies is fundamentally right looking at a few years. It's hard to see how they're... Unless they've got a, you know, good moat.”at 9:04

Power demand & the grid

• David Orr · Militia Capital

  Power plant shortages, not Texas policy, are the real bottleneck limiting AI data center buildouts.

  Orr says the Texas governor's victory was a headline but not the real issue. The real bottleneck is power plants, which were already severely limited, so the news did not change the outlook. Data centers can also be built in other states, so the policy news did not matter much.

  “the real bottleneck was just the power plants. So it was already gonna be severely limited either way.”at 10:45

US tax & spending policy

• David Orr · Militia Capital

  A Democratic sweep in 2028 could raise corporate taxes, so a 4% selloff is roughly fair, maybe 6-7%.

  Orr thinks the market may be pricing in risk that Democrats do well in 2028 and raise corporate tax rates. Such a tax increase would be deferred, maybe to 2030. He puts the chance at 30 to 50%. A 4% drop seems reasonable, and he thinks the drop should be closer to 6-7%. He says if the tax hike did happen it would not be enough to fully justify the selloff.

  “let's say you got also, like, a 30 to 50% chance that it happens. ... And 4% seems, you know, not too far off. Maybe it should be more like 6% or seven.”at 13:27

Antitrust & tech regulation

• David Orr · Militia Capital

  More antitrust action against Big Tech is long overdue and would hurt stocks, Google most exposed.

  Orr thinks the Democratic election results could bring more antitrust action against Big Tech. That would be good for the world but bad for stocks. Google is the company most at risk, though it has already been through antitrust cases.

  “Maybe they'll start doing more antitrust. That's long overdue. I think that'd be great for the world, but, yeah, probably pretty bad for stocks.”at 17:29

Pharma & biotech

• David Orr · Militia Capital

  Physical lab capacity, not compute, will bottleneck AI-driven drug discovery for many years.

  Orr thinks AI could drive medical research, but the bottleneck will be physical. AI would propose experiments, and humans or robots would have to run them and collect data. Scaling experiments by roughly 100 times would require building many new labs, which will take many years. So the physical bottleneck will hit much faster than compute limits.

  “if this is gonna increase the number of experiments being run by a 100 times, you know, actually physically building all of those labs and everything, I mean, that's just gonna take many, many years.”at 23:42

Frontier AI labs & model race

• David Orr · Militia Capital

  AI labs pitch cures because software customers won't sustain demand, so they need a new big customer.

  Orr reads the shift in AI lab messaging toward curing cancer and other outcomes as a sign of a demand problem. Software companies are the big customer for compute today and are unlikely to keep growing their purchases, so labs need a new big customer and must sell a new story. He hopes the pitch works but doubts it will.

  “they're the big customer today, and they need a new big customer. So they need to sort of sell this idea.”at 27:02

AI pricing & unit economics

• David Orr · Militia Capital

  Once AI systems are built they get copied, so prices fall to nearly free and AI makers profit little.

  Orr argues that once a company builds an AI system for a complex task, such as a personal tax return, it can make many copies, so the price will be close to free. The system makers will not profit from building it. He thinks the only winners are consumers. In competitive industries, margin gains from AI turn into a race to the bottom, where the old 'your margin is my market share' dynamic applies.

  “they'll probably get 10 of them made. And that means the price will be, like, basically free. They're not gonna make any profit from having even made the system”at 28:24

Enterprise AI adoption & AI software

• David Orr · Militia Capital

  Custom AI tools sold to corporations lack durable pricing power as they get simple and copied.

  Orr is skeptical that vertical or custom AI solutions sold to corporations will have durable pricing power. Such solutions may look like good businesses for a few years, but the underlying work is simple and easy to copy. He expects that once AI is good enough, in about five years, these tools will be cheap. He illustrates this with his own firm: it is building an in-house risk system and terminal that it plans to give away free, lowering prices as others copy it.

  “No. ... maybe for a few years, maybe it'll seem like a good business for a few years, but... in five years, I mean, I'm pretty sure.”at 33:55

Professional services & law

• David Orr · Militia Capital

  Big law firm work is pure rot; AI could free the people doing it for useful work.

  Orr says nobody likes working at big law firms and they stay mainly for the pay. He asks what happens if those people start doing something useful, and calls big law pure rot as an example. The point is that AI could redirect talent away from low-value professional work. Max Wiethe pushes back that high-status jobs such as big law also signal class, so a replacement like carpentry may not take their place. Orr's case rests on his view of the work and the talent it absorbs, with no specific data.

  “Like, nobody likes working at the big law firms. They all hate their jobs... But what if all those kinds of people started actually doing something useful? The big law is a good example because that's just pure rot.”at 43:09

US stock market outlook

• David Orr · Militia Capital

  Orr keeps his fund's beta near zero and does not bet on market direction.

  Orr says his fund's beta has recently read negative, but he does not fully trust the measure and aims to stay around zero. He says he is not worried about whether the market goes up or down, and that nobody knows where it will end the year. He points to 2022, when he was net short bad companies, but says his beta was still close to zero then. In this market, he says he is not trying to bet on direction and is staying out of the AI basket he thinks could fall. Because he runs the fund with this exposure, the stake is investor.

  “It's not super negative, but I don't really know if I trust it because you can never tell that much. I'm just aiming to be around zero. I just try to be in a spot where I'm not worried if the market's going up or down.”at 52:11

AI bubble & tech valuations

• David Orr · Militia Capital

  The AI trade is real, but some junky companies at 50x revenue could be wiped out.

  Orr says the AI trade is fundamentally different from past manias because the technology is real and there is a real bull case. He is still concerned about weak companies valued at about 50 times revenue, with a market value near $80 billion, and says a basket of these can collapse. He also says he could be wrong about the overbuild or glut roughly 40% of the time. That is why he is not betting on the theme and is staying out of it, which means he sees valuation risk in the weaker names rather than a broad crash.

  “The AI trade's way different, of course. This is fundamentally a real thing. There is a real bull case. But, you know, for some junky thing trading for 50 times revenue at $80,000,000,000 or whatever, and there... You know, it's like, okay. A basket of these things can kill it.”at 53:02