CITIC says AI cycle mirrors 2021 new energy, not dot-com bubble

The current AI trading cycle resembles the 2021 new energy boom more than the 2000 dot-com bubble, with upstream valuations yet to peak and earnings estimates just beginning to rise, according to CITIC Securities.

The AI investment cycle that began in June 2025 shares structural similarities with the 2021 new energy (electricity and new materials) cycle rather than the 2000 internet bubble, CITIC Securities said in a research report published late June. The brokerage identified four signals that would mark a top — and argued none have materialized yet.

"The current AI cycle is analogous to mid-2021 for new energy," the report said. "Upstream dynamic P/E ratios have not yet collapsed, and earnings expectations are only beginning to be revised upward."

CITIC's framework compares the AI supply chain to the 2021 new energy cycle, where the most commoditized input — electrolyte in 2021, silicon wafers today — serves as the canary in the coal mine. When prices for the least-scarce material peak, the cycle top is near. Silicon wafer pricing has not yet shown that signal, the report said.

The Four Top Signals

CITIC identified four conditions that would signal a peak in AI-related stocks. First, price peaks in the least-scarce upstream commodity — silicon wafers in this cycle, analogous to electrolyte in 2021. Second, widespread price increases and cost complaints from downstream buyers. Third, a rising density of overseas capital expenditure disclosures. Fourth, a collapse in crowding metrics and market breadth.

None of these conditions are present today, the report said, suggesting the AI trade still has room to run. The brokerage recommended favoring upstream stocks with low valuations and tight supply, as their risk-reward profile improves the longer the cycle extends.

Downstream and Cross-Sector Beneficiaries

CITIC also expressed a preference for downstream beneficiaries. The report named specific sub-sectors across the silicon and carbon-silicon supply chains: memory chips, gas turbine chains, optical modules, printed circuit boards, and cloud providers. In the carbon-plus-silicon category, the brokerage highlighted computing metals, fluorine chemicals, and phosphorus chemicals as beneficiaries of AI infrastructure demand.

The report's sector-level recommendations span multiple industries, reflecting the breadth of the AI buildout. Memory chips benefit from higher-bandwidth demand in AI servers. Gas turbine chains gain from rising electricity consumption at data centers. Optical modules and PCBs are direct beneficiaries of data center interconnect upgrades. Cloud providers capture the recurring revenue from AI inference workloads.

Investment Implications

For investors, the CITIC framework suggests the AI trade is in its middle innings rather than its final act. If the analogy to the 2021 new energy cycle holds, upstream suppliers with pricing power and constrained capacity could see the largest relative gains as the cycle matures. Downstream names offer more diversified exposure but may lag in a late-cycle rotation.

The key risk is timing: the 2021 new energy cycle peaked roughly 18 months after CITIC's mid-cycle analog, but the duration of AI's infrastructure buildout depends on capital spending trajectories from hyperscale cloud providers. Any slowdown in CapEx from Microsoft, Amazon, Alphabet, or Meta could accelerate the arrival of CITIC's four top signals.

This article is for informational purposes only and does not constitute investment advice.