CITIC Securities: Tech Sector Q4 Positioning Window Approaching, Four Key Themes for RSI Acceleration
nashnova research
CITIC Securities sees the next tech rally most likely starting in Q4 2026, contingent on a rate-hike expectation reversal, a new narrative consensus, and capital rebalancing — and lays out four priority positioning themes around accelerating RSI progress.
How is this pullback different from 2022?
In 2022 the Nasdaq's PE and EPS fell roughly 28% and 8% respectively — a double hit from valuation compression and earnings downgrades. This time the drawdown is mainly multiple contraction; EPS and cash flow remain the pricing anchor.
This means → the market is repricing "expensive," not repricing "broken" — the basis for further systematic selling is limited.
A-share TMT margin-finance balances have dropped about 20.5% from their July peak, showing early signs of a near-term floor.
Offshore capital is beginning to respond positively to new AI narratives; Anthropic's potential IPO window and the capital reallocation it could trigger deserve attention.
Why are cloud giants still spending so aggressively?
The big four cloud providers (CSPs) are on track to generate about $70 billion in external AI-cloud revenue in 2026 — yet their capex runs to roughly $750 billion. Revenue falls far short of covering spend.
In plain terms = these companies are not sizing capex to current AI income. They treat compute infrastructure as strategic arms spending — a race for model capability and cloud market share.
Looking ahead: combined capex is projected at nearly $1.4 trillion in 2027 (+85% YoY) and nearly $1.9 trillion in 2028 (+35% YoY).
This reflects a market where compute supply remains the binding constraint on capex, and 2027 EPS visibility for the offshore compute chain is relatively high.
How far has RSI actually progressed?
RSI — recursive self-improvement, where AI iteratively enhances its own capabilities — is already showing results on formally verifiable tasks: Anthropic's share of R&D led by Claude rose from 1% in March to 26% in August.
OpenAI's agent research capacity has reached 3.1 Agents per human workday; Chinese-developed models Qwen, GLM, and MiMo have also disclosed RSI progress.
Long-horizon tasks remain a weak spot — OpenAI's internal success rate for 32-hour-plus tasks with zero human intervention is only about 20%.
This means → enterprise use cases — AI for Science, chip design, collaborative workflows, finance — are likely to deliver commercial upside first. Q4 2026 is a key observation window for the next round of Chinese model upgrades.
Why does Meta Muse matter?
Meta launched its personal AI assistant Meta Muse on September 8. Within two weeks it topped both the US App Store and Google Play free-app charts, with initial downloads and DAU surpassing ChatGPT.
In plain terms = Muse funnels traffic from Facebook, Instagram, and WhatsApp's roughly 3.6 billion DAU ecosystem and integrates deeply with Slack, Shopify, and other third-party apps — pulling the distribution bar for consumer AI assistants extremely low.
Monetization runs on a free / $20/month / $100/month tiered subscription, with plans to explore transaction commissions and hardware tie-ins via Meta AI glasses and Muse Charm.
CITIC estimates 100 million Muse users would require roughly 200 million CPU cores, 800 PB of memory, 10,000 PB of SSD storage, and about 1.6 GW of power — further boosting compute demand.
What are the four positioning themes?
HK-listed model companies: valuation froth has been largely wrung out; watch for new model releases, RSI capability gains, and annual recurring revenue (ARR) delivery.
Internet platforms: these hold consumer distribution gateways and monetization infrastructure; watch for personal-agent product progress.
Hardware upgrades: priority on photonic neural-network processors (NPO), optical circuit switching (OCS), thin-film lithium niobate, 1.6T optical interconnects, and high-end PCB.
Domestic memory and advanced-logic capacity expansion: equipment investment opportunities, with Q3 earnings as the window to verify order conversion.
Can the Q4 rally launch on schedule?
CITIC Securities sets three preconditions: a marginal reversal in rate-hike expectations, narrative consensus around a new theme, and capital-flow rebalancing — all three must be met.
In physical AI, the focus has shifted from product demos to delivery and real-world performance; reliability, cost, and ROI in live deployments will determine demand sustainability.
This means → Q3 earnings delivery is the decisive checkpoint for whether the Q4 rally can start on time. The report's core advice: "structure over position size."
市场有风险，内容仅供研究参考，不构成投资建议。