Why Warren Pies Cut Equity Risk While Staying Bullish on AI

The show opened with its standard disclaimer that nothing said on Forward Guidance is a recommendation to buy or sell any investments or products. The host then introduced Warren Pies, founder of 3Fourteen Research and a repeat guest, and noted that Pies had recently published reports in which he meaningfully changed his core allocation weightings.

The host described Pies as someone who had generally faded market noise and hysteria over the last few years, and framed the discussion around his combination of a fundamental outlook and a technical outlook, along with the Iran war and resulting energy crisis, the AI buildout, and how the Fed was reacting to it.

Pies said he has followed a principle since his Ned Davis days: he builds conviction on fundamentals but manages risk with technicals. He said that when the market begins moving against a view, an investor needs the humility to shift gears, and that the two can be kept separate.

He said that since late 2023, after Waller signaled the Fed would begin easing, his equity stance had been a bullish overweight tilt. Some disinflation tailwinds from earlier in that period were, in his view, coming to an end, but he remained bullish on the AI story. He said it was striking to hear bears speak so confidently about an AI bubble as if it had already burst.

He acknowledged he could be right about where the AI story is going and still be wrong about the path, which he described as potentially very dangerous in this new technological era. That is why he said he trusts technicals and steps back when they signal a change.

Pies explained that the downgrade at the start of the previous week was driven by several signals. The first was that the market entered a consolidation, which he defined as a defined period of tight trading range with no new highs or lows, occurring near all-time or 52-week highs. He said such consolidations occur about 13% of the time, and that his team looks for a six or seven percent trading range over about 63 days, along with other requirements. Once a market has entered one, he said, it has lost momentum, and a trend change is likely either a breakout or a breakdown.

Within the consolidation, Pies said leadership trends are a key way to foretell how it will resolve. He uses the financials versus staples ratio, which his team relied on before the Liberation Day selloff and earlier this year, and in each case financials rolled over relative to staples after the market entered consolidation. He said the same pattern is appearing again. Staples rose 14% in October as money rotated aggressively into non-AI areas, which he described as unhealthy, noting that a market led by such groups has historically not supported a large bull advance.

He also pointed to hyperscaler credit spreads. He said those spreads broke back above their July levels before equities had fully corrected, which told him the default assumption should shift from a sustained uptrend to some kind of breakdown.

Pies said the midterms were that day and that he had risk-off reasons beyond the technicals. He said he has contacts plugged into the political side who sound very concerned about a possible blue tsunami, and he cited pressure on the Texas state legislature to slow the pace of data center buildouts. He said no one wants to put capital to work in that environment, and that October's tax-related selling by mutual funds and other issues had helped create the weak market environment.

On sentiment, he said his team would want to see a true sentiment or positioning washout before buying further weakness. They watch speculative or inverse ETF volume relative to total speculative volume, which has not shown a capitulation reading, and their internal sentiment models have not reached extreme pessimism. He said their modeled Vault target positioning was still about one standard deviation above the rolling mean, leaving room to cut exposure. His conclusion was that the market offered neither a breakout to buy nor extreme fear to buy, which is why he neutralized risk.

The host said market structure has changed a lot in recent years, with many systematic flows moving into markets. He noted that consumer staples and energy were leading, which he called an unhealthy situation, while Vault-targeting funds, options activity that appeared to be moving only up, and passive index investing were growing more powerful. He observed that top-line indices were still near all-time highs, while many assets further down the long tail had been struggling for months, and asked how Pies finds signal in such a noisy environment.

Pies said his team analyzes year-to-date performance of the S&P 500 on up and down days for each of roughly 50 industry groups. By far the most important, he said, was semiconductors. He explained that when semiconductors are up and an investor owns the S&P 500 on those days, and when they are down and the investor does not, the distance between those two equity lines is over 105% at this point of the year. He said no other industry has a spread that wide.

He said semiconductors now make up about 20% of S&P 500 market cap and are the lifeblood of the AI trade. Pod shops and multistrategy funds, he said, have been going long the AI trade and shorting something else. He joked that if someone wanted to know what Hershey's was doing this year, they could look at SMH, because it has been moving opposite to it for no clear reason.

He said his team has written about this all year and has pounded the semiconductor group in its macro and fundamental indicators. For years they have tracked GPU availability and rental rates to judge whether the compute market is loosening, weak, or tightening. Their key point was that semis needed to get back into gear. They worried coming through June that semis were topping and hoped July would mark the bottom, but the charts broke down instead.

Pies said getting good data is very difficult and that some faith in how the technology will diffuse is always built into projections. He said his firm's philosophy from day one has been to stay neutral, because the debate has two strong sides, bears and bulls, and it feels extremely emotional. To avoid being caught in that trap, they monitor key metrics.

The first metric, he said, was GPU availability, which his team has tracked since 2023. A colleague goes out each hour to see how easily a GPU can be obtained from neoclouds. Early this year, after the agentic explosion, availability fell sharply and has remained low through the year. B200 availability has been zero, which he said tends to lead rental prices, and rental prices have continued to creep higher.

He reminded listeners that when data center economics were modeled, a key assumption was that GPU rents would decline over time, possibly about 25% a year. Pies said that has not happened. As he put it: "Our core assumption was a 25% annual decline in rents, and we haven't seen that. Whenever we track these real-time daily things, we see a very tight market for GPUs for compute emerging now."

Pies said Silicon Data and other firms have begun tracking term structure, including commitments one year out. The data show that longer-dated rental prices are now above spot, which puts the curve in contango. He acknowledged that many people misread contango as a weak sign, which it is in the oil market, where it can signal inventories building. In his view, the GPU market is different: he described a convenience yield in physical commodities and a capacity premium emerging for compute.

