A recent essay by Citrini and Alap Shah titled “The 2028 Global Intelligence Crisis” presents a provocative scenario: AI keeps improving, productivity surges, margins expand—and yet the broader economy destabilizes.
It’s not a prediction. It’s a thought experiment. But it’s a serious one.
Read the original here: https://www.citriniresearch.com/p/2028gic
Their core question is uncomfortable: If AI continues to work exactly as promised, what breaks?
For those of us focused on data storage, compute infrastructure, and the rise of cloud computing alternatives (“neoclouds”), the implications are more structural than they first appear.
Citrini’s scenario unfolds in three stages.
In 2026, coding agents become capable enough that enterprises can replicate mid-market SaaS tools internally. Procurement teams start renegotiating contracts. “Build vs. buy” becomes real, not theoretical.
The result:
The key shift isn’t automation. It’s substitution.
By 2027, consumer-facing AI agents transact autonomously. They re-shop insurance annually, cancel unused subscriptions, price-match continuously, and route payments around interchange fees.
Entire business models built on:
begin to compress. The authors frame this as “friction going to zero.”
Here’s the spiral: AI improves → companies cut white-collar labor → spending weakens → firms reinvest savings into AI → AI improves again.
In their model, AI infrastructure keeps expanding because it substitutes for payroll. It becomes a cost structure decision, not discretionary innovation spend.
That creates a structural decoupling:
That tension is the heart of the essay.
The essay isn’t about storage. But the second-order effects are directly relevant.
If AI meaningfully replaces labor, it shifts from “digital transformation initiative” to “core margin protection.” Defensive spend behaves differently in downturns. Companies cut marketing budgets before they cut systems that reduce payroll.
For infrastructure, this implies:
Storage demand in this model is not tied to consumer GDP growth. It’s tied to enterprise efficiency mandates. That’s a different demand curve.
If application-layer differentiation collapses and SaaS pricing compresses, value shifts downward. The durable layer becomes:
This is the structural case behind neoclouds and AI-specialized infrastructure providers. When software margins compress, infrastructure becomes the leverage layer.
But there’s nuance: if enterprise budgets broadly shrink, total infrastructure spend may not rise in aggregate, even if AI remains critical. What grows is AI’s share of IT budgets. That distinction matters for operators modeling long-term demand.
In an agentic economy, AI systems transact continuously. That produces:
The bottleneck shifts from raw capacity to:
Storage stops being archival. It becomes operational infrastructure. The question becomes less “how many petabytes” and more “how fast can data move between inference layers?”
Citrini assumes AI infrastructure keeps booming even as the economy weakens. That’s plausible—but not guaranteed.
Neocloud operators are:
Neoclouds sit directly in that tension.
The essay spends significant time on private credit exposure to software. For infrastructure, the analogous risk is financing.
AI data center expansion increasingly relies on:
If macro volatility rises and funding spreads widen, expansion slows—even if demand remains strong. Infrastructure doesn’t fail because GPUs disappear. It fails when capital tightens.
The essay references proposals to tax AI inference or capture returns from intelligence infrastructure. If payroll tax receipts decline because labor is displaced, governments will look for alternative tax bases.
Compute is measurable. Inference is billable. Power draw is traceable. A compute tax is no longer science fiction.
If implemented, it could:
Policy risk now belongs in infrastructure models.
Citrini’s scenario challenges a long-standing economic assumption: technological productivity gains create new jobs that sustain consumer demand. If that chain weakens, infrastructure demand decouples from household income.
We could see a world where:
That’s not a traditional boom or bust cycle. It’s a structural rebalancing.
The market’s most fragile assumption right now is not that AI won’t work. It’s that AI will work—and everything else will remain stable.
For data storage and AI infrastructure operators, the opportunity is real. AI substitution drives persistent compute and data intensity. But durability depends on one question: Is AI augmenting labor, or replacing it?
The canary isn’t GPU supply. It’s wage data. Infrastructure investors should be watching both.
