Is The AI Selloff A Buying Opportunity? | Warren Pies

Anthropic

Across the shows: 86 speakers in other episodes, 106 bullish and 65 bearish calls, past 30 days

• Warren Pies · 3Fourteen ResearchBullish

  Anthropic's revenue per compute dollar rose sharply, so its margins look strong and inference costs fall as a share.

  Pies cites Anthropic's S-1 as hard evidence the AI business is fundamentally strong. Revenue per dollar of compute spend rose from 57 cents in Q2 of last year to $2.08. It dipped somewhat with the Q3 release but still shows strong margins. Inference is also a shrinking share of compute spend, and revenue per dollar of inference compute has risen every quarter, including Q3. He says these figures show the AI trade is intact, without relying on claims about what the technology can do.

  “gone from 57¢ in q two of last year up to $2.08. It dropped a little bit with their release here in q three, but it's still strong margins.”at 19:22

AI infrastructure & data centers

• Warren Pies · 3Fourteen Research

  Pressure from Texas lawmakers to slow data center build-outs is a real risk; capital is holding back.

  Pies says his political contacts are concerned about a midterm sweep, and pressure is building at places like the Texas state legislature to slow the pace of data center build-outs. He thinks that is a real risk, and it has led investors to hold back capital. Combined with the funding and rate concerns, it makes the near-term path for the build-out less certain, though he has no clear view on how the politics will unfold.

  “even the pressure that's coming, at places like, Texas state legislature to slow the pace of data center build outs, like, I think that's a real risk.”at 6:48

• Warren Pies · 3Fourteen Research

  GPU availability is tight and rental prices keep rising, contradicting the model's assumed 25% annual rent decline.

  Pies's firm tracks GPU availability daily through NeoCloud access. Availability fell early this year as agentic AI took off, and Blackwell B200 availability has been zero, which has tended to lead rental prices. Rents have crept higher as a result. His data-center economics models assumed GPU rents would fall about 25% a year, and that has not happened, so the market is tightening rather than overbuilding. Longer-dated rental contracts trading above spot, which he reads as capacity being locked in rather than weakness, reinforce the view.

  “B 200 availability has been zero. So that's... That has had a tendency to lead rental prices, and sure enough, that's what we've seen is that rental prices continue to creep higher.”at 16:11

• Warren Pies · 3Fourteen Research

  Satellite data shows data center build pace matches estimates, with 27+ GW due online in 2027.

  Pies's firm uses satellite data to track every data center under construction, from groundbreaking to energized shell, and compares the pace with semiconductor revenue and earnings estimates. So far the build-out matches the estimates and there is no supply shortfall yet. His models suggest an additional 27-plus gigawatts coming online over 2027, which supports the semiconductor group's top-line estimates. He says the market is anticipating rather than reacting to constraints. A real power, material or labor shortage would show up in the data, and would mean an underbuild rather than an overbuild.

  “We have our models suggest there's gonna be an additional 27 plus gigawatts coming online over 2027, which does, in our in our view, that supports the the top line estimates for the semiconductor group.”at 23:11

• Warren Pies · 3Fourteen Research

  Hyperscaler selloff alongside rising GPU rents makes no sense; he wants to own hyperscalers.

  Semis and hyperscalers used to move together as one AI trade. Now both are selling off while GPU rents rise. If the selloff is a supply worry, the owners of data center capacity that gets ahead of the crunch should earn strong economics, and the hyperscalers can cut capex while compute stays tight and capital needs fall. Pies therefore sees the price action as panicky and says he already wants to own hyperscalers, which would become more attractive still. He also argues the return-on-investment debate is being ground down, since all hyperscalers rallied on cloud revenue ramp in the Q2 reporting season and the returns are starting to emerge.

  “So, like, you need to own hyperscalers in that scenario. I could definitely see that being bearish for semis, but not for both.”at 24:46

Middle East & Iran

• Warren Pies · 3Fourteen Research

  Return to bombing campaign unlikely; energy-infrastructure strikes and ground invasion lack support.

