Has Memory Bottomed?

Summary

• Avi is all-in on the memory trade again. A week ago he flagged re-entry with memory off 30–40% from highs and said another 10–15% down would trigger him; it dipped 10% and he’s positioned — “at DRAM at 59, at Intel at 105, these things are going back to the highs.” His math: “another 2 to 3x before we get something stupid” against maybe -30% downside, so even at 50/50 odds “you take that trade. It’s expected positive value” — though “obviously, this is not financial advice. Obviously, I could be wrong.”
• The Soros lens from Alchemy of Finance frames the whole call: “when a bubble pops, if it doesn’t die, it actually comes back bigger the second time.” Memory just took a 40% drawdown while OpenAI announced more compute spend and Kimmy — the model that rocked the market — ran out of compute even with a distilled version. “It’s not over.”
• Jonah’s “shocking shocking chart”: only 2.2% of US households pay for AI subscriptions. His map is e-commerce, which was ~2% of US retail in 2004–05 when he wondered how high Amazon could go — it’s ~17% now and he asked whether Amazon had roughly 100x’d. He thinks paid penetration reaches 50% of US and eventually global households, flowing to Micron, MSFT, Meta, and Google. “It’s still inning number one.”
• Jonah said COVID-era leaders printed ~$30 trillion; Avi broadened the point: “everybody’s a populist now,” and equities became the inflation escape — then “we got ridiculously lucky” as AI, a deflationary productivity technology, arrived exactly when needed. Unlike 1999, there’s no hardware roadblock, only compute and a mental block, so adoption runs faster than the internet’s.
• Korea’s margin wipeout is the bubble tell: $26B of margin loans, 1.2 million accounts hitting margin calls when SKX and Samsung fell, ~300K liquidated, ~$1.5–2B of forced selling. Expect Bitcoin-2017-style fakeouts — rally 2x, drop 40%, repeat — so either hold until earnings actually miss or trade around a small slice, as they did trimming when Intel hit 125–130.
• With a lot of Millennium and Citadel PMs watching memory, Avi is hunting where nobody’s looking: the US-Saudi nuclear deal and uranium’s downstream — reactor operators and enrichers like Centress (likely Centrus), which “kind of looks nice” though he holds no position yet. Jonah’s arms-race read on US-China models: “nobody wins an arms race except the arms makers” — bullish components for a long time.
• On Iran, both see contained market risk: east-west pipelines around the Gulf mean “Iran used its leverage point but now it’s going to lose its leverage point forever over the next 5 years,” which is why oil isn’t ripping. Jonah expects the skirmish to grind on until regime change, but “markets wise, we’re safe.”
• Gold at 4,000 is Avi’s best-looking chart — six weeks of defended support, buy with a tight stop or wait for 4,300, target ~4,800 — because gold is a reflexive, flows-driven asset where TA actually works. Crypto “may have bottomed out temporarily” with Clarity Act odds ticking up; Avi’s closing rule: “shit can crash before it rips.”

Deep dive

1. The Bleecker Street lesson: save first, compound later

• Avi’s investment parable, as told: his 90-year-old landlord Archie Tarpinian — no email, no management company — owned “basically like two entire blocks of the West Village.” How? “Jonah, I was a janitor… I just saved my money” and bought blocks of Bleecker Street around the Depression era, “what now costs $20 million back then cost like five grand — and he did it with leverage.” The echo: “previous generations will look back on those of us who didn’t buy Bitcoin.”
• Jonah’s advice for new graduates: spend as little as possible for the first five years and plow savings into markets — “if you have a little bit of money at 24, it literally doesn’t matter. But if you have a lot of money at 29 to 35, you’re great.” Jonah concurs from experience: money he spent from 22 to 27 “actually just kind of didn’t matter.”

2. AI arrived exactly when the money-printing bill came due

• Jonah said COVID-era leaders printed ~$30 trillion; Avi broadened the macro framing: every bubble in history — dot-com, railroads — allocated capital faster than productivity could absorb it. “Everybody’s a populist now,” politicians won’t stop spending, and equities became the escape hatch from inflation. Then “we got ridiculously lucky”: AI, “the most transformative technology since the internet” and potentially deflationary, landed right as inflation returned inside a possible equities mega bubble. “It’s kind of bailing us out right now.”
• Why adoption outruns the internet’s — the point the bears miss: 1999 had a physical roadblock (a large chunk of society literally had no computer or connection), while AI’s only constraints are compute, “which we’re building a ton of,” and a mental block. “You download it onto your phone… the entire world already has it.”

3. All-in on memory: 2–3x upside against -30% downside

• The call, timestamped: a week ago Avi flagged re-entry with memory off 30–40% and said another 10–15% down would make him all-in — it dipped 10%, and he is. “At DRAM at 59, at Intel at 105, these things are going back to the highs” — with the hedge intact: “obviously I could be wrong,” but “the probability that the AI bubble is over is quite low.”
• The expected-value math: “another 2 to 3x before we get something stupid” versus maybe -30% down. “Even if you think it’s a 50/50 shot, you take that trade… it’s expected positive value.”
• The Soros model: “when a bubble pops, if it doesn’t die, it actually comes back bigger the second time.” Memory just drew down 40%, yet OpenAI announced more compute allocation the same day and Kimmy — the model that rocked the market — ran out of compute despite a distilled smaller version. “It’s not over.”
• Portfolio mechanics: he’s repositioned for this leg but will scale out into strength — “I’m taking them down, I’m taking them down” — recycling profits into the next stage: biotech (which barely moved in the whole drawdown) and energy.

