Jeremy Raper: Cycles, AI, and Activism

Episode Summary

Executive Summary: The conversation centered on two major themes: why AI/semiconductor stocks still behave like cyclical businesses despite structural changes, and how AI is transforming special-situation investing. Jeremy argued that peak earnings and valuation excesses in semis/memory create huge downside when cycles normalize. He also detailed how AI slashes legal/document work, expands investing leverage, and may compress market inefficiencies globally, especially in Australia and other less-trafficked markets.

Main Topics: AI semiconductors remain cyclical (Priority: 5/5): Jeremy argues that even if memory/semis are structurally better businesses than in past cycles, the market is still pricing peak earnings and ignoring the eventual collapse from extraordinary to merely good profitability. Valuation and cycle timing in memory stocks (Priority: 5/5): The discussion focused on how to value memory names when EBITDA/free cash flow can spike dramatically, and why price/book, earnings, and cash flow all become dangerous if investors underwrite peak conditions too long. AI as a force multiplier for special situations (Priority: 5/5): Jeremy explained that AI is already improving his work in activism, litigation, and corporate control situations by reducing document review and drafting costs from hundreds of thousands of dollars to near zero. AI's impact on market structure and geography (Priority: 4/5): He argued AI may reduce the advantage of developed markets by democratizing access to expertise, enabling cheaper corporate formation and activism globally, and potentially compressing DM/EM valuation gaps. Australia as a fertile special-situations market (Priority: 4/5): Jeremy described Australia as unusually rich in small-cap mispricings due to market structure, active capital scarcity, and the high friction of legal/regulatory value extraction, which AI can now reduce. Idea generation vs. idea execution with AI (Priority: 4/5): He said AI is currently weak at screening/finding ideas from scratch, but excellent at accelerating diligence, interpretation, and execution once a candidate situation is identified. Case studies in activism and event-driven investing (Priority: 4/5): The transcript included examples such as EDU Holdings, European Lithium/CRML, Domo, and prior Australian takeover-panel work to illustrate how tight timing and control of process can unlock value.

Key Arguments: Memory and semiconductor businesses may be structurally stronger than before, but trading them as if peak margins are permanent is a mistake. The market is not reacting to current earnings declines; it is repricing the second and third derivatives of demand, capex, and supply response. Peak-cycle valuations can collapse even when businesses remain good; the move from 150% ROE to 40% ROE can still destroy the stock. AI is most useful in document-heavy, legalistic, special-situation work because it can review, summarize, and draft in the required legal vernacular. AI lowers the frictional cost of activism and corporate governance challenges, making cash-box, delisting, and minority-holder situations easier to attack. AI may level global access to expertise, reducing the advantage of developed markets over emerging markets and increasing the attractiveness of cheap, overlooked jurisdictions. AI is not yet reliable enough for autonomous quantitative screening or for pulling clean financial data without strong user direction and trusted datasets. In markets like Australia, structural inefficiencies plus AI-driven execution speed could make small- and mid-cap special situations far more actionable. The best way to think about cyclical investments at peak earnings is like valuing an option on the duration of the boom plus a separate estimate of normalized post-cycle earnings.

Data Points: Micron long-term ROE: about 10% - Used to illustrate that long-term returns in memory can approximate the cost of equity, supporting a cyclical/commodity-like framework. Peak ROE mentioned for memory companies: 100% to 150% ROE - Jeremy described current run rates as extraordinary peak-cycle earnings that are likely unsustainable. SanDisk EBITDA (2024): $800 million to $900 million - Referenced as the low base before the boom in memory earnings. SanDisk EBITDA (peak/current consensus): $10 billion realized; $35 billion to $40 billion consensus for next two years - Illustrates how aggressively the market was extrapolating peak earnings. SanDisk EV: $250 billion - Used to show how stretched valuation became relative to potentially normal earnings. Micron price/book: 8.5x - Used to argue that even supposedly structurally better memory businesses were still being priced richly. Memory stock multiple mentioned: ~10x current free cash flow - Cited as evidence of extreme optimism despite cyclical risk. Hynix and related names: down materially after weak print - Example of market reaction as cycle concerns surfaced. DRAM ETF move: down 42% since June 18 - Used as an example of how quickly the memory trade reversed. Cyber semiconductor stock move: from 4.5-5 to 110, then back to 26 - Illustrates bubble-like volatility and rapid round-trips in semis-related stocks. AI legal campaign cost reduction: from $300,000-$350,000 to near $0 - Compared prior Takeovers Panel/legal work to current AI-assisted workflow. Domo cash balance after sale: ~$485 million net cash - Used as a cash-box/special-situation example. Domo share price discussed: $3.50 - Pointed out as trading at a deep discount to implied cash value. Australia special situation stakes: 5.7% and 10% ownership examples - Used to explain why position sizing and ownership thresholds matter in activism. EDU Holdings market cap: $15 million to $20 million - Example of a microcap that turned profitable and then pursued a delisting/buyback process. EDU share price move: from 5 cents to 15 cents; buyback at 30% premium; referenced later at about $1 - Shows the magnitude of the special-situation payoff and subsequent rerating. European Lithium market cap: $475 million AUD - Listed as the starting point for a complex merger/cash-box situation. European Lithium net cash per share: 18 cents per share - Used to argue the company’s hard asset backing versus current deal terms. European Lithium implied deal consideration: 58 cents then 26-27 cents per share - Deal value fell as CRML’s share price declined, making the transaction unattractive. CRML stock price move: from about $11-$12 to around $5 - Key driver of the deterioration in the implied merger consideration. Australian delisting vote threshold: 75% - Explained as the supermajority required to delist and why this creates activism leverage. Takeovers Panel legal spend: $300,000-$350,000 over six months - Prior extrajudicial proceeding cost before AI reduced the same type of work to near zero.

Pivotal Quotes: "The point was not that this is not post-cyclical or pre-cyclical or still cyclical or not, right?" — Jeremy: He clarifies that the debate is not whether semis are cyclical in some abstract sense, but how investors should trade them through the cycle. "I can do essentially all of it with AI, and that enables me to do more work, to find more opportunities." — Jeremy: Describing how AI transforms legal/document-heavy special situations and increases his investing leverage. "This is a gold rush. This is a temporary gold rush. Capacity will come online. It always does." — Jeremy: His core warning that AI/semiconductor enthusiasm is real but likely temporary and cyclical.

Implications: Listeners should expect more volatility in AI/semis as peak-cycle expectations unwind, while special-situation investors may gain a major edge from AI-assisted legal, regulatory, and cross-border work—especially in inefficient markets like Australia.