Jordi Visser — AI Speed Crash Opportunity Investment Thesis

Source: AI's Speed Crash Is Creating the Next Opportunity, Jordi Visser (YouTube), July 12, 2026.

The Framework: Short Abundance, Long Scarcity — Inside a Token-Demand Bull Market

Visser’s organizing lens: this is a structural bull market built on token demand, not a classic bubble. Parabolic ups produce speed crashes (violent, short-duration drawdowns). The AI midcycle second-derivative slowdown is ending as a surprise — it is now consensus, positioning has been flushed, and bears are reappearing (historically constructive). The investable rule remains unchanged:

Side	What It Means	Investable Expression
Short abundance	Anything built on code that computers can replicate	Legacy SaaS / seat-software multiples (e.g. Salesforce as contrast)
Long scarcity	Physically constrained by linear human supply	Memory (Samsung, Micron, SK Hynix), foundry capacity (TSMC), power/turbines
Token demand	Exponential digital demand vs linear physical supply	Next wave = consumer agents >> coding agents
Speed crash	Deleveraging flush inside the bull	Entry when vol/RSI/oversold cleanse completes

"You want to be short abundance. Anything built on code that can be replicated by a computer. And you want to be long scarcity. Anything that the spenders are spending on that is physically constrained by linear humans."

Investment Thesis #1: Memory Speed Crash = Buy Scarcity, Not Fade the Cycle

The argument: Samsung printed a 19-fold preliminary quarterly profit surge and still sold off — a 21% 10-day rate-of-change drawdown, the largest since COVID shutdown and Lehman. That is a speed crash inside a bull market, not a news-failure top. Samsung trades near a ~4 PE on 2026 estimates with operating profit potentially >$217B — more than cumulative operating profit over the past 40 years. Analysts already bake in slowdown (growth 524% this year / 43% next), yet multiples stay depressed. Memory shortages are now projected beyond 2030 (upgraded from “until 2030”); Micron’s long-term supply agreement is worth $100B; Dylan Patel: capacity grows only 20–30%/year while demand doubles.

"Don't treat this as a one-off event where AI is just doing this and everyone's over buying and then we're going to run into a problem. It's silly. We're not even in the first inning of token demand."

Contrarian element: Bears map memory to a normal hardware cycle or a “fracking” oversupply analogy. Visser rejects both: demand is digital/tokens/agents, not physical unit sales — so supply cannot chase demand the way oil supply chased fracking.

Trigger: Memory prices stay elevated; shortages language extends past 2030; SK Hynix / Samsung offerings remain heavily oversubscribed (he cites SK Hynix deal >7× oversubscribed and trading up).

Names: Samsung Electronics, Micron (MU) — he reduced mid-May and plans to buy back after the flush; SK Hynix; TSMC as the human-constrained foundry bottleneck that keeps scarcity prices high.

Investment Thesis #2: Positioning Has Been Cleansed — Concentrated AI Basket Reset

The argument: Tech-momentum 60-day realized vol hit 87. Systematic long/short (quant) managers had their worst stretch since December 2023 and gave back ~one-third of YTD gains. Retail 2× levered products gave back 62% on a ~23–38% underlying correction. 87% of S&P 500 semiconductor issues registered oversold on a 14-day basis — historically strong entry zones on Jeff deGraaf’s framing. His 100-name thematic book went from 73/100 above the 20-day (May 8) to 21 mid-week, then bounced to 31; 86/100 still above the rising 200-day (structural bull intact). Concentrated 10-name basket retested the 50-day with RSI reset — same setup that preceded the March liftoff.

"This is a structural bull market. It is based on token demand."

Contrarian element: Market reads Samsung/Micron selloffs on good news as cycle tops. Visser reads them as deleverage completion after Opus 4.5–era chase.

Trigger: Percent-above-50-day / RSI bottoms stabilize; SOXX and memory names attract large inflows again (he notes massive SOXX / SK Hynix inflows on the session he recorded).

