168X Talks with Frank Trading: After the AI Trade Ends, Where Is the Next Big Trade?

Summary

• Frank thinks the four-year AI bull market may be nearing its end, but a long-term bearish view does not rule out turning bullish in the short run. He sees OpenAI and Anthropic missing ARR expectations as a turning point for growth and financing confidence. Cloud providers’ free cash flow has turned negative, making it harder to sustain infrastructure spending at the same scale if investment appetite and financing slow. “Don’t fall in love with stocks”: believing AI is an industrial revolution does not mean believing AI stocks can keep rising.
• The easiest leg of the index short may be behind him; from here, smaller positions and key levels matter more. He disclosed building a large short near 31,250 on the Nasdaq and adding after SOXX broke below roughly 555. He now thinks the cleanest part of the decline may be over. He is watching 500, 515 and 530 on SOXX as successive rebound levels; he would need to see it reclaim and hold 515, including during Korean trading hours, before he sees a case for trying a long.
• His reverse-bottleneck trade targets weaker players as order spillovers fade, rather than shorting AI across the board. As large and small players expand capacity at the same time, supply will eventually catch up with demand and customers will return to the leaders. Pair trades such as Nvidia versus AMD are bets on relative strength. “What’s the difference between semiconductors and pigs? None. They’re the same.”
• The November 3 midterm elections are a policy fork; AI should not be assumed to get a reprieve afterward. Frank thinks Democrats “will definitely play the anti-AI card.” If they sweep both chambers, they could obstruct data center construction, while debt ceiling negotiations and efforts to ease regulation of financial firms and big banks could also reverse. Long bonds could be the best trade. If the two parties split control of Congress, he thinks the outcome would be more favorable for stocks.
• His base case for next year is a recessionary bear market, while whether Bitcoin can take the baton remains uncertain. He thinks the US economy could “go straight into recession” if AI capex falls. If oil prices and services inflation do not ease, the outcome could instead be stagflation. If rates fall and financial conditions remain loose, gold and Bitcoin could attract inflows; in a credit crisis, Bitcoin would be unlikely to escape the selloff.
• Among non-AI longs, he has bought India, while the DIA trade has run its short-term course. He thinks India’s fundamentals “haven’t deteriorated,” with strong domestic buying absorbing the shock from oil prices and the rupee. Unwinding Asian long-short trades could also bring in buyers. The Jio IPO and Oil Dollar Window are additional catalysts. DIA was only a valuation-recovery position after long-term yields fell, and he has now exited.
• His new consumer pair is long ROST and short XLY, betting on a shift to lower-cost consumption as the wealth effect fades. He recently opened a new XLY short and named Hilton, RCL and Starbucks: these companies are still near their highs, but a falling stock market could weaken consumers’ spending power. His ROST long, in a discount retailer, assumes households still have purchasing power. On the short side, he said: “If rates go up next week, all these companies are dead.”
• Open source is eroding the scarcity value of model capability, but long-term opportunities may still lie with leaders and on-device AI. His claim that “token billing is inevitably coming to an end” means revenue will shift away from token charges, which may currently account for 80%–90% of the total—not that token billing will disappear altogether. Smaller models and lower costs make him bullish on on-device names such as Qcom and ARM; for true long-term investors, he still sees Nvidia and TSM as leaders trading at attractive valuations.

Deep dive

1. A bearish view on a four-year bull market does not mean sitting out every rally

• Frank began by separating his long-term view from his short- and medium-term trading. He thinks the four-year AI bull market may already be over, but if there is a major positive catalyst from the Fed, Waller, Trump, Iran or within AI, he could still cover shorts and turn long. Trading should not let ego keep you out of a move that runs against your long-term view.
• He pointed to a trade in mid-September: he was already very bearish, but the Nasdaq 100 could not break lower and he thought the September FOMC meeting leaned toward a double cut. He still bought the dip in semiconductor stocks such as SMH and SOXX, rode the rebound, and exited near the highs. In his view, that did not “conflict” with his long-term bearish stance.
• He sees the latest rebound as concentrated short covering, not a trend reversal. “Don’t catch a falling knife”: he does not expect this internally driven selloff to snap straight back in a V-shaped recovery. Buying because prices look strong premarket or intraday is more likely to get you trapped than to catch the bottom.

