Citrini Black Tuesday — the press-applied name for the late-February 2026 tech-software selloff — was the cleanest case in living memory of an independent Substack memo moving the tape on an institutional cohort. On Monday February 23, IBM finished down 13 percent, its worst single day since 2000. DataDog, CrowdStrike, and Zscaler each fell roughly 10 to 11 percent. DoorDash, American Express, Blackstone, and KKR each finished the Monday session off 6 to 9 percent. By Thursday February 26, Citadel Securities had published a public rebuttal — a thing Citadel almost never does — and Howard Marks had released a new Oaktree memo. The essay that catalysed all of this was an explicit thought experiment, written as a fictional macro memo dated June 2028. Its authors were James Van Geelen, the founder of a small research shop called Citrini Research, and the AI entrepreneur Alap Shah. This is the story of how their memo moved markets, and the institutional debate the response produced.

Sunday to Thursday: the week the essay moved the tape

The essay went live on Sunday evening, February 22, 2026. By Monday morning, it was being forwarded inside every major buy-side credit and equity desk we have line-of-sight to. Monday’s session was already exposed to two separate same-day Anthropic catalysts — the launch of Claude Code Security (which directly threatened the cybersecurity software cohort) and a parallel announcement that Claude Code could now modernize Cobol systems (which was read as a hit to IBM’s services franchise). Three narratives, one direction.

By Monday’s close, the damage was visible. IBM finished the day down 13 percent — its worst single session since 2000 — on the Cobol-modernization news. DataDog ended off 11.3 percent; Zscaler 10.3 percent; CrowdStrike 9.8 percent — all read as direct casualties of the Claude Code Security launch. Across the broader names that get caught in any AI-narrative reversal, KKR finished off 8.9 percent, American Express 7.2 percent, DoorDash 6.6 percent, and Blackstone 6.2 percent. The market did not need to read the Citrini memo to fall — the Anthropic catalysts were doing structural damage on their own. But the Citrini memo gave the price action a frame, and frames are what make a sector decline tradeable as a narrative event rather than a day’s drift. By Tuesday, press coverage had named the broader episode Black Tuesday, even though the bulk of the single-session damage had landed on Monday. Tuesday itself saw stabilization in many of the names hit on Monday.

What the week revealed was the institutional plumbing’s new susceptibility to long-form independent research. Twenty years ago, market-moving research came from sell-side desks via morning notes. Today, a Substack memo can reach the same readership before the sell-side has caught up to the question being asked.

Who is Citrini Research?

Before walking through what the essay argued, the source needs context. Citrini Research is itself a recent phenomenon. Founded in 2023 by James van Geelen — a 33-year-old former Los Angeles paramedic with degrees in biology and psychology who built his original capital by selling an alternative-medicine startup to a private equity fund in 2018 — Citrini operates as an independent thematic-equity and global-macro research shop, headquartered in New York with roughly 10 employees, and distributes primarily through a paid Substack subscription. By early 2026 that Substack had crossed 119,000 paid subscribers, making it the top finance newsletter on the platform. The reputation that put Citrini on institutional Bloomberg chats came earlier: van Geelen publicly announced a short Silicon Valley Bank position in late 2022, months before the bank’s March 2023 collapse, and Citrini was demonstrably early on both the NVIDIA-AI thesis and the Eli Lilly GLP-1 obesity-drug trade. The house style is long-form thematic reconstruction with a second-order-thinking frame — what does the consensus story imply two steps further. The 2028 Global Intelligence Crisis is that frame applied to AI.

What the essay actually argued

“The 2028 Global Intelligence Crisis” is framed as a fictional document — a macro strategy memo dated June 2028, written by Citrini analysts looking backward at a world where AI displacement has already happened. The opening section sets the scene with deliberately specific numbers: national unemployment at 10.2 percent, the S&P 500 down 38 percent from its October 2026 peak, white-collar employment collapsed across software development, consulting, paralegal work, and most knowledge-economy categories. The memo’s authors call the failure mode “Ghost GDP” — their term for corporate output that registers in the national accounts but does not circulate through households because AI has displaced the white-collar workers who used to earn the wages that paid for the consumption that drove the output in the first place. Productivity goes up; aggregate demand collapses; the equity market re-prices the entire AI-software stack at sub-3× sales because the bull case actually got delivered.

