ELITE Strategist: How to position for the NEXT AI trade

Summary

• Warren Pies has flipped back to overweight equities after a brief August downgrade, targeting S&P 500 8,000 “within the next few months.” He cut exposure on August 10 at 7,753 fearing an unpriced September Fed hike; with the hike delivered and guidance he reads as “two hikes, followed by a long pause,” he sees a 1997 analog — one hike, pause, cuts in 1998 — where the market fell 6% after the first hike and then rose sharply. “We could be wrong, but that’s how we interpreted their statements in September.”
• The bull market lives or dies with AI, and semiconductors are its heart. Semis are nearly 20% of total market cap, and across ~60 S&P industries the spread between index performance on semi-up versus semi-down days is 90% — the widest anywhere. His key risk trigger: semis breaking back below July lows would make him “question our preference for stocks.” “No amount of Exxon will save this market.”
• His real-time compute metrics — GPU availability, rental rates, lab ARR — are “very optimistic,” the most bullish input he tracks. Blackwell is unobtainable on demand, B200 rental rates are up 30% this year against data-center models that assume annual declines, and availability leads rental rates. He thinks hyperscaler cloud estimates are too low and “the right tail of the distribution is underestimated… I don’t want to miss that.”
• The AI ROI debate is over, per Pies. After Microsoft, Amazon, and even Google’s reports, “the issue is closed. This discussion is over… you should come up with a new pessimistic scenario.”
• The one channel through which the Fed could break the AI trade is venture funding. The 2022 hiking cycle cut 12-month VC funding by almost 70%; today’s $500B trailing total ($290B ex-labs) bankrolls AI-native startups that are the labs’ super users, and in a worst case ~25% of labs’ ARR is at risk in a repeat — “you would see a market sell-off” if newly public labs missed ARR targets.
• AI capex — roughly 3% of nominal GDP annually — is raising the economy’s “speed limit,” i.e., the neutral rate. The SEP lifted the 2028 rate forecast 50bps to 3.8%, futures swung from three cuts pre-Iran-war to four hikes with oil stuck near $100/barrel, and “there has never before been a period when the Fed has cut rates amid a capital spending cycle.”
• What flips him bearish: a full 2022-style hiking cycle and political backlash against AI. He worries the market’s calm reaction “emboldens” the Fed and that “deep down Kevin Warsh wants to be a hawk”; meanwhile data-center opposition is going bipartisan into midterms and lab leaders’ messaging is “bad marketing” — the fix is pivoting from “we will destroy all jobs” to “we will cure cancer.” Portfolio: hyperscalers first, semis alongside, healthcare and an energy hedge; his models trade under tickers RAA and FCTE.

Deep dive

1. The whipsaw: out at 7,753 on Fed risk, back in for earnings and 8,000

• Pies walks Phil Rosen through the year’s positioning: optimism into January, shaken by the Iran war between February and April, then brought back into the market “in full force” by the Mythos model leak in early-mid April — “we really thought AI would be the leader in this market.” He downgraded on August 10 with the S&P at 7,753, judging the odds of a September Fed hike higher than the market priced.
• The hike came, and his read of the message — two hikes then a long pause — maps onto 1997: one hike, a pause, then a 1998 cut, with the market down 6% after the first hike before surging. “I think the reaction will be more like the 1997 cycle than some traditional rate hike cycle.”
• The re-entry recipe: a catalyst — “perhaps new model releases before Anthropic’s IPO” — that re-rates hyperscalers alongside semiconductors, with the 10-year bond yield rising to 5.1–5.15. Narrow leadership is fine: “the market can grow even with narrow leadership, given how it is currently structured,” and that mix “could take us to the 8,000 mark for the S&P 500 within the next few months.”

2. Why macro risk has shrunk — and why low correlations still make it dangerous

• The September SEP calmed him: the Fed penciled core PCE at 3.4% by year-end, but a pending data revision (portfolio-management fees being the main overstated component) should “cut a few tenths” — so the Fed set itself a high bar. Unemployment is forecast flat at 4.1% through 2028, giving cover to pause on any labor wobble. Going beyond two hikes “would require inflation to reach a very high level, which I think is unlikely.”
• Remaining risks: Warsh himself — “such a slippery guy… I still don’t think his communication was anything special” — and midterms, where semis have already traded on Democrats’ Senate odds and “opposition to data centers is real, it’s really starting to get bipartisan support.”
• The mechanism behind his August caution: implied correlations near historical lows suppress index volatility, so a single macro factor can make previously unsynchronized stocks fall together. That’s exactly the setup coming out of reporting season, when investors rotate from micro to “the Fed, the economy, and more general topics.”

