Bubble Bubble
Jensen Huang continues to reject the idea of an imminent AI bubble. He sees no classic bubble forming over the next 5–10 years for a simple reason: infrastructure constraints (chips, power, data centers) still leave demand far ahead of supply. Wall Street, in his view, keeps misreading the buildout. Free or cheap Chinese models such as DeepSeek and Kimi do not reduce demand for Nvidia hardware; they accelerate it.
Huang points out that AI is already turning profitable in high-value areas like coding agents, and the economic value of AI tokens will keep rising as models grow more capable. He believes the semiconductor industry needs roughly 5–10× growth over the next decade just to meet demand. The real risk is not a bubble, but under-investment or fear that slows adoption.
On regulation he draws a clean line: regulate the applications (medicine, autonomous vehicles, high-stakes uses), not the underlying technology. AI is dual-use. Advancing it quickly is safer than throttling it. Excessive fear-based regulation risks leaving America behind in an arms race where both U.S. and Chinese officials privately prefer a balance of power.
Of course the current correction can run deep especially since some AI-related names have skyrocketed. Markets never rise in a straight line. During the dot-com boom, the NASDAQ had a severe correction in late 1998 taking down tech names with it. Anyone who stayed invested would have suffered large losses.
In today's market, many stocks such as MU and SNDK have posted massive gains. Such names can easily lose one-half to two-thirds of their value in weeks after boasting 5-10x gains. A 45% pullback looks minor on a log chart.
That said, timeframe is everything. Longer-term holders who respect the 50-day moving average often ride through 33–45% corrections and still capture the bulk of the advance. Shorter-term traders take smaller bites and avoid the deeper drawdowns. As Bill O’Neil and Livermore both understood, the big money is usually made by sitting, not by constant trading. But for those who have longer timeframes such as the 50-dma on their sell stops, even such investors would be largely in cash by now with ample profits.
So while the stocks that gained the most can easily lose half their value in just a few weeks, the broader markets continue to trend higher overall such as seen in the NYSE Composite which is near new highs.
On the short side, continue to keep an eye out for weak rallies into resistance for potential short-sale entries. On the long side, keep watch on any formations in the coming days and weeks such as constructive volume dry-ups, undercut & rallies, pocket pivots, and buyable gap ups. Let the market come to you. Historically, true bubble bursts end multi-year uptrends and produce deep, multi-year bear markets (1929–1932, 2000–2003). The average bull market lasts 3 years. We are 3 1/2 years into this one.
Meanwhile the fundamentals keep improving. The S&P 500’s net profit margin is tracking toward 15.7% for Q2 2026 — the highest since at least 2009.
Estimated 2027 EPS has cracked $400, a roughly 4× rise since 2012. And valuation multiples are nowhere near the 2000 extremes: the Nasdaq 100 traded at 86× earnings in May 2001 (well after the peak in Mar 2000); it sits near 24× today.
Critics will keep insisting AI is not yet earning enough to justify current prices. That argument mistakes the early innings of a multi-year infrastructure and productivity buildout for a finished product. Further, AI adoption is high enough to matter. Quality-adjusted AI output grew over 2,000% per year since 2024, while labor productivity growth has been materially stronger compared to pre-AI, with higher-AI-adoption sectors showing faster productivity gains.
Singularly, the companies spending the most intelligently today are positioning themselves for the profits that arrive once the capacity is in place and the applications scale. In every major technology wave, the biggest returns went to those who built through the skepticism, not those who waited for perfect current-period earnings. That said, AMZN was a loss leader before and after the dot-com bubble burst. It was also a price leader before and after. There is no reason why leading tech names can't find their footing but they could also continue deeper into their bear markets or trade sideways as the AI capex vs productivity debate continues. Capital flow thus price/volume rules the day.
Lastly, the Fed kept a firm stance on focusing on 2% inflation as Warsh testified, so CME FedWatch futures now price in a 66% chance of a rate hike when they next meet in September. In consequence, stock market indices and bonds plunged intraday to close near the day's lows.
