Warren Pies: AI selloff is a technical var shock, not a fundamental break

Pies cut equity risk despite staying fundamentally bullish on AI

Warren Pies of 3Fourteen Research describes a framework where he builds conviction on fundamentals but manages risk with technicals. He says his firm downgraded equities at the start of the prior week despite remaining fundamentally bullish on the AI build-out. "I think that there are it's funny to me to hear the bears so confident... I'm not there. I'm pushing against that still," he says of AI bears.

The downgrade was driven by three technical signals: the market entering a "consolidation" (a tight trading range with no new highs or lows, which he says happens about 13% of the time), financials rolling over relative to staples (with staples up about 14% in October), and hyperscaler credit spreads breaking above the wides seen during what he calls the "Leopold liquidation lows" in July. He also cites political context, saying contacts are worried about a "blue tsunami" in the midterms and pressure in places like the Texas state legislature to slow data center build-outs.

Pies says he hasn't seen a genuine sentiment washout yet — speculative/inverse ETF volume and his internal sentiment models haven't hit capitulatory readings, and vol-target positioning is still roughly one standard deviation above the rolling mean, meaning there's room for further selling before a buy signal emerges.

Semiconductors remain the key swing factor for the broader market

Pies says semiconductors have driven the single biggest spread between S&P up-days and down-days of any industry group this year — over 105% between the two equity lines — reflecting how pod shops and multi-strategy funds have used being long semis and short other sectors as their AI exposure trade.

He and colleague Fernando track GPU availability and rental rates daily, including from "Neo Clouds." Availability for Blackwell B200s has been near zero, and rental prices have kept rising rather than falling the modeled 25% annual decline data-center economics assumed. He also notes longer-dated GPU rental contracts trading above spot (contango), which he interprets not as weakness but as a "capacity premium" from users locking in future compute.

Pies sees no evidence of an AI overbuild

Pies points to hyperscaler cloud revenue acceleration and cites revenue per dollar of compute spend rising from 57 cents in Q2 of the prior year to $2.08, dipping slightly in Q3 but staying strong, alongside inference costs falling as a share of compute spend. "The objective facts tell me that the AI trade isn't intact still," he says, pushing back against return-on-investment skepticism.

Using satellite data tracking data-center construction pace, his firm models about 27-plus gigawatts of additional capacity coming online over 2027, which he says aligns with current semiconductor revenue estimates rather than signaling a supply shortfall. He argues the recent simultaneous selloff in both semis and hyperscalers doesn't fit a genuine supply-crunch scenario — in that case hyperscalers should rally even as semis fall — concluding the price action looks more like a "liquidation var shock" than a fundamental shift.

The Fed, not AI fundamentals, is now the main risk

Pies says the correlation driving the 10-year yield has shifted from oil to the Nasdaq, meaning the Fed is now effectively reacting to the AI build-out rather than the Iran-driven oil shock. He notes the market has gone from pricing 75 basis points of cuts before the war to pricing as many as six hikes out through 2027.

"The risk is that the Fed doesn't stop until they've broken something in the AI trade in the funding markets." — Warren Pies

He flags that debt issuance, including CoreWeave's bonds, has become a more important funding channel now that free cash flow alone can't cover the build-out, making credit spreads a key risk gauge. He also estimates VC-funded startups account for roughly 25% of AI lab ARR, meaning higher rates could squeeze that funding source similarly to 2022.

Pies doubts the Fed will let the AI trade break, and sees limited recession risk

Pies argues political and strategic incentives make it unlikely the Fed will deliberately crash the AI build-out, suggesting a credit wobble would likely prompt the Fed to stop hiking or even pivot to cuts fairly quickly. He frames this as part of a broader political dynamic since COVID-era deficits. Still, he calls this dynamic a background factor rather than something openly discussed by policymakers.

On the broader economy, Pies expects more chopping than a sharp downturn, citing a still-wide fiscal deficit. He expects at most one more Fed hike before a pause, which could let the two-year yield and yield curve flatten somewhat, reaccelerating parts of the economy crowded out by the AI build-out.

Oil markets are frozen ahead of the next Iran-related move

Pies says oil positioning is unusually squared, with traders unwilling to commit ahead of the next geopolitical move. He lays out four possible US paths on Iran — discriminate bombing (already done), indiscriminate bombing (threatened but not executed), ground invasion (no political appetite), and the current blockade strategy, which he believes is working by reducing Iranian floating storage. He thinks a negotiated deal is more likely after the midterms than before, and notes China appears to be leaning on Iran as it tires of drawing down its own inventories.

He estimates over 60% of the global VLCC fleet was laden at last check, calling shipping "totally unbalanced," with Brent-WTI spreads and crack spreads also blown out. Pies maintains a general stance of staying overweight energy within a balanced portfolio as a hedge against this uncertainty.

Diesel crack spreads reflect genuine structural tightness, not just cyclical strength

Pies says blown-out middle distillate crack spreads, including Singapore gas oil cracks above $70 a barrel, typically resolve through refiners retooling output and demand destruction, which he believes is already occurring at diesel prices above $200 a barrel sustained for months. He compares the current period to the early shale era, when a crude export ban (while refined products could still be exported) created unusually durable high crack spreads for refiners.

He expects a mix of both forces — some demand destruction and refiners shifting capacity toward middle distillates — to gradually narrow the spreads, while oil prices themselves stay "sticky" and possibly move higher. He also suggests the conflict could accelerate a longer-term shift toward reduced US presence in the Middle East and a more "multipolar" global energy and security order, calling the intervention "objectively... a strategic miscalculation" from a US soft-power standpoint.