Brief
The proximate trigger for this debate is Gartner's own research: its 2026 Hype Cycle for Generative AI places the category in the Trough of Disillusionment, and a separate, first-ever standalone Hype Cycle for Agentic AI, published April 2, 2026, places agentic AI at the Peak of Inflated Expectations heading toward its own trough. Gartner published its first standalone Hype Cycle for Agentic AI on 2 April 2026, with agentic AI sitting at the Peak of Inflated Expectations, careening toward the Trough of Disillusionment. Commentary built on Gartner's framing treats the sequence as confirmation of the model: Gartner's Hype Cycle report for GenAI, published in July 2025, predicted a specific scenario for GenAI, and now, almost nine months later, GenAI would enter the Trough of Disillusionment, which 2026 data overwhelmingly confirms.
The evidentiary anchor most commonly cited for the trough thesis is MIT's Project NANDA research. The numbers are brutal: 95% of enterprise GenAI pilots are failing according to MIT research published in 2025, representing analysis of over three hundred real implementations and organizations deeply interviewed. Proponents of the trough framing argue this is not fatal but developmental: companies that invest wisely during the trough will have a significant advantage when the slope of enlightenment arrives, and the companies that survive the trough will define the next era of business technology. On this reading, the trough is functioning exactly as the model predicts — disappointment concentrated among naive adopters, with disciplined ones positioned to capture the eventual slope.
A second camp argues the empirical picture looks nothing like a technology heading into a trough and cooling off — it looks like acceleration. The capital evidence is the crux: Goldman Sachs projects combined hyperscaler capex 2025–2027 will reach $1.15 trillion, more than double the $477B deployed across 2022–2024. Azure's AI-related revenue is growing at 39% year-over-year with over $120 billion in 2026 capex; Amazon is investing $200 billion in 2026, the largest single-company technology capital commitment in history; and Google Cloud grew 48% year-over-year in Q4 2025, the fastest growth rate among the three major cloud providers. Nvidia's own executives frame this explicitly as acceleration, not disillusionment: "Demand is accelerating," Jensen Huang, founder and CEO of Nvidia, said in a statement. Nvidia's CFO has pointed to "global VC funding in AI, roughly 70% of which is spent on compute, exceeded $400 billion in the first half of 2026, surpassing the $265 billion raised in all of 2025." and to "nearly 20 companies, including Cursor, Figma and Together AI, now exceed $1 billion in annualized run-rate revenue, up from 13 companies in Q4 of last year." If demand and capital deployment are both accelerating simultaneously, the argument goes, the technology cannot simultaneously be in a Gartner-defined trough, where the defining feature is that expectations and investment cool as disappointment sets in.
A third camp — spanning finance researchers, academic hype-cycle scholars, and some practitioners who use the model informally — argues the contested question is malformed because the framework doesn't map onto generative AI's cost and deployment structure at all. Academic literature on the model itself has long flagged this gap: despite its popularity for over 20 years, the scant academic literature has offered little insight beyond verifying the existence of hype cycle phenomena. A peer-reviewed longitudinal study of 46 Gartner-tracked technologies found incongruences connected with the reports of Gartner, thus questioning the reliable applicability of the hype cycle model. On the financial side, critics argue the capex-revenue relationship generative AI exhibits is structurally unlike a normal hype cycle: historically, bubbles narrowed the capex-revenue gap as adoption caught up, but in AI the opposite is happening — hyperscalers are doubling GPU orders despite sluggish AI revenue growth, meaning capex is accelerating while revenue growth is stalling. This camp also flags a circularity the hype-cycle model has no vocabulary for: a structural peculiarity critics highlight is that AI companies receive investment capital, immediately spend it on compute from cloud hyperscalers, which counts as revenue for the hyperscalers, which raises their valuations, which supports continued AI investment — money moving in a loop that can look like growth from the outside. Even practitioners who use the Gartner lens conversationally concede its limits: Gartner's Hype Cycle is a well-known model that tries to map the chaos, but frankly, it's a flawed model, though a useful one for framing the conversation.
What is not contested: enterprise pilot failure rates are high, hyperscaler capex is at an unprecedented scale relative to GDP, and Gartner itself has split generative AI and agentic AI into separate hype-cycle instruments rather than treating them as one trajectory — implicitly conceding the original single-curve framing undersold how differently sub-technologies within GenAI are behaving.