2. ARR misses first undermine confidence in continued financing

• Frank sees OpenAI and Anthropic both missing ARR expectations as a turning point for AI growth. Early investors who entered below a $300B valuation may not be worried; someone who bought Anthropic in June this year at a valuation of around $1.5T and is now back at breakeven or underwater may be less willing to put in more money when asked to take another allocation. Weaker financing appetite could, in turn, affect future fundraising and expansion.
• He said AI companies’ capex in September was only a little over 40% of August’s and 10%–20% of June’s. His logic is that cloud providers’ free cash flow has long been negative, and without borrowing it is difficult to sustain infrastructure at the same scale, let alone expand it. ARR misses weaken investors’ “confidence and resolve” to keep putting money in.
• The host asked who would backstop Anthropic next month and the related plans, including an OpenAI IPO in 2027. His answer: “No one.” He said that when OpenAI tested the waters with Trump last October, Trump made clear there would be no backstop. There are other AI companies in the market; there is no buyer of last resort obliged to save any one model maker.
• The question still to resolve is whether ARR missed because companies have orders but lack capacity, or because customers are switching to Qwen, DeepSeek or Gemini. He listed paid demand, revenue growth, incremental revenue, enterprise customer model switching, traffic split across multiple models and private deployments as things to track. He also wants to know whether cloud providers are willing to lock in capacity with companies such as Micron early, and whether delivery times are shortening.

3. The beta short shifts to smaller positions; 515 is the threshold for trying a long

• He disclosed building a large short near 31,250 on the Nasdaq and adding again after SOXX broke below roughly 555. The cleanest part of the decline may now be over, while the end of the AI bubble is still not the market consensus. News could whipsaw both sides, so he would rather keep the short small and adjust around technical levels.
• His bar for turning long is for SOXX to reclaim 500 and 515, with 515 the key level to hold, including during Korean trading hours. He sees 500, 515 and 530 as successive rebound levels. Investors who are underwater do not have to wait for all 3 levels to play out: trim some if the rebound fails, and take another step back if a key level breaks, rather than waiting until there is no way out.
• The speed and scale of this selloff were similar to, or even faster than, the one in June and July, yet VXN remains nearly 10 points below its July high. Frank’s read is that many people simply had not bought puts this time; the market plunged because people were actually selling. He saw the June–July pullback as leverage unwinding with fundamentals intact. This time, he thinks fundamentals have deteriorated and many people are selling.
• Asked what would count as a genuine turning point, he said they had already seen one: ARR misses, a nearly 20% drop in SOXX over 2 weeks and a collapse in the Korean market. Bulls should instead ask what turn would make it worth going long, rather than continuing to ask when the market will officially be in a bear market.

4. Cash is an option; protecting gains matters more than proving your conviction

• When he cannot make sense of the market, his response is: “Do I have to make this money?” He mentioned keeping dollars on Robinhood at a 7% annual yield and said he had put quite a bit there himself. “7% a year is not nothing”; there is no need to play every hand.
• He said he was up more than 440% year to date, with virtually no major drawdowns in his account. Some friends were up 10x or 20x in the middle of the year, only to end up flat or negative year to date, and some were wiped out. What he regrets is not that they lacked faith in AI, but that they did not protect the money they had already made.
• The host inferred that he might stay in cash for the next 6–9 months. Frank immediately replied: “I can’t see that far ahead into next year.” What he confirmed was a basic expectation of a recessionary path next year; he was not projecting his current positioning forward as an unchanged allocation for the next 6 months.

5. The reverse-bottleneck trade is about fading scarcity rents

• Frank used pigs to explain the semiconductor cycle: when supply is short, everyone starts raising pigs; when all that supply arrives at once, prices collapse. “What’s the difference between semiconductors and pigs? None. They’re the same.” He does not think supply and demand settle into an exact balance and stay there; capacity expansion ultimately pushes scarcity into oversupply.
• In his view, since 2025 Nvidia and TSM have been among the names that rose less. Intel, AAOI, GSS and TTMI, by contrast, benefited from spillover orders and could charge more because leaders lacked capacity, driving a turn in fundamentals. Once supply catches up, if customers can meet their needs by buying Nvidia, why allocate the extra budget to AMD?
• “When supply catches up with demand, past winners become losers.” He used the Nvidia/AMD and Lumentum/AAOI share-price ratios to illustrate relative strength: Lumentum itself has been plunging, but the ratio has recently risen. He thinks that makes for a reasonable pair trade.
• Models will get smaller and cheaper, and may no longer require as many semiconductors. Even if the US needs a lot of chips, it has no power. Frank expects weaker demand and rising supply to create a mismatch. Since “markets trade expectations,” he is starting the trade now; at minimum, it is an idea to keep in mind, not necessarily to act on.

