Why the AI economy is like a bad dating app — drowning in decks, pilot purgatory — and it’s playing out like the dotcom bubble, except worse

The article warns of a potential “reckoning” in corporate AI spending as companies ramp generative AI pilots without measurable outcomes, with Webb expecting cracks as early as next year. It cites Bain survey data: ~40% of companies targeting 11%–20% AI cost savings landed below 10%, while 90% still plan to increase AI budgets—suggesting enthusiasm outpacing realized returns. The overall message is that AI adoption may deliver cheap experimentation upfront but compounds costs later (integration, legal/IT, and productivity/decision overhead), increasing risk of missed targets and renewed scrutiny from Wall Street.

Analysis

The market is likely still underpricing the distinction between AI spend and AI monetization. The near-term beneficiaries remain the infrastructure toll collectors — cloud, networking, and compute — because enterprises can keep paying for experimentation even when workflows never harden. That favors AMZN relative to software names: AWS can harvest pilot churn, but the more important issue is whether the current wave of AI workloads becomes durable consumption or just a noisy pull-forward that later lapses into budget scrutiny.

The first real pain point should show up in discretionary software, consulting, and internal tools budgets, not in headline AI capex. If executives start demanding ROI after two failed planning cycles, the weakest models will be high-multiple vendors whose value proposition depends on "productivity" rather than hard cost takeout. That sets up a 1-3 month catalyst around earnings calls and 2026 budget planning, where commentary on renewal rates, seat expansion, and pilot conversion matters more than raw AI adoption language.

The contrarian read is that this is less a burst-bubble setup than a reallocation event. Spending likely shifts from pilot-heavy experimentation toward integration, security, data plumbing, and workflow orchestration. That means the obvious shorts may be the wrong ones if the selloff hits the wrong layer of the stack; the structural loser is not AI itself, but vendors selling undifferentiated features without embedded distribution. The thesis breaks if management teams begin quantifying realized savings or if cloud AI revenue remains structurally above baseline through the next two earnings seasons.

Key Decisions for Investors

• Favor AMZN over high-multiple software proxies: long AMZN / short WCLD into the next 1-3 earnings cycles; thesis is that infrastructure monetizes pilot churn even if application ROI disappoints.
• Short a basket of enterprise software names with AI-heavy narratives but weak operating leverage versus cash generation (e.g., SNOW/DDOG/HUBS) over the next 1-3 months; look for guide cuts or softer net retention, with 15-25% downside if budget scrutiny surfaces.
• Use IGV as the cleaner macro hedge: buy IGV put spreads into Q3/Q4 earnings if you want expression on 'pilot purgatory' without single-name blowup risk; risk/reward is best if software multiple compression starts before reported deceleration.
• Watch AWS and cloud commentary closely; if AMZN management quantifies AI conversion to production workloads or reaccelerating AI consumption, cover short software exposure and rotate back into the compute layer.
• No aggressive short on AMZN unless AWS growth or AI attach rates visibly stall for two consecutive quarters; otherwise the stock may benefit from the spending itself even if end-demand ROI is weak.