Positions (8)
GartnerForHigh leverage
Stated Position: Gartner maintains that generative AI is currently in the Trough of Disillusionment on its 2026 Hype Cycle, with a separately published Hype Cycle for Agentic AI showing that sub-technology at the Peak of Inflated Expectations heading toward its own distinct trough.
Underlying Interest: Gartner's advisory and research business is built on the hype-cycle methodology as a proprietary, licensable analytical product; validating the model's continued applicability to the defining technology of this cycle protects both its intellectual property and its credibility with enterprise clients making multi-year technology bets.
Gartner frames the trough placement as the natural result of early adopters hitting performance and ROI problems after initial excitement, consistent with the model's defined stages.
Leverage: Gartner's Hype Cycle reports are cited as the reference framework by trade press, vendors, and enterprise IT buyers, giving it outsized influence over how the market narrates and times its own AI investment decisions.
How It's Evolved: Gartner has moved from a single generative AI hype cycle to splitting out a standalone Agentic AI hype cycle in April 2026, an implicit concession that the original single curve could not capture how differently sub-technologies within the category are behaving.
Watch Signals:- [Likely] Gartner publishes further sub-category hype cycles (beyond generative AI and agentic AI) as it did in April 2026, splitting the category further rather than defending one curve.
- [Possible] Gartner revises its 2027 Hype Cycle placement for generative AI if enterprise ROI data materially improves or worsens versus the MIT NANDA findings.
- [Unlikely] Gartner abandons or substantially redesigns the hype-cycle methodology itself in response to academic critiques of its reliability.
Open Question
Can Gartner's Hype Cycle for Generative AI reconcile its Trough of Disillusionment placement with hyperscaler capex data showing accelerating, not cooling, capital deployment into the same technology category?
MIT Project NANDAMixedHigh leverage
Stated Position: MIT's Project NANDA research documents that despite tens of billions in enterprise investment, the overwhelming majority of generative AI pilots have produced no measurable P&L return, a finding widely cited as empirical support for the trough narrative.
Underlying Interest: As an academic research initiative, NANDA's credibility rests on methodological neutrality; its findings are used by both trough-camp and framework-skeptic commentators to support opposing conclusions, which is itself evidence the underlying data doesn't map cleanly onto the hype-cycle stages.
The research is presented as a rigorous empirical measurement of enterprise deployment outcomes rather than a position on the hype-cycle model itself.
Leverage: The study's data — a 95% enterprise pilot failure rate based on analysis of over three hundred real implementations — has become the single most-cited statistic in the entire debate, meaning any revision or extension of this research directly moves the public narrative.
How It's Evolved: Stable since the original 2025 study; commentators have increasingly folded the finding into hype-cycle narratives Gartner did not originally construct it to support.
Watch Signals:- [Possible] Project NANDA publishes a follow-up 2026 or 2027 study tracking whether the same cohort of enterprises has moved pilots into production.
- [Possible] Rival research groups attempt to replicate or challenge the NANDA methodology given how heavily it is now cited across the debate.
Open Question
Does MIT Project NANDA's 95% enterprise generative AI pilot failure finding indicate a temporary trough consistent with Gartner's model, or a structural cost-and-integration problem that a hype-cycle framework is not designed to diagnose?
NvidiaAgainstHigh leverage
Stated Position: Nvidia's leadership states that AI demand is accelerating, not cooling, and that the technology is already producing profitable, productive enterprise work rather than chasing a future milestone — a picture inconsistent with a technology sliding into a Gartner-defined disillusionment trough.
Underlying Interest: Nvidia's valuation and continued hyperscaler capex commitments depend on the market believing AI demand is durable and still ascending rather than entering a cooling phase; a trough narrative directly threatens capital markets' willingness to fund the buildout that drives Nvidia's data-center revenue.
Nvidia points to concrete customer and revenue evidence: nearly 20 companies now exceeding $1 billion in annualized run-rate revenue on its platforms, and named enterprise deployments in trading, drug discovery, and computational lithography as evidence of productive, not disillusioned, deployment.
Leverage: As the dominant supplier of AI accelerators, Nvidia's quarterly earnings calls and demand commentary function as the most closely watched real-time indicator of AI investment health across the entire supply chain.
How It's Evolved: Nvidia's public framing has hardened from cautious optimism to explicit acceleration language across 2026 earnings calls, directly countering the popular narrative that AI adoption has peaked.