Pies interpreted the further-dated futures prices as people locking in capacity for the longer term. He said the hyperscalers' cloud revenues were ramping up, as seen in the Q2 reporting season, and that all the hyperscalers had rallied. He argued the return-on-investment debate was being ground down and that the returns are beginning to emerge. He also called the fundamentals in the Anthropic S1 pretty impressive.

The host said the emotional nature of the discussion stems partly from people misreading the fundamental use cases and value of these tools. Nobody, he suggested, is seriously debating whether the value is there, which he called absurd in its size. What may be more open to debate, he said, is who is financing the buildout and how long the runway lasts.

The host pointed out that one reason the recent hyperscaler capex estimates were missed may be supply crunches in other parts of the buildout that make it hard to actually build. He also noted the midterms and a Texas governorship race, arguing that the winner would matter a great deal for how many data centers can be built. Pies replied that it is hard to disentangle supply limits from demand, pointing to capital market limitations as the Fed hikes.

Pies said earlier this year the common pattern was semis up and hyperscalers down, and that his theory was these would converge because they are part of one trade. Instead, both are now selling off. He said a supply worry would not explain that, because owners of data center capacity ahead of a supply crunch should see strong economics continuing, and restricting capex would lead to great returns for compute providers while their capital requirements fall.

He concluded that the price action looks panicky and does not fit a supply-crunch scenario. In that scenario, he said, he would want to own hyperscalers, and they would become more attractive. He said the price action more likely reflects a market that is liquidating and degrossing, and that if it is a liquidation Varock-type event, the response goes back to his technical overlay, managing risk, and maintaining fundamental conviction.

The host framed the discussion around the economic impact of the buildout, describing an investment boom with many financial assets being issued to fund it, and asked how Pies distinguishes what drives economic growth from the rest of the economy, and how the Fed navigates that. Pies said that in his framework, macro is the intermediate-term risk, and the situation had become more bearish. Before the war, he said, the Fed was still in a cut cycle, but the war quickly shifted policy into a hike cycle.

Pies said the rate drivers had also changed. Treasury yields used to track oil, but he said they now track the Nasdaq more closely. He said the 10-year yield and the stock market have stayed more than 90% correlated throughout. He also said that for the first time ever, the Fed was cutting rates through last year as capex was ramping, and that this could be pointed to as a policy mistake, though he did not say so definitively. As he put it: "What's moving the Fed? They were focused on oil. Their speech is focused on oil, and now they're focused on AI and the buildout. And that's the big change."

Pies said the AI buildout had been funded by free cash flow, which made it hard for the Fed to dent it. Now that external funding markets are involved, he said, the Fed can limit the supply side through funding costs. He said this becomes a real risk if the Fed keeps hiking and inflation data does not calm down, and that he expects the cycle to slow at some point. He said this would not be a death blow to the AI trade, but would be a nasty path, especially for semiconductors, which he said lean bearish.

He also described a second channel. His team estimated that venture-backed startups account for about 25% of lab annual recurring revenue, so a squeeze on VC funding, like the one seen in 2022, could hit that revenue. He said he was rethinking how interest rate policy interacts with the AI buildout, having earlier believed that the Fed could not dent it.

Pies said he thinks there is something to the idea that the Fed will not tighten too much, because he does not believe the Fed will kill the AI trade. He said the market has already shown instability and minor credit stress, and that Fed Chair Powell said he was unable to say financial conditions were tight, which Pies read as a wobble from his Jackson Hole speech. He said another credit wobble could end the hiking, and cuts could follow quickly.

He also said the Fed is in a tough spot, hiking now in response to political dissatisfaction that has built since COVID and deficit spending, along with the inflation story and the Iran war. Pies argued that if the Fed damages the AI trade in a meaningful way, it would face more political heat, and he said he thinks this is a real factor operating in the background.

Asked how sensitive the economy is to an AI wobble, Pies said he sees the outcome as more chopping than recession. He said a fiscal impulse is fading, but deficits remain quite wide. When the AI trade sells off, rates come down and money flows into staples and rate-sensitive pockets, which he said is a bit overdone but reflects some economic reality.

Pies said his view is that the Fed has one more hike in it and then will likely hold a hawkish bias. If that bias goes away and the Fed says it is done, the two-year yield could fall and the curve could flatten, reopening parts of the economy crowded out by the AI buildout. Given the deficits and the way an AI credit wobble would interact with rates, he said he has a hard time reaching a true recessionary scenario in the current environment.

Pies said he expected China to eventually lean on Iran for crude and that he is now hearing that chatter. He said China is tired of drawing down its own internal inventories. He described the shipping as unbalanced, saying more than 60% of the VLCCs in the global fleet were laid up when he last checked. He said crack spreads are blown out and that the diesel market is a mess, with middle distillates in the same condition, which he said is why Brent and WTI are also dislocated.

He said positioning has become squared, with no one wanting to take a strong stance ahead of the next move. He argued that investors should keep an overweight in energy within a balanced portfolio because there is no other way to hedge the risk that the situation goes sideways. He said the next key deadline was roughly 45 days away.

The host asked why distillate crack spreads had not tightened, given that refiners appear extremely profitable. Pies said these spreads typically resolve when refiners retool to produce more of the product and supply comes on, and when some demand destruction occurs. He said diesel has been above $200 a barrel for months, which is causing some demand destruction, though it takes time for that to work through the system. Users can switch between products such as heating oil, creating some switching ability.

Pies said high crack spreads persist for a sustained period only when there are structural changes. He cited the early shale-era crude export ban, which kept refined product exports open, as an example of such an oddity. Singapore gas oil cracks were above $70 a barrel at last check. He said he expects oil prices to remain sticky and not collapse, because refiners will bid up crude. He described refining as the most cyclical business in the world and cited former Exxon CEO Lee Raymond's view that it is a bad business.