3. A “short funding” market where semiconductors are the heart of the trade

• Pies’s structural read: funds are long the AI-development basket — semis, equipment, and industrial names such as CAT — funded by shorts in financials, consumer staples, utilities, and software. The one leg he questions: long semis, short hyperscalers, “which I don’t think makes much sense.”
• His centrality evidence: semis are almost 20% of market cap, and across ~60 industries the spread in S&P performance between semi-up and semi-down days is 90% this year — no other industry comes close. “Semiconductors are the heart of the AI trade… they cannot fall apart if we want the market to truly reach stable new highs.”
• His contrarian take on breadth: the bull market’s rhythm is semis/AI leading and the equal-weight index catching up in bursts; his rule of thumb is the equal-weight must confirm any S&P high within three months. But “I really don’t think this market expansion will be able to take us to new significant heights. We rely completely on the AI trend.”

4. GPU availability, rental rates, and lab ARR say the right tail is underpriced

• The metric’s origin was skepticism: since 2023 his colleague Fernando has been reaching out to neoclouds and pay-as-you-go cloud providers to ask whether a GPU can be obtained within a day, indexed across A100, H100, H200, and B200. Availability determines rental rates — “availability determines rental rates” — and the third leg, lab ARR, matters because it’s “how external money comes into the ecosystem” against circular-financing worries.
• Today’s readings are “very optimistic”: you can’t get Blackwell on demand. H200 availability increased last month, but high rental rates for older chips pushed inference economics toward Blackwell; availability of the older chips is now dropping as well. B200 rental rates are up 30% this year against models assuming annual declines. “The cloud revenue estimates for hyperscalers are underestimated.”
• On the tired “where’s my productivity?” question, his answer is rising profit per worker in knowledge-based fields, rising revenue per employee in software, and rising profit per employee in finance — and on ROI, the last reporting season was “decisive”: after Microsoft, Amazon, and Google, “the issue is closed. This discussion is over. In my opinion, you should come up with a new pessimistic scenario.”

5. The Fed’s real channel into AI: venture funding and a higher neutral rate

• Rosen’s sharpest question — doesn’t a hiking cycle choke the startups that generate frontier-lab revenue? Pies concedes the common line that “it would take more than 25 basis points to stop trillions of dollars of data center construction” is “mostly true, but we wanted to explore this question further.” The 2022 cycle cut 12-month venture funding by almost 70%; today’s trailing total is ~half a trillion, ~$290B ex-labs, and the top 1% of AI spenders outspend the next 10% by orders of magnitude. Worst case: ~25% of labs’ ARR at risk, which “would be a huge obstacle to investment in AI” — so the AI story “is not completely independent of the Fed.”
• His capex-versus-Fed chart: futures swung from three cuts pre-Iran-war to four hikes, oil is “stuck at around $100 per barrel,” and the SEP raised the 2028 rate forecast 50bps to 3.8% — the Fed admitting neutral is higher. AI, “like a deus ex machina,” has companies that used to buy back shares now borrowing and injecting ~3% of nominal GDP annually into the economy, raising its “speed limit.” “There has never before been a period when the Fed has cut rates amid a capital spending cycle.”

6. What breaks the bull — and how Pies is positioned

• Two things flip him bearish: a full 2022-repeat hiking cycle — he worries the market’s calm post-meeting reaction “emboldens the Fed” and that “deep down Kevin Warsh wants to be a hawk, even though he was supposedly selected to lower rates” — and political backlash against AI itself.
• His unvarnished take on the labs’ image problem: Dario Amodei “only makes it worse, he buries himself even deeper,” the average person “has already made their verdict that they don’t trust Sam Altman,” and Anthropic’s effective-altruist AI-supervisor team is “bad marketing… sometimes you look at this and think that maybe it would be better to have a situation like in The Terminator than to give these guys control over your life.” The fix — and part of his healthcare bull case — is pivoting the rhetoric from “we will destroy all jobs” to “we will cure cancer.”
• Positioning: “I’m more of a hyperscaler right now. I like big tech owners of computing power,” with semis alongside, neo-clouds only “if you want to add spice,” healthcare as the ex-tech sector, and an energy overweight as “a great hedge.” His firm is 314research.com; funds tracking his models trade as RAA and FCTE.