Watch Signals:- [Likely] Nvidia's subsequent quarterly earnings calls continue emphasizing accelerating demand language to counter bubble and trough narratives.
- [Possible] A material slowdown in Nvidia data-center revenue growth in a future quarter would validate the trough camp's reading and undercut Nvidia's own framing.
Open Question
If Nvidia's data-center demand and named enterprise revenue milestones continue accelerating through 2026 and 2027, can Gartner's Trough of Disillusionment placement for generative AI survive as an accurate description of the same period?
Hyperscalers (Amazon, Microsoft, Alphabet, Meta)AgainstHigh leverage
Stated Position: The major cloud and platform companies are not behaving as though generative AI is in a cooling, disillusioned phase: collectively they are guiding toward the largest capital expenditure commitments in corporate history, explicitly betting on continued rather than declining demand.
Underlying Interest: Having already committed hundreds of billions in capex and, in some cases, taken on significant new debt and halted share buybacks to fund it, hyperscalers have a direct financial interest in the market believing AI demand justifies this spending rather than concluding the category has peaked prematurely.
Each hyperscaler frames its capex as necessary infrastructure investment to meet current and forecast AI workload demand, not as speculative overbuilding against a fading trend.
Leverage: Their combined capex decisions set the demand floor for the entire AI hardware and infrastructure supply chain, making their spending behavior itself a data point that outweighs any single analyst framework.
How It's Evolved: Capex commitments have escalated sharply through 2026, with Alphabet and Meta halting share buybacks and Apple and Microsoft scaling theirs back as the group shifts to a capital-intensive model — Alphabet turned free cash flow negative in the second quarter of 2026 for the first time, and Amazon's long-term debt jumped 81% to $119 billion in the first quarter alone.
Watch Signals:- [Likely] Continued quarterly capex guidance increases or holds through the remainder of 2026 despite trough narratives in the press.
- [Possible] A hyperscaler capex pullback or guidance cut in a future quarter would be the clearest falsification signal for the 'still ascending' camp.
- [Possible] Further increases in hyperscaler long-term debt issuance, given Morgan Stanley and J.P. Morgan's estimate that the sector will need to issue roughly $1.5 trillion in new debt over three years.
Open Question
Can Amazon, Microsoft, Alphabet, and Meta sustain capex guided toward a combined $1.15 trillion from 2025 through 2027 if enterprise generative AI adoption follows the slower, trough-then-slope trajectory Gartner's model predicts rather than continuing to accelerate?
U.S. Federal ReserveUndeclaredModerate leverage
Stated Position: The Federal Reserve has flagged AI-linked capital expenditure concentration as a systemic financial risk, without taking an explicit position on whether generative AI itself is in a hype-cycle trough, still ascending, or on some other trajectory.
Underlying Interest: As a financial stability regulator, the Fed's institutional interest is in flagging concentration and leverage risk regardless of which hype-cycle camp is correct — its silence on the trajectory question itself is a structural feature of its mandate, not evidence for either camp.
The Fed's concern is framed narrowly around financial contagion risk if AI capex contracts sharply, not around the underlying technology's maturity curve.
Leverage: Federal Reserve statements on systemic risk directly move credit markets and can influence the cost and availability of the debt financing hyperscalers are increasingly relying on to fund AI capex.
How It's Evolved: Stable — the Fed's public commentary has centered on capex-driven financial contagion risk rather than adjudicating the hype-cycle debate directly.
Watch Signals:- [Possible] Further Federal Reserve financial stability reports naming AI capex concentration as a monitored risk category.
- [Possible] Statements from Fed officials specifically addressing the capex-to-revenue gap that hype-cycle-skeptic researchers have flagged.
Open Question
Would the Federal Reserve's identification of AI capital expenditure concentration as a systemic risk change if generative AI enterprise ROI data materially improves, or is the Fed's concern independent of where the technology sits on any hype-cycle framework?
Academic hype-cycle researchers (innovation-diffusion scholars)AgainstLow leverage
Stated Position: Peer-reviewed research on the Gartner Hype Cycle model itself, predating the generative AI debate, has repeatedly found that the model's stage placements do not reliably correspond to empirical technology-adoption data, questioning whether it is fit for purpose for any single technology, including generative AI.
Underlying Interest: Academic researchers in technology innovation management have an intellectual and disciplinary interest in establishing a more rigorous, falsifiable model of technology adoption than a proprietary consulting framework that has faced limited independent scientific validation.