The structural move is what made the essay viral. The bull case for AI investing — that the technology genuinely transforms productivity at a scale unseen since the personal computer — is the case the long side of every AI-software portfolio is making implicitly. Citrini took that case and ran the second-order consequence: if productivity gets delivered AND the productivity gains accrue to the capital side of the labour/capital split, the demand-side of the economy has nowhere to land. Software earnings do not exist in a vacuum; they exist because consumers and businesses pay for software with money they earned from labour. Remove the labour, and the demand for software collapses two quarters later. Same bull thesis, opposite portfolio implication.

The essay is explicitly framed as a thought experiment, not a forecast. The point is not to predict 2028 — it is to make the long-AI-software position specify what demand-side assumption it is implicitly carrying. Most positions had never asked. The full memo is on Citrini’s Substack: The 2028 Global Intelligence Crisis.

Why a Substack moved the tape

Three structural conditions had to be in place for one essay to move tape this way. They all were.

The first was that Substack had reached institutional inboxes faster than the sell-side. By 2026, every major bank credit and equity desk read a short list of independent research the day it dropped — often before their own analysts had absorbed it and certainly before the morning notes went out. Independent voices had been moving institutional money for at least two years; the Citrini essay was the moment that became visible in the tape. By Monday morning, the memo was being shared inside Citadel Securities, Citadel’s response team had read it, and Bloomberg chat lines were full of it.

The second was that the software sector was already vulnerable to a narrative shock. According to coverage circulating in mid-February, hedge funds had built roughly $24 billion in software-sector shorts during the first eight weeks of 2026, against a backdrop of the sector bleeding roughly $1 trillion in market cap since January on AI-substitution fears. The shorts were positioned; the narrative was missing. Citrini provided the narrative. A long-form essay does not move a sector that nobody was already short of, but the inverse holds with force: it can move a sector that everyone is short of and waiting for a reason. The market had loaded the gun. The essay pulled the trigger.

The third was the same-week compounding from Anthropic. Monday’s Claude Code Security launch and Tuesday’s Cobol-modernization announcement — both small in isolation — landed inside a narrative that was already being read as “software is being eaten by what software built.” The DataDog, CrowdStrike, and Zscaler drops on Monday were not Citrini-driven, and IBM’s Tuesday drop was about Cobol services, not Ghost GDP. But the essay’s narrative absorbed all three into one story. By Tuesday afternoon, financial Twitter was treating the entire week’s software damage as if it had a single cause, and that cause was a Substack post. That telescoping is exactly how essays move markets — they unify what would otherwise be noise.

By the standards of 2020s capital markets, this is structurally new. Sell-side analysts have always moved individual names. Magazine cover stories have always marked tops and bottoms. But an explicitly fictional thought-experiment essay, distributed through a paid Substack with no institutional credentialing behind it, has never — in our reading — been the proximate cause of a coordinated multi-name single-session selloff of this scale in tech-software equities.

The Citadel rebuttal: empirical pushback from the market-maker

The most institutionally surprising response came two days after Black Tuesday. On Thursday February 26, Citadel Securities published a public rebuttal — authored by Citadel macro strategist Frank Flight — arguing that the Citrini scenario was a profound misreading of macro fundamentals, not a probabilistic tail. Citadel almost never publishes public macro rebuttals; market-makers usually let prices do the arguing. The fact of the rebuttal was itself the signal — they had been forced to respond.

Flight’s empirical lead was Indeed job-posting data showing that demand for software engineers was up 11 percent year-over-year in early 2026. The St. Louis Fed’s Real-Time Population Survey, Flight noted, found that daily use of generative AI for work was “unexpectedly stable” and showed “little evidence of any imminent displacement risk.” The structural argument behind the data was that Citrini had conflated recursive technology with recursive economic adoption. Technology compounds quickly. Economic deployment of technology compounds slowly, because organisational integration is costly, regulation emerges, and diminishing marginal returns dominate the later stages of any adoption curve. Markets, Flight argued, often extrapolate the early acceleration phase linearly, but historical pace of adoption plateaus.

For the institutional reader, the Citadel response did two jobs at once. It provided the data set the long-AI-software portfolios needed to hold position through the Black Tuesday volatility. And it implicitly confirmed that Citrini’s essay had moved enough institutional capital to warrant a written response from one of the world’s largest market-makers. Rebuttals are status assignments. Citadel had assigned status. The full Fortune coverage of the rebuttal is here: Citadel Securities demolishes viral AI doomsday essay.

