The AI Mirage: Inside the Hype Cycle Where Paper Trillions Meet Gravity
Oct 28, 2025
The numbers glitter. But a good share of them are scaffolds—warrants, options, reservations, and promises—stacked to look like revenue. That’s not a sin; it’s how early markets breathe. The risk is pretending scaffolding is steel. The AI hype cycle thrives on that confusion: headlines imply capital has landed; balance sheets tell you it hasn’t. The gulf between the two is where gravity does its work.
Enterprises are learning the same lesson from a different angle. Not every problem deserves a trillion‑parameter model. As one MIT Sloan panelist put it, an LLM is often a wildly expensive way to solve a narrow task; the DeepSeek example shows you can do more with less . Constraints forced DeepSeek to get crafty—both in training and inference paths—proving that architecture and discipline can beat raw scale when cost pressures bite . That is the first fissure in the mirage: if smaller models solve real work at sane cost, the “all roads lead to mega‑model” thesis is a story, not a state.
Paper Growth, Real Cash
Cash flows are where hype meets bone. OpenAI’s reported annualised run‑rate headlines created the sense of a money gusher, yet back‑of‑the‑envelope maths put monthly revenue near $167 million at that point, and the business still ran at a loss, with talk of needing tens of billions more to keep going . When investment banks and consultancies pitch multi‑trillion‑dollar “opportunities,” your scepticism should sharpen, not soften. The bigger the promised total addressable future, the more you should ask: who pays, when, and with what margins?
There’s a second bill coming due: energy. Researchers estimate gen‑AI systems can consume up to 33 times the energy of task‑specific software—an expense that doesn’t vanish when the press release fades . If the unit economics depend on cheap power and infinite patience, they’re not unit economics; they’re a wish.
The Hype Mechanics
Every technology swings through elation, disappointment, and recovery—the familiar curve from peak of expectation to trough of disillusionment, then a plateau where reality finally catches up. AI isn’t one cycle; it’s many nested cycles—the current burst in generative models is a mini‑cycle riding a broader, decades‑long drift upward. When the promises stack faster than proof, mood breaks hard and fast. Then the useful parts settle in and compound quietly.
Right now the signals are split. Some voices declare “higher, immediately.” Others insist “sell, yesterday.” When faith bifurcates like that, my own read goes neutral. Our Vector Mass Psychology System (VPS)—15 sentiment and flow inputs—stays silent until coherence returns. It requires an 80+ score to speak. At times like these it hovers near zero—one chorus cancels the other . That fragmentation, paired with complacency that’s high but not manic, is ideal terrain for a sharp shakeout that feels like a crash but isn’t—punishment for emotion, not positioning .
How Paper Becomes a Story
Here’s how the mirage assembles itself without any villains. A hyperscaler reserves compute capacity from a vendor; a vendor books that reservation as backlog; a startup announces “access” to that capacity as a partnership; a bank packages the three as “growth”. No one lies. Everyone flatters the future. Capital circles the system, touching each balance sheet long enough to leave a smudge. If real demand grows into the scaffolding, no harm done. If it doesn’t, warrants expire out‑of‑the‑money, reservations roll forward, and the bonds holding the narrative together lose their sheen. The clean question is always the same: where did cash arrive, when, and at what margin?
There’s a reason this phase feels familiar. We’ve run this play before—with “big data,” with “mobile”, with “cloud.” Each cycle minted winners. Each cycle also minted a long tail of companies that were really capital structures in fancy clothes. The phrase you want in a mania is not “we’re late.” It’s “show me the receipt.”
State Over Story: A Five‑Dial Overlay
Markets are mirrors. They reflect belief until belief becomes price, then they reverse when belief exhausts itself. To avoid trading the mirror, I read five dials that ignore narrative and describe state. Breadth: is participation wide, or just the mega‑caps holding indices aloft? Credit: are high‑yield spreads tightening (healing) or widening (strain)? Real yields/USD: is the macro tide favouring long‑duration cashflows or pressing them? Volatility term structure: is the curve steep (calm digesting risk) or inverted (fear). Leadership: do cyclicals and quality carry the tape, or are we down to one theme doing all the lifting?
When three or more align, probability improves; when they argue, it’s noise. That discipline is how we acted when the crowd was paralysed. On 7 May, our MOAB (Mother of All Buy) trigger fired—rare, built for extremes. From that day, the S&P 500 climbed roughly 12%, the Nasdaq 100 more than 15%, and the Composite over 16%—process outlasting panic, arithmetic beating adrenaline . In April the VIX spiked then faded; fear flashed, never caught. No manic greed, no speculative fever, no true‑top delusion. Pullbacks were invitations because the crowd wasn’t euphoric and the trend was up .
Scenario Trees, Not Slogans
If you want to trade the AI hype cycle rather than be traded by it, sketch scenarios instead of speeches. Demand softens 10–15%: capacity reservations roll, guidance gets haircut, option packages miss performance triggers, multiples compress. Margins compress 300–500 bps: GPU rentals reveal thin unit economics; buyers shift to owned capacity; paper growth reverses. Funding tightens: special‑purpose vehicles seize up; “leased to self” structures stop pencilling; second‑tier names vanish. None of this requires a crash. It only requires gravity and time.
Positioning flows from that map. Underweight stories whose bull case is “everyone knows”; overweight the cash generators upstream—power, materials, boring picks and shovels—where real constraints live. Trade strength without needing applause. Exit into unanimous optimism. Re‑enter when dislocation offers resilience at a discount. That’s not cynicism. It’s adulthood.
Why the Mood Will Shift
Consultancies and banks will continue to pitch the opportunity in trillions; they sell picks and guidance into any gold rush. Boards will keep asking for transformation on a deadline. But boardroom mirages often hide grinding realities: the bill for inference, the talent costs, the data errors, the governance risk, the dull limits of real workflows. As costs climb and quick wins plateau, the psychology will tip. That isn’t the end of AI; it’s the end of this phase of fantasy. The steady state on the far side is better—boring products that actually earn their keep.
When that shift comes, watch the five dials. If breadth narrows to one theme, respect gravity. If credit widens while “AI” rallies, something under the surface is cracking. If the vol curve inverts into “good news,” risk is being mispriced. If leadership rotates back to defensives while headlines shout revolution, step aside. Markets telegraph regime change long before narratives admit it.
Clarity Without Cruelty
There’s no virtue in sneering at dreams. Some of the best work will survive this and shape the next decade. But money is a teacher, and it teaches with pressure. The projects that learn to live inside constraints—cost, energy, latency, accuracy, privacy—will keep climbing once the lights dim on the stage. The rest will blame the audience and hand back the set.
The AI hype cycle isn’t an enemy. It’s a season. The task is to see it as weather, not prophecy. Read state, not story. Follow cash, not slogans. In mania, hesitate; in panic, prepare. When mood splits and VPS goes quiet, resist the urge to guess direction; let the dials converge first . When the next true flush arrives, you won’t need a headline to tell you. Gravity is its own press release.
You don’t have to predict the end of the mirage. You just have to stop confusing scaffolding for steel. When belief finally meets the bill, the companies with receipts will still be standing. The rest were always paper, glittering in borrowed light.
