CITIC SEC has released a research report maintaining its view that the Federal Reserve will not raise interest rates in the medium term. It suggests that sectors currently discounted due to liquidity concerns, specifically those with long-term carbon and silicon-based pricing, have room for valuation recovery.

Regarding direct-exposure, extreme silicon-based cyclical products, discussing a bubble burst now is somewhat premature. However, after valuations have been elevated, performance becomes the key battleground, moving out of a comfortable trading range. The preference shifts towards sectors with strong supply constraints and low valuations, narrowing the focus. From an industry chain perspective, mid-to-downstream segments are more recommended.

Consequently, the firm is most optimistic about sectors with longer narratives, AI exposure yet discounted by macro factors—carbon-based cyclical products like computing metals, fluorochemicals, and phosphorus chemicals—as well as low-valuation, mid-to-downstream segments such as the storage chain, gas turbine chain, optical modules, PCBs, and cloud service providers.

Comparing AI to the Lithium Battery Cycle

Why does this AI wave resemble the lithium battery cycle more than the dot-com bubble? Both cycles belong to a "cyclical product rally under accelerating penetration rates." The difference is that new energy vehicles were driven directly by consumer demand, while AI involves indirect transmission: token consumption acceleration → Capex → mid-to-upstream hardware. This makes it more sensitive to penetration rate fluctuations with more acceleration points.

The volatility's essence in both is "supply-demand gap + price increases," fitting a cyclical stock framework rather than a valuation framework. Unlike the dot-com bubble driven purely by valuation expansion (Nasdaq forward PE of 60x), both lithium batteries and AI are built on substantial Capex/consumption and real high profitability, with current valuations not showing significant froth (Nasdaq dynamic PE under 30x).

Both can be broken down into upstream, midstream, and downstream sub-sectors (resources → materials/components → end applications). In terms of cycle stage, AI is approximately at the 2021 point of the new energy cycle—the midstream (e.g., Nvidia) led the initial surge, while the upstream (e.g., storage/materials) is not expected to ignite until late 2025.

Medium-Term Performance and Valuation Outlook

Using an aggregate method to calculate market cap changes through a full cycle, the mid-cycle typically sees upstream segments far outperform mid-downstream before eventually regressing. The starting point is when industry capacity becomes large enough to impact the supply-demand balance sheet (lithium batteries: Sep 2020–Sep 2021; AI: post-June 2025 for materials, with amplitude and coverage already exceeding the new energy cycle).

More crucially, the market cap peak for the new energy sector led the earnings peak by about six quarters. Core mid-upstream companies saw their aggregate calculated PE compress from 137x to 18x during the earnings surge, with most genuine peaks appearing in Q3 2021. This indicates the selling point is not at earnings/gross margin realization, but when market cap, valuation, and positioning first loosen.

Currently, the global AI mid-upstream aggregate PE_ttm has retreated from 100x to around 40x, similar to mid-2021 in the new energy cycle. Earnings will still be realized, but further valuation expansion is unlikely. Furthermore, sell-side expectations show significant bias. According to Visible Alpha consensus, analyst models are consistently more conservative on upstream early in the cycle and more aggressive later. Once upstream profit/ASP consensus is大幅上调, the stock price is often near its peak.

For AI, foreign consensus earnings expectations for mid-upstream are not projected to see widespread上调 until the first half of 2026, and the dynamic PE for mid-upstream has not yet begun to collapse, reflecting stronger earnings爆发力 and tighter medium-term supply expectations this cycle.

Short-Term Monitoring Indicators

Four key signals for a potential top: 1) Closely watch the price peak of the "least tight"品种. In the new energy cycle, prices for the tightest commodities like lithium carbonate and cobalt lagged stock prices by nearly a year, offering little reference. The真正的领先 indicators were less tight品种 (e.g., electrolyte in 2021: the loosest supply-demand, its price first flattened in Q4 2021 as demand growth marginally slowed, with stock prices weakening simultaneously).

Mapping to AI: DRAM ≈ lithium carbonate (direct exposure, restrained capacity, leading price increases). Silicon wafers ≈ electrolyte (latest and smallest price increases, structural oversupply, large consumption exposure). As NAND spot price increases have already shown边际放缓, the focus should be on硅片.一旦 it enters a price plateau, it could signal a short-term top.

2) Monitor price博弈 between upstream and downstream. Upstream stock prices in the new energy cycle peaked in Q4 2021. Tesla's first price hike was in late March 2021, with密集提价 by domestic and international automakers and长城汽车欧拉停产 concentrated in Q4 2021–Q1 2022. For AI, watch for events like Apple raising prices on new models and its procurement price博弈 with美光 on memory, anticipated around June 2026.

3) Observe the density of Capex announcements. In the new energy cycle, this explained upstream stock price movements far better than commodity prices (disclosure immediately impacts medium-term supply-demand expectations). However, this AI cycle is overseas-led with harder expansion and tighter monetary conditions. Domestic Capex announcements may have weaker forward-looking power than in the new energy cycle. Large-scale overseas capacity expansion would be a warning signal.

4) Finally, the坍缩 of拥挤度 and扩散度 signals the end, neither of which has been triggered yet.拥挤度 is a necessary but not sufficient indicator; its松解 marks the终点.扩散度 looks at the proportion of stocks hitting new highs (its turn领先 the price peak). The new energy cycle saw a peak新高占比 of 62%,领先 the price peak by 3–4 months. While AI拥挤度 is approaching historical highs, the新高占比 has not yet declined, and leader抱团 remains intact.

Identifying Sectors with Better Risk-Reward

Two patterns dictate risk-reward in the latter half: First, the larger the泡沫涨幅前期, the deeper the fall post-burst. Second, in the new energy cycle, mid-stream processing materials with low barriers and easy capacity expansion saw their average market cap at lows within two years post-peak仅为约 21% of the peak. For upstream resources with supply/quota constraints and slow expansion, this ratio was about 32%.

Macro liquidity expectations have tightened over the past half-month, essentially due to a more ambiguous FOMC stance and加息预期 driven by May's CPI (influenced by high oil prices and the World Cup) and non-farm payrolls. CITIC SEC maintains its view that the Fed will not hike rates in the medium term. It believes sectors with carbon and silicon-based long-term pricing, currently discounted due to liquidity, have room for valuation repair.

For direct-exposure, extreme silicon-based cyclicals, discussing a bubble burst is still early. However, with elevated valuations demanding a博弈 on业绩, it is no longer a comfortable trading zone. The preference is towards strong supply-constrained, low-valuation sectors, with a recommendation for偏中下游的品种 within the industry chain.

Accordingly, CITIC SEC is most看好 sectors with longer narratives and AI exposure yet discounted by macro factors—carbon-based cyclical products (computing metals, fluorochemicals, phosphorus chemicals)—as well as low-valuation, mid-to-downstream segments (storage chain, gas turbine chain, optical modules, PCBs, cloud service providers).

Key Charts: