Citi Sees Potential in AI Inference and Optical Communications, Reiterates Buy Ratings on Four Stocks Including Apple

Overview

This report, issued by Citigroup, focuses on cutting-edge developments in the AI inference layer, with a particular emphasis on model performance, power supply constraints, data center location logic, and investment value of specific beneficiary stocks. The report argues that AI scarcity is now rapidly being monetized, with top models managing demand through higher pricing and restricted access. Meanwhile, power supply has become a critical bottleneck, driving data centers to shift toward low-cost, high-renewable-energy regions. At the company level, Citi reaffirms its buy ratings for Apple (AAPL), Ciena (CIEN), Lumentum (LITE), and MiniMax (0100.HK), providing detailed logic behind their target price estimates.

Core views

Models and Inference Layer: The 'vertical wall' effect of AI models is becoming evident, with the performance gap between top proprietary models and open-source models widening to 9-10 points. The latest top models have improved their intelligence index score by 4 points to 65, but their hybrid pricing has roughly doubled and requires access via usage credits, indicating that scarcity has become a product feature. Google’s DiffusionGemma achieves four times faster inference speed than Gemma 4 by shifting the decoding bottleneck from memory bandwidth to computation, though at the cost of quality in non-sequence generation. Additionally, new models like Poolside’s Lagauna and Odyessy’s Starchild-1 are driving advances in agent coding and multimodal world models, expanding the boundaries of AI applications. Power and Infrastructure: Power supply is the core constraint on AI expansion. A private neo-cloud company has signed up for 4.9 GW of capacity, yet its pipeline exceeds 40 GW+, with a signing rate of only 12%, reflecting an explosive surge in demand coupled with supply challenges. Data center locations are highly concentrated in regions with retail electricity prices of 9-12 cents/kWh, and states with over 25% renewable energy share are attracting disproportionate capacity. As component costs rise, capital expenditure (CapEx) per H100 equivalent unit steadily increases, and power costs are shifting from post-construction operational expenses to pre-construction capital costs. Stock Valuation and Logic: 1. Apple (AAPL): Target price of $315, corresponding to a 33x forward P/E ratio based on 2027 expected earnings per share. This valuation is 20% above historical levels, justified by gross margin expansion (excluding tariff impacts), growth in service sales mix, gradual adoption of Apple Intelligence, and a strong balance sheet. 2. Ciena (CIEN): Target price of $658, corresponding to a 45x forward P/E ratio based on fiscal year 2028 expected earnings per share. This valuation is higher than peers in the networking sector (average over 30x), but lower than optical companies with similar growth expectations (average around 49x). Supporting logic includes growth in cloud and data center interconnect (DCI) opportunities, as well as an estimated two-year EPS compound growth rate of around 50%. 3. Lumentum (LITE): Target price of $1100, corresponding to a 40x P/E ratio, 14% above the average forward P/E ratio of its AI optical peers (35x). The report considers this premium reasonable given its projected EPS growth rate of 2-3 times that of peers from 2025 to 2028, along with potential from new products such as CPO and OCS. 4. MiniMax (0100.HK): Target price of HK$1,330, corresponding to a 30x forward P/S ratio based on 2028 expectations. Although seemingly high, considering its projected 2025-2030 revenue CAGR of 128%, the price-to-revenue CAGR ratio is only 0.23x, placing it within a reasonable range.

Analysis framework

The report employs a multi-dimensional analytical framework. First, it tracks model ranking scores, inference speeds, pricing, and developer usage (such as PyPI downloads and SDK adoption rates) to quantify the technological evolution and market acceptance of AI models. Second, it combines macroeconomic geographic data (such as state electricity prices and renewable energy shares) with micro-project data (such as signed capacity and pipeline ratios) to analyze physical constraints and location logic for data center infrastructure. For stock valuations, different valuation methods were applied according to each company’s industry characteristics: for the mature tech giant Apple, the report uses a P/E ratio and compares it with historical levels; for high-growth network equipment providers Ciena and Lumentum, it uses forward P/E ratios combined with PEG logic (growth-adjusted valuation) for peer comparisons; for the early-stage, high-growth AI-native company MiniMax, it uses a forward P/S ratio and references its own trading history and revenue growth rate to validate its reasonableness.

Methodology notes

• Valuation MethodPE/PEG valuation

  A valuation method combining P/E ratio with earnings growth rate

  In evaluating Ciena and Lumentum, the report not only looks at the absolute P/E ratio but also considers their projected high EPS compound growth rates (e.g., 50% or 2-3 times that of peers) to justify giving them a higher P/E premium—a typical PEG valuation approach.

• Valuation MethodPS valuation

  Price-to-Sales valuation, suitable for high-growth but unprofitable or early-stage companies

  For AI companies like MiniMax, which are in the early stages, experiencing rapid revenue growth but may not yet be stable in profitability, the report uses a 2028 expected P/S ratio for valuation and validates its relative cheapness through the ‘price-to-revenue CAGR ratio.’

• Industry/Industrial Analysis FrameworkSupply-demand framework

  Analyzing the balance between industry supply and demand

  By analyzing AI chip rental prices, model pricing changes, and the large gap between data center power signing rates (12%) and the potential pipeline (over 40 GW+), the report reveals the current state of supply shortage in the AI infrastructure sector—a key factor in judging industry prosperity and pricing power.