6. A Democratic sweep of both chambers could favor long bonds over stocks

• As of the November 3 midterm elections, Frank’s estimate was that a Democratic House win was all but certain, while the odds of winning the Senate were a little over 60%. He laid out 3 scenarios: Democrats sweep both chambers, the two parties split them 50–50, or Republicans sweep. He saw very little chance of a Republican sweep and set that aside.
• His policy logic starts with local opposition: residents think data centers take up electricity and bring little positive externality to the area, or even impose a significant negative one; workers worry AI will take their jobs. “The Democrats will definitely play the anti-AI card” while in opposition. If they sweep both chambers, he thinks they could block data center construction at the local-government level, citing jobs, residents’ power bills, land and water.
• The risks extend beyond AI. Debt ceiling negotiations and policies to ease restrictions on financial firms and big banks could also be rolled back. In a Democratic sweep, he thinks long bonds would be the best trade and other sectors could fall. If the two parties split control, he thinks that could instead be a relatively favorable environment for stocks.

7. Even a peace deal would not immediately fix the refining bottleneck

• The host proposed another scenario: what if the Russia–Ukraine war suddenly quieted down or Iran reached a peace deal in the next 2 weeks? Frank disagreed: “If he wanted a deal, he would have made one already.” With less than 2 weeks left, he thinks Trump cannot get it done.
• He distinguished crude from refined products: having some crude exports and enough supply does not resolve gasoline and diesel shortages. Refinery bombings have pushed crack spreads wider, and gasoline and diesel prices are still climbing. Refineries take a long time to recover; even if Iran surrenders, the US surrenders or crude supply returns, it cannot immediately improve fuel availability for households and industry.

8. The base case is recession or stagflation; Bitcoin depends on how investors define it

• Frank described the US as a K-shaped economy and noted that the Sahm Rule was triggered in the third quarter of 2024. In his view, AI capex has been entirely responsible for US economic growth over the past few years, so a drop in capex could mean “straight into recession.” His base case for next year is a recessionary path, perhaps even an “unprecedented recessionary bear market.”
• If a recession leads the Fed to stop raising rates or cut them, gold and Bitcoin could attract hot money, though he said it was “hard to say” how much Bitcoin’s smaller market could absorb. If oil prices and services inflation stay high and the Fed remains hawkish, the outcome could be worse: stagflation. He thinks the odds of a soft landing or “Goldilocks” scenario are very low.
• The host recalled a correlation of around 0.3–0.4 between Bitcoin and the Nasdaq and was unsure whether Bitcoin could outperform on its own. Frank admitted he did not “see it particularly clearly”: stocks face earnings pressure from consecutive misses and lowered expectations, while crypto is more like a game of chips. If big money trades Bitcoin as “digital gold,” a weaker economy and lower rates would reduce the cost of holding it, and Bitcoin could instead enter a bull market.
• An extreme scenario is different: if CoreWeave implodes or goes bankrupt, and Oracle also goes bankrupt, creating something like a financial crisis, “everything will fall, including Bitcoin.” He also clarified that his earlier comment that Bitcoin was “going to zero” was an emotional reaction and “not what I really think”; it should not be taken as his current firm view.

9. The DIA long was just a valuation rebound trade after rates fell

• Asked about the DIA long the host had seen, Frank first clarified that he had already exited. He treated DIA as a position dominated by mature companies: expectations for high AI growth pushed real and nominal rates higher, compressing their valuations. After OpenAI and Anthropic missed ARR expectations, AI growth expectations cracked and US long-term yields fell, creating an opportunity for a valuation rebound.
• He does not see DIA as a safe haven in a recession or stagflation. Big banks hold corporate debt; a substantial share of CAT’s growth comes from AI-related purchases; and GS’s M&A and market-making revenue could also be affected. If AI investment and stock market activity keep cooling, these companies’ earnings will be in trouble too. “DIA was just a short-term punt.”