Longitudinal comparison of Gartner's reported technology placements against independent adoption and attention data found systematic mismatches rather than confirmation of the model's stages.
Leverage: Peer-reviewed critique doesn't move markets directly, but it supplies the intellectual ammunition that finance researchers, journalists, and skeptical practitioners draw on when arguing the hype-cycle framework itself is the wrong lens.
How It's Evolved: Stable — this critique predates generative AI by over a decade and has simply been re-applied to the current debate rather than newly formulated.
Watch Signals:- [Possible] New peer-reviewed studies specifically testing Gartner's generative AI and agentic AI hype-cycle placements against independent adoption or search-attention data.
- [Unlikely] A definitive academic consensus forms either validating or rejecting the hype-cycle model's applicability to generative AI specifically.
Open Question
Does the systematic mismatch between Gartner's historical hype-cycle placements and independent adoption data found in prior peer-reviewed studies extend to its current placement of generative AI in the Trough of Disillusionment?
AI capex bears (financial researchers, e.g. Man Group)AgainstModerate leverage
Stated Position: Institutional financial researchers argue that the generative AI capex cycle exhibits a capex-revenue divergence that is the opposite of what typically resolves a hype-cycle trough, meaning the technology is not simply cycling through Gartner's stages but is exhibiting a distinct, potentially unsustainable financial pattern.
Underlying Interest: As institutional investment researchers, firms like Man Group have a mandate to flag risk to their own capital allocation decisions and clients, giving them incentive to name structural anomalies (debt-funded buildout, circular capital flows) that a simple hype-cycle narrative would smooth over.
Historically, technology bubbles narrow their capex-revenue gap as adoption catches up with prior overbuilding; the current AI cycle shows the gap widening even as spending accelerates, which the standard hype-cycle vocabulary has no mechanism to explain.
Leverage: Institutional research of this kind directly informs asset allocation decisions among large investors exposed to AI infrastructure equities and debt, giving it real influence over capital flows into the sector regardless of which narrative dominates retail or trade press.
How It's Evolved: This critique has sharpened through 2026 as hyperscaler debt issuance and negative free cash flow at Alphabet became visible in company financials, moving from a theoretical concern to one grounded in reported balance-sheet data.
Watch Signals:- [Likely] Continued institutional research reports tracking the hyperscaler capex-to-revenue ratio through subsequent 2026 and 2027 quarters.
- [Possible] A credit-rating action or debt-market repricing event tied to hyperscaler AI-related debt issuance.
Open Question
If hyperscaler capex continues accelerating while AI cloud revenue growth stalls relative to spending, does that pattern falsify the hype-cycle model's applicability to generative AI, or does it simply describe an unusually long trough phase?
Enterprise CIOs / IT buyersMixedHigh leverage
Stated Position: Enterprise technology buyers broadly report continuing, and in some categories accelerating, generative and agentic AI adoption, even as they acknowledge a low rate of pilots reaching measurable production ROI — a position that partially confirms and partially contradicts the trough narrative simultaneously.
Underlying Interest: CIOs face internal pressure to show AI ROI to boards and shareholders while managing real budget and governance constraints, giving them incentive to describe the technology as maturing and improving (consistent with a slope-of-enlightenment narrative) even when production deployment lags pilot activity.
Buyers describe moving from experimentation to deployment in specific use cases (coding, legal, financial tasks) while citing governance, cost predictability, and integration difficulty as the practical barriers slowing broader rollout.
Leverage: Aggregate enterprise purchasing decisions are the ultimate demand signal that determines whether hyperscaler and vendor capex bets are validated or stranded, making this the single most consequential stakeholder group for resolving the debate empirically.
How It's Evolved: Survey data shows agentic AI adoption specifically rising from an experimentation phase in 2025 to full deployment in select use cases by early 2026, per Nvidia's own enterprise survey research.
Watch Signals:- [Likely] Further enterprise AI adoption surveys (Nvidia's State of AI series, Gartner CIO surveys) tracking the pilot-to-production conversion rate through 2026 and 2027.
- [Possible] A widening or narrowing of the MIT NANDA-documented 95% pilot failure rate in any updated study.
Open Question
Will enterprise CIOs' documented shift from AI pilot experimentation to production deployment in specific use cases in early 2026 continue at a pace consistent with a slope-of-enlightenment trajectory, or stall in a way that confirms a prolonged trough?