  Pies is skeptical that the US will go back to a bombing campaign. Discriminate bombing has already been done. Indiscriminate strikes on energy infrastructure have been threatened but not executed, because the regional response would be devastating and Gulf countries are not on board. A ground invasion would bring mass casualties with no political appetite, and he thinks a Democratic midterm sweep would reduce the scope for one. Trump's threats are not a reliable signal of action.

  “I've always been a little skeptical of that viewpoint that we're gonna just go right back to a bombing campaign.”at 41:03

• Warren Pies · 3Fourteen Research

  The US blockade is working, choking Iranian exports and cutting Iranian floating storage.

  Pies judges the blockade to be working because Iranian exports are being choked off and Iranian floating storage is down. He qualifies this by noting that a blockade needs a lot of patience to execute properly, and he questions whether Trump has that patience to see the strategy through.

  “I think the blockade is working. I think we are choking off Iranian exports. Iranian floating storage is down.”at 42:27

• Warren Pies · 3Fourteen Research

  An Iran deal is more likely after the election than before, and may look politically bad.

  Pies thinks the consensus is overly crowded into the view that the conflict will inevitably restart. He expects the next geopolitical move to dictate oil prices. A deal is more likely after the midterms because some elements of any deal will look politically bad compared with the status quo. He notes that SPR and G7 releases are running low, so the situation needs to be resolved within roughly the next 45 days.

  “I think a deal post election is more likely than a pre... deal pre election, because more than likely, the deal is gonna look pretty... There's some elements to it that are gonna look pretty bad politically.”at 42:51

Oil & fuels

• Warren Pies · 3Fourteen Research

  Keep energy overweight in a balanced portfolio; it is the only real hedge against a sideways conflict.

  Pies has long argued for an overweight in energy within a balanced portfolio. Given how uncertain the geopolitical path is, he sees no other way to hedge the risk that the conflict goes sideways. He presents energy as the hedge that fits this unresolved situation.

  “I've always said, you have to have an overweight energy within a balanced portfolio because there's no other way to hedge the risk that this thing goes sideways.”at 44:45

• Warren Pies · 3Fourteen Research

  Elevated diesel crack spreads are temporary and will revert as refiners retool and demand destructs.

  Pies explains that high crack spreads normally resolve as refiners retool to produce more of the product and supply comes back. Demand destruction at roughly $200 diesel and fuel switching also help. He says sustained high spreads need a structural oddity, such as the early shale-era crude export ban that favoured refiners. Current spreads are cyclical, and refining is not a business to get married to, so he expects them to work themselves out.

  “So I think we're gonna end up a little mix of both, and you're gonna see a reversion of these spreads.”at 50:27

• Warren Pies · 3Fourteen Research

  Oil prices will stay sticky high; a collapse is implausible with $70 middle-distillate cracks.

  Pies argues that refiners will eventually bid up crude to meet product demand, so oil prices will stay sticky high. He says a collapse in oil is not realistic while middle-distillate cracks are around $70 a barrel, citing Singapore gasoil cracks as a reference. He also notes demand destruction at roughly $200 diesel and fuel switching as partial offsets, so the outcome is a mix of demand destruction and refiner supply.

  “You'll see refiners also will ultimately bid up oil. You'll see oil prices stay sticky high.”at 50:30

US stock market outlook

• Warren Pies · 3Fourteen Research

  Downgraded equities to neutral: consolidation, staples rotation and credit stress argue against chasing this market.

  Pies's firm downgraded equities after the market entered a consolidation, a tight trading range with no new highs or lows near highs. Consolidations signal lost momentum and a likely trend change. Leadership also points to a breakdown: financials have rolled over relative to staples, as they did before the Liberation Day and earlier 2025 selloffs. Staples are up 14% in October, and a market led by energy and staples is historically unhealthy and lacks the tailwind needed for equity beta. Hyperscaler credit spreads breaking above July levels added to the case for a shift to breakdown risk. Sentiment has not washed out and positioning still has room to fall, so the firm neutralized risk and will wait for a breakout or extreme fear before repositioning.

  “So either going to buy a breakout, that's how we reposition now, or you're going to buy extreme fear. And we don't have either of those right now. So that's that's why we neutralize risk.”at 8:06

• Warren Pies · 3Fourteen Research

  AI selloffs push rates down and lift staples and rate-sensitive stocks, though a bit overdone.

