We Are Undoubtedly In An AI Bubble

The article argues that “AI beneficiaries” meet textbook criteria for an AI bubble, citing hyperscaler capex growth of over 80% to about $750B this year and weaker free-cash flow. However, it counters bubble fears with S&P 500 fundamentals: ~88% of Q2 earnings completed with blended y/y EPS growth near ~50%, versus 5-year/10-year EPS growth of 15.2%/11.2%, and expects ~30% EPS growth on a ~20x forward multiple. It also flags rate hikes as a potential “pricking” catalyst (with AI-related debt issuance projected to rise ~25% to ~$2.25T), concluding investors should stay positioned—using a more selective, barbell approach across AI infrastructure and software.

Analysis

The real tradeable split is not "AI vs no AI" but cash-rich platform owners versus supply-chain levered names whose earnings depend on hyperscaler willingness to keep pre-paying for capacity. AMZN and MSFT still have the cleanest path because their AI spend is absorbed inside broader franchise growth; the market will tolerate heavy capex as long as operating income and free cash flow re-accelerate within the next 1-2 quarters. By contrast, memory and component names such as SNDK and MU are more exposed to a stop-start capex cycle: a one-quarter pause can translate into a much larger air pocket in pricing, utilization, and multiple support.

The second-order winner is not the most hyped application names, but those with visible monetization and high switching costs. PLTR, NET, TEAM, and TWLO can keep working if AI features show up in net retention and sales efficiency, yet their rerating is fragile because they still trade on durable growth assumptions rather than near-term cash flow. If investors start demanding proof of AI payback, the low-quality "AI beneficiaries" should underperform first, while software with real workflow lock-in can outperform even in a de-risking tape.

The bubble thesis needs a catalyst, not just rich multiples: the likely prickers are higher-for-longer rates, a widening credit spread, or hyperscaler guidance that shows capex is outpacing monetization longer than the Street expects. Over the next 1-3 months, the key falsifier is continued upward revision to 2026 AI spending paired with stable margin guidance; over 6-18 months, the structural risk is that capital intensity rises faster than the revenue pool, compressing ROIC across the ecosystem. Bottom line: this is a dispersion market, not a clean index short.

Key Decisions for Investors

• Prefer a long AMZN/MSFT basket on pullbacks versus the broader Nasdaq for the next 1-3 months; these are the best insulated names if AI capex stays elevated because they can amortize spend across multiple profit pools. Falsifier: any sequential slowdown in cloud operating margin or free-cash-flow guidance.
• Initiate a tactical short basket in SNDK and MU, or buy 1-3 month put spreads, to express the view that memory is the first place AI capex excess turns into inventory and pricing pressure. Best entry is after a post-earnings rally or on renewed momentum in semis. Cover if hyperscaler capex commentary re-accelerates.
• Long PLTR/NET only on weakness, sized as a barbell against the hardware trade; these names have the best chance of converting AI interest into durable software spend, but they need proof in billings and net retention. If growth decelerates two quarters in a row, the multiple should compress quickly.
• Avoid a broad 'short AI' index expression until rates or credit actually tighten; bubble talk alone is not a catalyst. Watch 10Y yields and IG/HY spreads as the real triggers for de-rating over the next 1-6 months.
• If you want convexity, use downside put spreads on the most crowded high-beta AI names rather than outright shorts; the tape can remain momentum-driven longer than fundamentals justify. Reassess if implied volatility becomes cheap relative to realized and forward revisions turn negative.