4. Only 2.2% of US households pay for AI — inning one

• Jonah’s “shocking shocking chart”: paid AI subscriptions sit at 2.2% of US households. His map is FRED’s e-commerce data — ~2% of retail in 2004–05 when he first asked how high Amazon could go, ~17% now, and he asked whether Amazon had roughly 100x’d. Even Avi was floored: “e-commerce is not even 20% of total sales in the US… I would have been so far off” — though Avi kept the hedge: “I might be full of shit… I need to double check my priors.”
• Avi’s thesis: he doesn’t think the free tier is where most token consumption comes from — “I think you’re going to have to pay for a subscription in order for it to be truly useful” — and he thinks penetration could reach 50% of US, eventually global, households, flowing through to Micron, MSFT, Meta, and the hyperscalers. “It’s still inning number one.”
• Avi’s edge argument: crypto-damaged retail who can stomach a 30% drawdown gain an advantage over financial incumbents and most investors who panic-sell or get force-stopped at 5–10%, plus “lettuce hands… unbattle-tested retail.”

5. Korea’s margin wipeout is the bubble tell — trade it like Bitcoin 2017

• The tape: $26 billion of margin loans in the Korean market; when SKX and Samsung fell, 1.2 million accounts hit margin calls, 300K were liquidated, 2.3 trillion won ($1.5–2B) of forced selling. Avi’s read: mass leverage means people “feel reasonably safe taking these bets… they’re not as terrified as they should be” — classic bubble psychology, but sitting on a real productivity engine.
• That mix produces fakeouts “like the way Bitcoin traded in 2017” — rally 2x, off 40%, repeat. The playbook: hold until earnings actually miss (“cleanse and expunge your mind of your peak net worth”), or trade around a small slice — as they did warning when Intel crossed 120 toward 130 and when Micron gave back an 11% post-earnings pop in a day.

6. Nobody wins the arms race except the arms makers

• On the US-China model fight: Jonah said it seemed like Chinese labs were “basically at the frontier or slightly behind it,” Beijing is pouring in funding and reshored what he thought was Z.AI from Singapore to the mainland for “a stranglehold on the technology.” Jonah’s conclusion: “nobody wins an arms race except the arms makers” — structurally bullish components for a long time, but hard to express: no Anthropic or OpenAI shares outside insane private valuations, and Micron is “crazy volatile, probably low Sharpe but high return on capital if you can stick with it.”
• Avi’s counter-move — look where PMs aren’t: “everybody’s watching AI stocks… the whole world is watching memory” at Millennium and Citadel. So he’s digging into the new US-Saudi nuclear deal and uranium’s downstream — reactor operators and enrichers like Centress (likely Centrus), which “kind of looks nice” though he has no position yet pending research on what the US actually agreed to and whether the company joins foreign construction.

7. Iran loses the strait forever — and markets are safe

• The WSJ headline — Trump says the US will bomb Iran’s power plants if Tehran strikes ships — “literally sounds like they’re in kindergarten,” which Jonah maps onto his own kids’ toy fights. His base case: the skirmish continues until regime change, because if Iran was near a weapon (he takes “the middle of the bell curve” on that debate), war only accelerates its ambitions. “This war is not going to stop.”
• Why oil isn’t ripping: east-west pipelines are being built on both sides of the Gulf, so “Iran used its leverage point but now it’s going to lose its leverage point forever over the next 5 years” — the strait “is not going to matter in 10 years.” And Trump is markets-sensitive — “he does not want his name on a great depression” — easing off when oil hit the danger zone, letting “stocks build, as we say in oil land,” before having back at it.
• Jonah’s military reframe of why a modern navy can’t just secure Hormuz — which “blows my mind,” per Jonah: “the world is huge”; Avi adds that you need eyes on every vessel when Iranian ships hide themselves and you can’t risk an international incident. Plus the likely-Ahmadinejad twist: Israeli intelligence cultivated the bitter ex-president as a possible inside man before Iran caught him — Avi’s razor being that the US “would not have gone in… unless they had a plan to end the war reasonably quickly. Clearly the plan went wrong.”

8. Gold at 4K: trade the chart, not the fundamentals

• After weeks of chop, the tape finally stabilized — NASDAQ down 40 pips, Bitcoin off 1%, ZEC down 4% — and Avi’s best-looking setup is gold, off 26% from the highs but defending 4,000 for six weeks: “I’m actually just not going to overintellectualize this. That’s a f*ing good-looking chart.” Buy the level with a tight stop, or wait for 4,300 — the yearly moving average — as the signal; target prior resistance ~4,800. Kicker: Japan’s inflation pressures might prompt reserve-building.
• Why TA works for gold: gold’s driver is central banks — “low information people” whose decisions are “quite literally often determined based on price level.” “Gold is a reflexive asset” — whereas “if you tried to do technical analysis on Nvidia, go see a doctor.” Jonah’s caveat, followed by Avi’s risk rule: reflexive buyers chase strength, not downtrends — so if 4,000 breaks, “your thesis has been disproven… you have to sell,” don’t “sit idly by while gold burns.”
• On crypto: Clarity Act odds are “starting to tick back up” in Washington — bullish — and “aside from Saylor being Saylor, I think crypto may have bottomed out, like, temporarily.” If memory goes nowhere for a week or two, Avi expects rotation back into gold and Bitcoin. Avi’s closing rule: “shit can crash before it rips.”