Jensen Huang continues to reject the idea of an imminent AI bubble. He sees no classic bubble forming over the next 5–10 years for a simple reason: infrastructure constraints (chips, power, data centers) still leave demand far ahead of supply. Wall Street, in his view, keeps misreading the buildout. Free or cheap Chinese models such as DeepSeek and Kimi do not reduce demand for Nvidia hardware; they accelerate it.
Huang points out that AI is already turning profitable in high-value areas like coding agents, and the economic value of AI tokens will keep rising as models grow more capable. He believes the semiconductor industry needs roughly 5–10× growth over the next decade just to meet demand. The real risk is not a bubble, but under-investment or fear that slows adoption.
On regulation he draws a clean line: regulate the applications (medicine, autonomous vehicles, high-stakes uses), not the underlying technology. AI is dual-use. Advancing it quickly is safer than throttling it. Excessive fear-based regulation risks leaving America behind in an arms race where both U.S. and Chinese officials privately prefer a balance of power.
Of course the current correction can run deep especially since some AI-related names have skyrocketed. Markets never rise in a straight line. During the dot-com boom, the NASDAQ had a severe correction in late 1998 taking down tech names with it. Anyone who stayed invested would have suffered large losses.
In today's market, many stocks such as MU and SNDK have posted massive gains. Such names can easily lose one-half to two-thirds of their value in weeks after boasting 5-10x gains. A 45% pullback looks minor on a log chart.
That said, timeframe is everything. Longer-term holders who respect the 50-day moving average often ride through 33–45% corrections and still capture the bulk of the advance. Shorter-term traders take smaller bites and avoid the deeper drawdowns. As Bill O’Neil and Livermore both understood, the big money is usually made by sitting, not by constant trading. But for those who have longer timeframes such as the 50-dma on their sell stops, even such investors would be largely in cash by now with ample profits.
So while the stocks that gained the most can easily lose half their value in just a few weeks, the broader markets continue to trend higher overall such as seen in the NYSE Composite which is near new highs.
On the short side, continue to keep an eye out for weak rallies into resistance for potential short-sale entries. On the long side, keep watch on any formations in the coming days and weeks such as constructive volume dry-ups, undercut & rallies, pocket pivots, and buyable gap ups. Let the market come to you. Historically, true bubble bursts end multi-year uptrends and produce deep, multi-year bear markets (1929–1932, 2000–2003). The average bull market lasts 3 years. We are 3 1/2 years into this one.
Meanwhile the fundamentals keep improving. The S&P 500’s net profit margin is tracking toward 15.7% for Q2 2026 — the highest since at least 2009.
Estimated 2027 EPS has cracked $400, a roughly 4× rise since 2012. And valuation multiples are nowhere near the 2000 extremes: the Nasdaq 100 traded at 86× earnings in May 2001 (well after the peak in Mar 2000); it sits near 24× today.
Critics will keep insisting AI is not yet earning enough to justify current prices. That argument mistakes the early innings of a multi-year infrastructure and productivity buildout for a finished product. Further, AI adoption is high enough to matter. Quality-adjusted AI output grew over 2,000% per year since 2024, while labor productivity growth has been materially stronger compared to pre-AI, with higher-AI-adoption sectors showing faster productivity gains.
Singularly, the companies spending the most intelligently today are positioning themselves for the profits that arrive once the capacity is in place and the applications scale. In every major technology wave, the biggest returns went to those who built through the skepticism, not those who waited for perfect current-period earnings. That said, AMZN was a loss leader before and after the dot-com bubble burst. It was also a price leader before and after. There is no reason why leading tech names can't find their footing but they could also continue deeper into their bear markets or trade sideways as the AI capex vs productivity debate continues. Capital flow thus price/volume rules the day.
Lastly, the Fed kept a firm stance on focusing on 2% inflation as Warsh testified, so CME FedWatch futures now price in a 66% chance of a rate hike when they next meet in September. In consequence, stock market indices and bonds plunged intraday to close near the day's lows.