The dean of credit weighs in: Howard Marks, AI Hurtles Ahead

The same day Citadel published its rebuttal — Thursday February 26 — Howard Marks released a new Oaktree memo titled AI Hurtles Ahead. Marks publishes on his own cadence; the timing was not coordinated with the Citrini cycle in any visible way. But it functioned as the institutional-credit response to what had been an institutional-equity provocation, and the timing is what gave it weight. Marks is the dean of credit memos for a reason: when he writes, every LP-led credit allocator in the Western institutional system reads.

His position was explicitly two-handed. On one hand, AI is not a fad. The power, speed, and autonomy of the systems are genuine, the implications for capital allocation are real, and the period since his previous memo on the subject — Is It a Bubble?, December 2025 — had only sharpened that conviction. On the other hand, the appropriateness of asset prices is a separate question, and at current AI-asset valuations, the transformational potential and the appropriate price were being evaluated as if they were the same question. They are not. Investors must view AI’s transformational power against the appropriateness of the prices being asked for AI assets. The memo’s structural move was the separation of “is this real” from “is this priced.”

The portfolio posture Marks recommended is restraint — not absence. If you accept that AI is genuinely transformational, you cannot be wholly absent from it; the opportunity cost of zero exposure compounds against you. If you accept that current valuations price the entire transformation, you cannot be at policy-level exposure either; the downside is asymmetric. The position is the middle: own AI at thoughtful weights, hedge the high-multiple expressions, and reserve incremental capital for the moments when the price-of-transformation re-rates downward, which will happen episodically regardless of whether Citrini’s tail scenario plays out. Within a week of publication, Brookfield Private Wealth had republished the memo through its institutional-client Marks-memo channel — the routine signal that the wealth-distribution layer treats a Marks memo as required reading the moment it drops. The memo lives at Oaktree’s site. The same “real-but-priced” discipline transfers to other emerging-tech categories, and for the parallel case study see our breakdown of the quantum computing investment landscape.

The strategist’s reframe: Ben Thompson, Agents Over Bubbles

Three weeks after Black Tuesday — Monday March 16, 2026 — Ben Thompson published Agents Over Bubbles on Stratechery, the most-read tech-strategy newsletter in finance. His answer to the question Citrini had implicitly forced was a clean no on the bubble framing, but with an amendment that mattered: the current AI capex cycle is not a bubble because agentic AI systems (language models bound to task-execution frameworks with persistent memory and tool-use) represent a genuine paradigm shift that creates demonstrable economic value at the agent layer. The bubble framing, Thompson argued, fits the model-layer trade — which he conceded was increasingly priced — but the agent layer is where genuine strategic differentiation, and therefore genuine excess return, still lives.

The structural move was to separate “AI” into three layers (compute, models, agents) and ask which layer the bubble accusation actually applied to. By the time Thompson wrote, the equity-research notes circulating through institutional desks were treating “AI” as a single thing, which made “is it a bubble” a single question. Thompson’s essay disaggregated the question, and the disaggregation became the default vocabulary for how the strategic AI conversation has been conducted since. By April, it was hard to find an earnings-call Q&A for an AI-adjacent software company that did not reference the compute / models / agents distinction at least once.

The portfolio implication was granular. If you accept Thompson’s three-layer framework, you reduce exposure to the model layer (where margin compression from open-weights competition is now the consensus expectation) and increase exposure to the agent layer (orchestration platforms, evaluation infrastructure, vertical-specific agent products). The compute layer is its own conversation — see our separate coverage on the DRAM pricing cycle and the power-grid bottleneck on AI data centers for that. What Thompson’s essay did was make explicit the strategic question every AI-exposed portfolio had been implicitly carrying: which layer of the stack do you actually own? Most portfolios had been “all of it via large-cap exposure” without admitting it. Agents Over Bubbles lives at Stratechery.

The macro counterweight: Adam Tooze and the K-shaped economy

The fourth response to the Citrini debate arrived on a delay, from a different vocabulary. On May 10, 2026 — ten and a half weeks after Black Tuesday — the macro historian Adam Tooze published Chartbook 447, subtitled The US economy in May 2026: How much cognitive dissonance can you handle? The essay does not engage Citrini directly. It does not need to. Its central argument is that the US economy is the canonical case study in K-shaped distribution — at the top of the K, private-AI valuations compound vertically (Anthropic from $183 billion in September 2025 to $380 billion by spring 2026); at the bottom, a broader economy delivering contradictory signals on labour, prices, and household finances. The cognitive dissonance of the subtitle is the central diagnostic.

