Key Takeaways
- • Citadel founder Ken Griffin said there are "almost certainly bubbles within this moment in time" in AI stocks, telling Bloomberg on October 20 at Citadel's Future of Global Markets Conference in New York that transformative technologies are generally intertwined with bubbles.
- • Alphabet, Amazon, Microsoft and Meta guided to more than $700 billion of 2026 capital expenditure in April, about $100 billion above expectations a quarter earlier, per Sherwood News.
- • The Shiller CAPE reached about 41 in September, the highest since the dot-com era and close to the 44.2 record set in December 1999, according to CryptoBriefing as reported by KuCoin.
- • UBS projects hyperscalers will spend $4.1 trillion on AI infrastructure from 2026 to 2028, while cited cloud backlog across Amazon, Alphabet, Microsoft and Oracle totals $2.3 trillion, according to 24/7 Wall St.
- • Investors should diversify and rebalance concentrated AI exposure, and executives should require clear return measures on AI capital spending, the article argues.
Summary
Ken Griffin, founder and chief executive of Citadel, has added his voice to the debate over AI valuations. Speaking to Bloomberg on the sidelines of Citadel's Future of Global Markets Conference in New York on October 20, he said there are "almost certainly bubbles within this moment in time," adding that "any radical transformation of our economy is generally intertwined with a bubble." His view is more nuanced than a sell signal: he argued that bubbles often accompany genuinely transformative technologies and that companies that survive them can lead markets for decades. Reflecting on the internet era, he told Bloomberg, "I just wish I'd kept my Amazon stock in that moment."
The scale of spending behind the debate is substantial. In April, Alphabet, Amazon, Microsoft and Meta together guided to more than $700 billion of capital expenditure for 2026, roughly $100 billion more than expected a quarter earlier, according to Sherwood News. Amazon guided to about $200 billion, Microsoft to about $190 billion, Alphabet to $180 billion to $190 billion and Meta to $125 billion to $145 billion. Most of this goes into data centers, chips and power infrastructure. According to 24/7 Wall St., UBS projects $4.1 trillion of hyperscaler AI infrastructure spending from 2026 to 2028, and the same report cites $2.3 trillion of cloud backlog across Amazon, Alphabet, Microsoft and Oracle. The open question is whether AI product revenue will grow quickly enough to earn an adequate return on this capital. Griffin raised that point, comparing AI computing capacity to airline seats, which lose pricing power once supply exceeds demand, and questioning whether businesses are getting enough value from expensive AI models, per 24/7 Wall St.
Valuations are near historic extremes by several measures. The S&P 500's Shiller CAPE, which compares prices with ten years of inflation-adjusted earnings, reached about 41 in September, according to CryptoBriefing as reported by KuCoin. That is the highest since the dot-com era and close to the record 44.2 of December 1999. High valuations do not predict the timing of a decline, but they have historically been associated with lower long-term returns from current prices.
The dot-com comparison cuts both ways. The S&P 500 eventually surpassed its 2000 peak, and Amazon, adjusted for splits, traded near $5 a share at its dot-com peak, fell below $1 after the bust, and now trades above $260, according to the Motley Fool. The downside record is also severe: from its March 2000 peak to its October 2002 low, the S&P 500 fell about half and the Nasdaq Composite fell close to 80%, and the Nasdaq took about 15 years to regain its 2000 high. The article notes that many internet companies failed outright or never recovered their peak valuations.
The article argues the AI cycle differs in important ways. Leading companies are highly profitable, generate large cash flows and fund much of their investment internally, unlike many dot-com firms that depended on equity markets. However, risks are larger in some respects: spending is concentrated in a few companies that carry outsized weight in major indices, capital intensity is far higher, and multi-year power, chip and data center commitments are hard to reverse. Paul Tudor Jones has flagged a wave of new AI equity supply as a risk for 2027. Griffin's comments do not mean Citadel has exited AI stocks. As of June 30, its amended filing showed holdings of about $2.6 billion in Amazon, $2.4 billion in Nvidia, $1.4 billion in Microsoft, and about $1.1 billion each in Broadcom and Alphabet, according to 24/7 Wall St. Those are snapshot figures, and Citadel also reported options positions that may offset exposure.
The scale of spending behind the debate is substantial. In April, Alphabet, Amazon, Microsoft and Meta together guided to more than $700 billion of capital expenditure for 2026, roughly $100 billion more than expected a quarter earlier, according to Sherwood News. Amazon guided to about $200 billion, Microsoft to about $190 billion, Alphabet to $180 billion to $190 billion and Meta to $125 billion to $145 billion. Most of this goes into data centers, chips and power infrastructure. According to 24/7 Wall St., UBS projects $4.1 trillion of hyperscaler AI infrastructure spending from 2026 to 2028, and the same report cites $2.3 trillion of cloud backlog across Amazon, Alphabet, Microsoft and Oracle. The open question is whether AI product revenue will grow quickly enough to earn an adequate return on this capital. Griffin raised that point, comparing AI computing capacity to airline seats, which lose pricing power once supply exceeds demand, and questioning whether businesses are getting enough value from expensive AI models, per 24/7 Wall St.
Valuations are near historic extremes by several measures. The S&P 500's Shiller CAPE, which compares prices with ten years of inflation-adjusted earnings, reached about 41 in September, according to CryptoBriefing as reported by KuCoin. That is the highest since the dot-com era and close to the record 44.2 of December 1999. High valuations do not predict the timing of a decline, but they have historically been associated with lower long-term returns from current prices.
The dot-com comparison cuts both ways. The S&P 500 eventually surpassed its 2000 peak, and Amazon, adjusted for splits, traded near $5 a share at its dot-com peak, fell below $1 after the bust, and now trades above $260, according to the Motley Fool. The downside record is also severe: from its March 2000 peak to its October 2002 low, the S&P 500 fell about half and the Nasdaq Composite fell close to 80%, and the Nasdaq took about 15 years to regain its 2000 high. The article notes that many internet companies failed outright or never recovered their peak valuations.
The article argues the AI cycle differs in important ways. Leading companies are highly profitable, generate large cash flows and fund much of their investment internally, unlike many dot-com firms that depended on equity markets. However, risks are larger in some respects: spending is concentrated in a few companies that carry outsized weight in major indices, capital intensity is far higher, and multi-year power, chip and data center commitments are hard to reverse. Paul Tudor Jones has flagged a wave of new AI equity supply as a risk for 2027. Griffin's comments do not mean Citadel has exited AI stocks. As of June 30, its amended filing showed holdings of about $2.6 billion in Amazon, $2.4 billion in Nvidia, $1.4 billion in Microsoft, and about $1.1 billion each in Broadcom and Alphabet, according to 24/7 Wall St. Those are snapshot figures, and Citadel also reported options positions that may offset exposure.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.