10. The India long bets on a divergence between fundamentals and flows

• Frank thinks India’s fundamentals have not deteriorated: growth was rapid in the first half, and bank bad-loan ratios are near historic lows. The selloff came from oil shortages and spending dollars on Middle Eastern crude, which weakened the rupee and scared off foreign investors. By his estimate, foreign investors are selling around $50B a month; yet the market is down only in the low teens, not the 30% he had expected, which shows strong domestic buying.
• India is also the long leg in Korean shorts across many Asian long-short trades. If Korea starts to collapse, unwinding those pair trades could bring a rebound in India.
• Another catalyst is Jio, which he said would list “next week.” He called it the biggest IPO in Indian history, at around $100B or more, with very high subscription demand. He described Jio as the company of India’s richest man, with businesses spanning services like Taobao, WeChat, jewelry, video and telecom. He thinks global macro funds will participate; the process of converting into rupees will bring money back into India and lead to investment in other Indian companies.
• He also likes the Oil Dollar Window: India’s central bank uses designated banks to provide several large state-owned oil companies with the dollars they need to buy crude. Fundamentals have not worsened, domestic buying is strong, and these catalysts are in play. That is why he bought India last week. He called India the standout market that week and said he made some money too.

11. Long ROST, short XLY: betting on lower-cost consumption as the wealth effect fades

• He opened a new XLY short over the past 2 days and has also been long ROST. The short positions include Hilton, RCL and Starbucks. This is another way he expresses his view that the K-shaped economy will break down: once the wealth effect from the stock market fades, household spending power will drop sharply and discretionary consumption will come under pressure.
• ROST is a discount apparel retailer in his description. The thesis is not that consumers will be richer, but that people who can still spend may choose cheaper goods.
• He emphasized that Hilton, RCL and Starbucks have barely fallen and remain near, or not far from, their highs. His view: “If rates go up next week, all these companies are dead.” Other short candidates include Ralph Lauren, Airbnb, Booking and Marriott, all extensions of his recession call into consumer-facing names.

12. Open source threatens capability scarcity and the token billing model

• Frank used the spread of technology to explain why closed-source advantages may narrow: Britain invented the steam engine and spinning machine first and kept them tightly guarded, but the gains of the Industrial Revolution still spread to France, Germany and the rest of the world. The gap between open- and closed-source models may widen or narrow at different stages, but he thinks closed-source models’ relative advantage will keep shrinking.
• For tasks that do not involve proving mathematical theorems or deriving formulas no one has studied before, he asked whether people can really feel a difference between DeepSeek, Qwen and Gemini, on the one hand, and Opus 5.5 and Extra 6, on the other. “It’s actually hard.” Open source does not have to beat closed source in every domain; it only needs to handle more everyday tasks at lower cost for customers to have reason to shift workloads.
• He thinks what open source really changes is the “scarcity of capability,” giving large customers more control over model choice, deployment and pricing. Model capabilities are becoming widely available faster than many customers are willing to raise their spending. OpenAI and Anthropic’s development of cheaper models such as Haiku to compete also affects their own profitability.
• “Token billing is inevitably coming to an end” is his strong claim, but he immediately qualified it: token billing will not disappear entirely. Rather, the bulk of reported revenue, perhaps 80%–90% today, will shift substantially toward private deployments and the use of open-source models. He does not think token billing alone can support 2 companies valued above $1T each.

13. He remains bullish on leaders over the long term; the next leg of profits may shift to on-device AI

• He did not end by rejecting AI. Although he does not think LLMs will lead to AGI, he still sees them as one of the most important industrial revolutions in human history. For true long-term investors, he said, there is no need to panic about being underwater in Nvidia or TSM. He thinks both are cheap, especially Nvidia, and recognizes their leadership in GPU design and foundry.
• For Intel, AMD, MU, SK Hynix, AAOI and GFS, he urged investors who are underwater to reconsider whether they can live with the downside. “You can make 10,000%, but it only takes a 100% loss to go to zero.” He is not spreading his optimism about AI’s future evenly across every semiconductor stock.
• On-device AI is another long-term direction he still likes: models will get cheaper and smaller, while privacy and low-latency use cases will support on-device products. He named Qualcomm, ARM and NXP, and thinks the AI value chain’s profit pool could shift from semiconductors to on-device AI. If on-device AI really takes off, companies such as Lattice may not do as well.