  Pies observes that when the AI trade sells off and rates come down, staples and rate-sensitive parts of the market tend to outperform. He thinks this move is somewhat overdone but still reflects real economic conditions.

  “to the extent rates come down when the AI trade comes off, you see a bit at things like Staples and rate sensitive pockets of the market.”at 38:27

US Treasuries & yields

• Warren Pies · 3Fourteen Research

  The 10-year yield now tracks Nasdaq more than oil, as the AI build-out replaced oil as the inflation driver.

  The 10-year yield used to track oil, so a war-ending oil price drop could have brought relief in rates. That relationship has flipped: the 10-year is now more correlated with the Nasdaq than with oil. Pies sees a handoff of inflationary causes from the war and oil to the build-out, which he calls a problem. The SOFR-to-10-year correlation has stayed above 90% throughout, which reinforces that Fed expectations drive the curve.

  “the correlation's flipped more towards, like, Nasdaq. So Nasdaq and ten years now much more correlated than the ten year is with oil.”at 29:49

• Warren Pies · 3Fourteen Research

  After one more hike the Fed is done; dropping its hawkish bias lets the two-year yield fall.

  With the two-year around 4.6%, Pies thinks the Fed has one more hike left and will then keep a hawkish or hiking bias. If that bias goes away and the Fed says it is done, he sees room for the two-year yield to move down and the yield curve to flatten. That would reaccelerate parts of the economy that have been crowded out by the AI build-out.

  “if that hawkish bias goes away and they say, oh, here we're done we're done done, then I think that you have room to move the two year down.”at 39:08

US rates & the Fed

• Warren Pies · 3Fourteen Research

  Fed hikes now target the AI build-out; it may tighten until it breaks AI funding markets.

  The Fed moved from a cutting cycle before the war to a hiking cycle that has now realized. The Fed's focus has shifted from oil, which could be eased by a supply-side resolution, to the AI build-out. That leaves a less clean path out of the hiking cycle. Macro is the intermediate-term risk and a bearish situation. The Fed cut into a capex cycle for the first time ever, and inflation pressure is now emerging late in that cycle. Pies says the Fed may not stop until it has broken something in the AI trade or in funding markets. A credit wobble could end the hiking, though he doesn't see the Fed killing the AI trade.

  “the risk is that the Fed doesn't stop until they've broken something in the AI trade in the funding markets”at 30:53

• Warren Pies · 3Fourteen Research

  Fed is squeezed: hiking to appease political backlash risks more heat if it damages the AI trade.

  Pies says the Fed is in a tough spot. It started hiking in response to the Iran war and to political dissatisfaction that has built up since COVID, which he links to blown-out deficits. He sees the Fed becoming self-conscious about the inflation story and the political backlash. He argues that if the Fed damages the AI trade in a meaningful way, it risks even more political heat. He calls this a real factor operating in the background even if it is not widely discussed.

  “when you boil it all down, they actually risk getting more heat, if they damage the AI trade in a meaningful way.”at 37:35

AI bubble & tech valuations

• Warren Pies · 3Fourteen Research

  Bears are too confident the AI bubble has burst; the AI story still has farther to run.

  Pies finds it striking how confident the bears are that the bubble has already burst, as if it were settled. He pushes back, citing fundamentals: GPU rents rising, tight availability, rising revenue per compute dollar at Anthropic, and cloud revenue ramping at hyperscalers. He says the objective facts show the AI trade is still intact, and he still believes there is farther to go in the AI story. He admits the path could be dangerous even if he is right about the destination.

  “They talk about the bubble as if it's burst already, and it's like a known thing... I still believe that we have farther to go in this AI story.”at 2:37

Credit markets

• Warren Pies · 3Fourteen Research

  Hyperscaler credit spreads above July wides signal AI financing stress and a shift toward breakdown.

  The AI build-out needs a lot of external financing. After the July turbulence, hyperscaler credit spreads widened, then broke back above those July wides before equities had fully corrected. Pies reads that as a warning from the debt market that funding for the build-out is getting harder. It made him change his default assumption from a sustained uptrend to some kind of breakdown, and it is one of the metrics his firm watches for managing risk.