What Tooze contributed to the post-Citrini conversation is the macro denominator. Citrini argued that the AI bull case would eventually destroy demand through displacement. Tooze argued that the K-shape is already doing slow demand destruction on a less visible time horizon — through the distribution of the gains, not through their absence. You do not need Citrini’s 2028 displacement scenario to get to a similar place; you only need to extend the current K-shape forward for a few more cycles. Two days after Chartbook 447, Tooze extended the K-shape framing into a Foreign Policy conversation with Cameron Abadi — one of his many cross-channel amplifications that keep the Chartbook arguments inside the foreign-policy audience as well as the finance one.

For investors, Tooze is the single best long-form macro counterweight to the AI-only narrative dominating finance Substack. He provides the wages, deficits, and distribution context against which the AI-thesis numerator gets evaluated. The portfolio implication is to apply a haircut to any thesis that treats US “GDP growth” as a single number, because the May 2026 chartbook makes the case that aggregate GDP has become a low-information statistic in a K-shaped economy. For Luna3’s separate read on the broader US macro picture, see our recent piece on the US expansion against the warning signals. The chartbook itself is at Tooze’s Substack.

What investors actually learned

The four responses do not produce a single answer. They produce a dialectic. Citrini is the bear thesis; Citadel is the empirical rebuttal; Marks is the institutional discipline that refuses to take a side and treats both real and priced as separate questions; Thompson is the strategist’s reframe that disaggregates the question itself; Tooze is the macro counterweight that gets to a similar bear conclusion through a different mechanism. None of the four positions is wrong; each is engaging with a different layer of the same problem.

The portfolio takeaway from reading all four against the Citrini provocation is granular, not directional. Holding long-duration AI software at 12× sales requires conviction that human-displacement does NOT compound into demand destruction. The investors most exposed to the Citrini outcome (high-multiple SaaS at 10–15× sales) should be the ones who read the essay most carefully. The investors most exposed to the Marks-style mispricing risk should hedge the high-multiple expressions. The investors carrying broad large-cap AI exposure without specifying which layer of the Thompson three-layer stack they own are now in a worse risk-adjusted position than the targeted alternatives that did not exist 18 months ago. The investors using aggregate US GDP as the denominator for their macro view should haircut that denominator per Tooze’s K-shape evidence.

None of this requires Citrini’s scenario to play out. It requires only that the question Citrini forced — what is the demand-side assumption your AI-software portfolio is implicitly carrying? — be answered explicitly rather than implicitly. That is the discipline shift the essay produced. Whether the productivity gains accrue to capital or to labour over the next five years is the macro question that determines whether Citrini’s tail becomes a base case. We do not know the answer. Neither does Marks, neither does Thompson, neither does Tooze, and neither does Frank Flight at Citadel. What we know is that the question is now being asked by everyone, and was being asked by approximately nobody until February 22.

What Citrini Black Tuesday actually changed

The clean retrospective frame is that the Citrini Black Tuesday episode was the cleanest case in living memory of independent long-form research moving institutional capital faster than sell-side research could respond. But the durable change is not the price action. The price action recovered fast: by Wednesday February 25, most of the names hit on Monday had closed back at or above their pre-event Friday levels. The durable change is that the institutional reaction time to a viral essay is now measured in days, not quarters. Citadel published a public rebuttal within 96 hours. Marks released a memo on the same news cycle. Brookfield Private Wealth republished within a week. Thompson disaggregated the question within three weeks. Tooze provided the macro denominator within ten.

The next Substack moment is coming. The same conditions that made Citrini Black Tuesday possible are still in place: independent research reaches institutional inboxes before the sell-side, sectors get crowded into narrative trades that wait for catalysts, and same-week compounding catalysts amplify whichever narrative they touch. The investor question is not whether the next Citrini moment happens but how prepared the institutional plumbing is to absorb it. The Black Tuesday response cycle suggests the plumbing is now fast. It does not suggest the plumbing is right.

That is what the week of February 22–24, 2026 actually changed. Independent research now sets the agenda before institutional research absorbs it. The dialectic that emerged is more durable than the price action that produced it. And the canonical name for what happened — Citrini Black Tuesday — is itself the marker that the institutional reading class now treats a Substack post the way it once treated a sell-side downgrade. The reading class has changed. The plumbing has caught up. The next essay is already being written.