• Industry/Industrial Analysis FrameworkValue Chain Upstream-Midstream-Downstream Transmission

  Analyzing value distribution across various links in the value chain

  The report points out that rents/savings generated by model layering and routing architectures will flow to the compilation inference layer and the tier holding profiling data, indicating that value is shifting from simple model training toward inference optimization and routing management.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Apple (AAPL.US)

  Benefit: Gradual adoption of Apple Intelligence and growth in service mix drive valuation premium

  Strengths

  Expanded gross margins, strong balance sheet, rising proportion of service revenue

  Weaknesses

  High dependence on Taiwan and mainland China supply chains

  Comparison

  Valuation 20% above historical levels, reflecting AI empowerment expectations

  Risks

  Macroeconomic weakness leading to contraction in the smartphone market, Sino-US tensions impacting supply chains, European regulatory risks

• Ciena (CIEN.US)

  Benefit: Growth in cloud and DCI opportunities, demand for AI-driven network upgrades

  Strengths

  EPS two-year compound growth rate around 50%, technical edge in optical communications

  Weaknesses

  Top 10 customers account for over half of sales, high customer concentration

  Comparison

  Valuation above peers in the networking sector but below pure optical peers, reflecting its intermediate positioning and high growth

  Risks

  Slow recovery in telecom and cable TV orders, loss of market share, loss of key customers

• Lumentum (LITE.US)

  Benefit: Adoption of new AI optical modules, CPO, and OCS products

  Strengths

  Projected EPS growth rate 2-3 times that of peers, strong potential from new product lines

  Weaknesses

  If AI ramp slows down, EML ramp could be affected

  Comparison

  Enjoying a 14% P/E premium over peers, based on higher growth expectations

  Risks

  Intensified AI competition limiting market share growth, slower-than-expected new product adoption, weaker-than-expected margin expansion

• MiniMax (0100.HK)

  Benefit: Leading Chinese AI-native company with rapid revenue growth

  Strengths

  Projected 2025-30E revenue CAGR of 128%, agentic evolution and global abstraction capability

  Weaknesses

  Short trading history, no direct comparable listed companies

  Comparison

  Valued using its own historical P/S mean discount, with low PEG ratio

  Risks

  Rapid model iteration and intense competition, selling pressure after lock-up period ends, ongoing financing needs, currency fluctuations, talent retention, and IP risks

Key data

• Apple Target Price$315Corresponds to a 33x P/E ratio based on 2027 EPS, 20% above historical levels
• Ciena Target Price$658Corresponds to a 45x forward P/E ratio based on 2028 EPS, with EPS two-year CAGR around 50%
• Lumentum Target Price$1100Corresponds to a 40x P/E ratio, 14% above the average peer P/E of 35x
• MiniMax Target PriceHK$1,330Corresponds to a 30x forward P/S ratio based on 2028 expectations, with 2025-30E revenue CAGR projected at 128%
• Top Model Intelligence Index65 pointsUp 4 points from the previous release, hybrid pricing roughly doubled
• Data Center Electricity Price Range9-12 cents/kWhRetail electricity price range in regions where data centers are concentrated
• Neo-Cloud Signing Rate12%4.9 GW signed against a pipeline of over 40 GW+, reflecting supply bottlenecks

Impact & implications

The report believes that the segmentation and monetization of AI inference layers will bring excess returns to companies with efficient inference compilation and data routing capabilities. For hardware and infrastructure providers, power costs and availability will become key variables determining competitiveness, favoring operators and equipment makers located in regions with low electricity prices and high green energy shares. Among specific stocks, Apple, thanks to its ecosystem and service transformation, is expected to maintain valuation premiums; Ciena and Lumentum, as leaders in optical communications and network equipment, will directly benefit from increased capital spending on data center interconnects and AI cluster construction; MiniMax, as a leading Chinese AI-native company, sees its high growth expectations supporting a relatively high P/S valuation—but risks related to competition and commercialization need to be closely monitored.

Risks

• Macroeconomic weakness or shifts in consumer demand leading to smartphone market contraction (Apple)
• Sino-US tensions impacting supply chain stability (Apple)
• European Digital Markets Act and other regulations reducing app store revenues (Apple)
• Slow or delayed recovery in telecom and cable TV orders (Ciena)
• Loss of key customers significantly impacting sales (Ciena)
• Intensified AI competition limiting market share growth (Lumentum)
• Slower-than-expected adoption of AI workloads affecting new product ramps (Lumentum)
• Rapid model evolution and fierce competition (MiniMax)
• Selling pressure following the upcoming six-month lock-up period ending (MiniMax)
• Ongoing capital financing needs and currency fluctuation risks (MiniMax)

What to watch

• Progress on enterprise-level routing primitives at Databricks Data+ AI Summit (June 15-18) and AWS Summit NYC (June 17)
• Technical announcements at AMD Advancing AI (June 22-23) and other AI-related conferences
• Impact of Texas’ ERCOT Shares Batch Zero Inclusion Decisions (August 7) on the electricity market
• Adoption rate of Apple Intelligence and changes in service sales mix
• Order trends and market share changes for Ciena and Lumentum in cloud and DCI sectors
• New product adoption and revenue growth realization for MiniMax