  “seeing those credit spreads jump above the the July levels before equities had corrected all the way back down there tells me, again, our default assumption should be changing from a sustained uptrend to some kind of breakdown.”at 5:47

AI chips & semiconductors

• Warren Pies · 3Fourteen Research

  Semis must regain momentum to lead the market; a Fed-driven slowdown would hit them hardest.

  Across 50-plus S&P 500 industry groups, semis show the biggest spread between up-day and down-day performance, over 105%, and no other group comes close. Semis are about 20% of S&P 500 market cap and the lifeblood of the AI trade, which is why pod shops are long them and short other things. The team feared semis were topping through June and hoped July was the bottom, but the chart has broken down again. Pies says semis need to get back into gear. If the Fed keeps hiking and slows the cycle, semis would suffer most, which would be a nasty path for the group and bearish for the market.

  “But our key key point is that semis have to get into gear.”at 12:15

Venture capital & startups

• Warren Pies · 3Fourteen Research

  VC-funded startups are about 25% of lab revenue, so Fed hikes that choke VC could hit lab ARR.

  Pies's firm estimates that venture-backed startups account for about 25% of AI lab annual recurring revenue. If higher interest rates from Fed hikes killed the VC funding cycle, as in 2022, a large chunk of lab ARR could disappear from that group. He presents this as one of the channels through which interest rate policy could hit the AI trade, alongside the funding of the build-out itself.

  “we estimated that VC startup companies are about 25% of lab ARR. So if we really killed the VC funding cycle with the Fed hikes and higher interest rates, then similar to 2022, then you could see a pretty big chunk of lab ARR come out”at 34:11

US growth & recession risk

• Warren Pies · 3Fourteen Research

  Hard to see a true recession: deficits and AI credit wobble interacting with rates limit downside.

  Pies sees an AI selloff as chopping rather than a trigger for a growth slowdown. The fiscal impulse is fading, but the deficit is still wide. When the AI trade sells off, rates fall and staples and rate-sensitive pockets of the market rally, which he calls a bit overdone but a reflection of economic reality. Taken together with the deficits and how an AI credit wobble would interact with interest rates, he has a hard time getting to a true recessionary scenario.

  “So I think that between the deficits and the way the AI credit wobble would interact with interest rates, I have a hard time getting to, like, a true recessionary scenario in the current environment.”at 39:26

Market sentiment, positioning & volatility

• Warren Pies · 3Fourteen Research

  Oil positioning is squared with no extremes; nobody wants a strong stance before the next move.

  Pies normally trades by waiting for positioning extremes and fading them. He sees no such extreme now. Positioning has been squared, and no one has conviction ahead of the next geopolitical move, so traders are left guessing where Trump is heading. He describes the oil market as a mess with no clear positioning signal.

  “positioning has been totally squared, and no one wants to take a strong stance ahead of the next move.”at 44:35

Renewables & clean energy

• Warren Pies · 3Fourteen Research

  Iran conflict could accelerate the decarbonization transition already underway.

  Pies thinks there is a good chance the oil shock accelerates the decarbonization transition that was already underway. He pairs this with a reduced US presence in the Middle East and around the Gulf Coast as part of the wider multipolar shift. He hedges the point, calling it a likely outcome rather than a certainty.

  “[The Iran conflict] could accelerate, I think, the... decarbonization transition that was already underway.”at 46:06

US global standing & alliances

• Warren Pies · 3Fourteen Research

  Iran war was a strategic miscalculation; the US has lost soft power and its hard power is questioned.

  Pies argues that the US has been policing the Strait of Hormuz, the South China Sea and every other region at once, and the conflict has stretched and exposed those capabilities. To end the war, Washington is now bringing China and Russia into multi-power discussions. He sees this as the multipolar shift accelerating, and he calls the Iran war an objective strategic miscalculation that has cost the US soft power and put its hard power in question.

  “This was a strategic miscalculation to do this from The US's perspective. So I think we've lost soft power. Our hard power is in more question.”at 47:50