Names: NVIDIA (NVDA) — one of his 10 concentrated names; closed above 20-day first time since June 2, holding the 200-day, MACD buy signal at lowest valuation in a decade. Broader AI beneficiaries basket after the flush.

Investment Thesis #3: Consumer Agents Are the Next Token Supercycle Catalyst

The argument: Coding agents (Opus 4.5) were the first visible token explosion. Consumer agents are the next — and larger — theme for the coming 12 months. Meta CTO Andrew Bosworth: consumer AI has been slow because the product layer is immature, not because models are weak. When the interface disappears into daily life (search/maps/messaging/camera analog), frequency, context length, and multi-step reasoning drive a step-function in tokens. Goldman reportedly estimates agents could drive token consumption up roughly 24× to ~120 quadrillion/month. Dylan Patel’s SemiAnalysis firm: token spend from <$100K (Nov) → ~$4M annualized by January → ~$11M annualized today (peak week ~$14M) for a ~90-person firm. Meta selling some capacity would reflect consumer-agent product lag, not excess compute.

"When the consumer AI interface problem is solved, token usage should explode. AI moves from being something people occasionally summon to something that is working in the background."

Contrarian element: Consensus treats Meta capacity sales / “air pocket” narratives as proof of overbuild. Visser: hyperscaler backlogs already total nearly $2T (Microsoft $627B, Google $462B, Amazon $225B, Oracle $638B); SemiAnalysis capex forecast $11.1T vs Goldman $7.6T; Meta itself said it was compute-constrained and underestimated demand.

Trigger: Consumer-agent product breakthroughs at Meta / Apple (habit formation, always-on assistants); Apple sustaining strength despite 20% price hikes from memory costs (he notes AAPL closing ATH on that print).

Names: Meta (META), Apple (AAPL), NVIDIA / memory stack as token-demand leverage; Anthropic / frontier labs as high-ASP assistant workload winners (Patel: smartest model can be cheaper overall via fewer iterations).

Capacity reality check: Microsoft — “strong customer demand continues to exceed capacity”; Google allocating under robust ROIC with Trainium largely sold out; Meta compute-constrained and admitted underestimating demand. SemiAnalysis: compute market continues to tighten; rental curves look like bullish oil curves. The “too much capacity” narrative fails the latest earnings tape.

Investment Thesis #4: Bessent Reframe — Crypto Becomes US Financial Statecraft (AI Macro Nexus Phase Two)

The argument: AI Macro Nexus phase one was AI disrupting productivity, labor, inflation, and earnings. Phase two: financial guardrails change. Scott Bessent’s New York Economic Club speech — and Mohamed El-Erian’s NYT op-ed — reframe digital assets, stablecoins, tokenization, and new payment systems as tools of US economic/national security, not speculative sideshows. Parallel to Andreessen’s 2014 “Why Bitcoin Matters”: mainstream acceptance lags; agentic commerce is the accelerant once agents transact, settle invoices, and move money.

"You can no longer have no view on crypto. You cannot just call it speculative and walk away."

Contrarian element: Crypto sentiment is as bleak as AI was ~15 months earlier when he was buying Micron below 100. Michael Saylor sold ~3,500 BTC — Bitcoin closed higher (bad-news absorption).

Trigger: Clarity Act progress; Bitcoin reclaiming 50-day then 200-day; DOGE (retail energy proxy) above 200-day; Fed — July hike risk case is one-and-done 25 bps; no hike = Bitcoin tailwinds.

Names: Bitcoin (his “S&P 500 of the future of tokenization”); Circle (CRCL) as public equal-weight name in his 40-name crypto index; broader tokenization / stablecoin rails.

The Ecosystem Map: Where Visser Is Positioning

Concentrated AI basket (10 names) — reduced into the midcycle slowdown; now reviewing buybacks after 50-day / RSI reset; NVIDIA explicitly retained as a core name
Memory — exited / cut Micron mid-May ahead of the top; planning to re-enter post-flush; constructive on Samsung / SK Hynix scarcity math
Consumer AI — rising research focus on Meta and Apple as the product-layer unlock
Palantir (PLTR) — still in the 100-name book and personally owned (~$130 cost basis; consolidating)
Crypto sleeve — silver/Bitcoin rotation earlier underperformed; now treating Bessent speech as the regime-change catalyst; building a crypto-forward YouTube track covering stablecoins, tokenization, and the new payment rails
Value-chain view — most value accrues to physically scarce inputs (memory, foundry, power) while token demand compounds through coding → consumer → FSD/humanoid waves; each capability threshold creates a new consumption surface humans price linearly and miss

Key Risks

Classic memory cyclicality — if demand reverts to hardware-unit economics, 4× PE Samsung is a value trap, not scarcity alpha
Midcycle second-derivative persistence — rate-of-change slowdown continues even if absolute spend is large; returns compress vs. the prior 20-bagger memory leg
Consumer-agent delay — product/habit-formation friction lasts longer; Meta capacity sales get read (correctly, temporarily) as demand destruction
Token-index narrative abuse — seasonal/bill-shock dips misread as structural token-demand failure
Fed hawkish surprise — a hike path beyond one-and-done 25 bps would pressure Bitcoin / risk assets
Algorithmic efficiency hail Mary — if model efficiency collapses memory intensity fast enough to close the 20–30% supply vs doubling demand gap

Investment Opportunities at a Glance

Tier	Name / Category	Core Thesis	Conviction Signal
1	Samsung Electronics (SSNLF)	Speed-crash entry into scarcity memory; ~4 PE vs $217B+ OP potential	19× profit print + largest ROC selloff since COVID/Lehman
1	Micron (MU)	Re-entry candidate after mid-May cut; shortages beyond 2030; $100B LTA	Explicit buyback timing after flush
1	SK Hynix	HBM/memory scarcity; deal >7× oversubscribed	Same scarcity framework as Samsung
1	NVIDIA (NVDA)	Concentrated-basket core; decade-low valuation + MACD buy + 200-day hold	Named as one of 10 concentrated names
2	Meta (META)	Consumer-agent product unlock → next token explosion	Bosworth interview + research focus
2	Apple (AAPL)	Consumer-agent distribution + ATH resilience despite 20% memory-driven price hikes	Closing ATH on cost-pass-through news
2	TSMC	Human-constrained foundry that prevents overbuild / keeps scarcity prices high	Named as bubble-prevention constraint
2	Bitcoin / Circle (CRCL)	Bessent statecraft reframe; BTC as tokenization benchmark	Saylor-sale absorption; 40-name crypto index
3	Gas turbines / power capacity	Physical constraint keeping compute scarce and prices elevated	Named alongside TSMC as supply-side governor
3	Palantir (PLTR)	Own-your-stack enterprise AI; consolidating in 100-name book	Personal holding ~$130; still owned
4	Anthropic / frontier labs	High-ASP assistant workloads; approaching operating profit	Patel spend curve + margin commentary

Monitoring Checklist

Bottom Line

Buy the memory speed crash. A 21% Samsung ROC flush on a 19× profit print is deleveraging inside scarcity, not a cycle top — shortages now run beyond 2030 against 20–30% capacity growth and doubling demand.
Consumer agents > coding agents for the next token leg. Goldman’s ~24× token estimate and Meta/Apple product-layer progress are the underpriced catalysts; Meta capacity sales are product lag, not overbuild.
Hyperscaler backlogs already refute the “too much capacity” meme — nearly $2T committed demand and SemiAnalysis $11.1T capex vs Goldman $7.6T.
You no longer get to have “no view” on crypto. Bessent + El-Erian reframe stablecoins/tokenization as US financial statecraft; agentic commerce is the accelerant — Bitcoin absorbed a Saylor sale and closed higher.
Short abundance / long scarcity still rules. Salesforce-style code abundance compresses; physically constrained memory, foundry, and power capture the spend.