Delivery delays weigh on near-term upside, but the long-term domestic AI accelerator thesis still supports maintaining Overweight

Overview

This report is Morgan Stanley’s 2Q26 earnings preview and valuation update for Cambricon Technology Corporation. The key change is that slower-than-expected MLU580/590 deliveries are delaying near-term revenue recognition, prompting reductions to 2026-2028 revenue, EPS and target price forecasts. However, the report maintains Overweight, believing that demand for China’s domestic AI accelerators, capex by CSPs and sovereign AI buyers, domestic substitution and system-level competitiveness will support Cambricon’s long-term growth.

Core views

The report argues that the near-term issue is supply-chain execution and insufficient system-level delivery maturity, rather than weak end-market demand for AI accelerators. Cambricon’s 2Q26 revenue is expected to be Rmb3.8bn, up 31% sequentially, with EPS of Rmb2.09; 2026/27/28 revenue forecasts are cut by 2%/5%/8%, respectively, while EPS forecasts are reduced by 6%/5%/8%. Even so, the report emphasizes that China’s AI computing market is shifting from single-chip specification competition toward system-level scaling, real token economics, TCO, software migration costs and supply availability. Cambricon remains attractive over the long term due to its domestic ecosystem position, MLU580 ramp-up and MLU690 roadmap.

Analysis framework

The report combines supply-chain checks, a 2Q26 earnings preview, revisions to 2026-2028 financial forecasts, China AI GPU TAM estimates, forecasts for AI chip localization, relative valuation and a residual income model. At the industry level, it applies an “economics × execution” framework focused on TCO, token cost, TPS, performance per dollar, foundry availability, software ecosystem maturity, CSP relationships and product roadmap credibility.

Methodology notes

• Valuation methodsResidual income model

  Deriving the base-case target price using a residual income model

  The report continues to use a residual income model as its base-case valuation method. The target price reduction primarily reflects downward revisions to 2026-2028 EPS forecasts. Key assumptions include an 8.4% cost of equity, 1.06 beta, a 2.0% risk-free rate, a 6.0% equity risk premium, a 16% medium-term growth rate and a 6.0% terminal growth rate.

• Industry analysisEconomics × execution

  Competitive framework for domestic AI accelerators

  The report believes China’s AI GPU competition should not be assessed solely by policy themes or single-chip specifications, but by deployment economics and execution capabilities, including TCO, token cost, TPS, cost-performance, foundry resources, software ecosystem, CSP relationships and roadmap credibility.

• Financial forecastsMorgan Stanley ModelWare

  Framework for financial forecasts and consensus comparisons

  Unless otherwise stated, the financial metrics in the report are based on the Morgan Stanley ModelWare framework, while some consensus data comes from Refinitiv Estimates.

Asset mapping & comparison

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

• Cambricon Technology Corporation (688256.SS)

  Covered company; Overweight rating maintained

  Strengths

  Deep engagement with major CSP customers, with MLU590 deployed in relevant workloads; progress in transitioning to the domestic AI chip supply chain; MLU690 expected to launch in 4Q26, with potential performance improvement of approximately 2.2x; benefits from demand for domestic AI accelerators and domestic substitution.

  Weaknesses

  MLU580/590 deliveries are slower than expected, while production ramp-up, memory availability and system-level delivery remain immature; absolute valuation is elevated.

  Comparison

  The report believes China’s domestic chips remain approximately two generations behind the US at the single-chip level, but the practical gap is narrowing through multi-chip designs, advanced packaging, rack-level system architecture, optical networking and hardware-software co-optimization. Compared with Nvidia products available in China, domestic accelerators have the potential to offer lower TCO and similar token costs.

  Risks

  Capacity and yield constraints, customer concentration, slower technology iteration, intensifying competition, price competition, rising wafer costs, and slower-than-expected domestic AI capex and LLM commercialization.

• China AI chip/AI GPU industry chain

  Industry backdrop for Cambricon’s long-term growth

  Strengths

  Demand from CSPs, sovereign AI buyers, SOEs and local governments for AI infrastructure supports TAM expansion; inference demand is more frequent and utilization-intensive, benefiting domestic solutions with better cost efficiency and availability.

  Weaknesses

  Demand is concentrated among a small number of large buyers, making the industry highly sensitive to order timing and capex changes; software ecosystem and system-level capabilities remain key barriers.

  Comparison

  Industry competition is shifting from peak single-chip performance toward system-level scaling, real token economics and deployment costs, so the sector should not be valued simply as a policy theme.

  Risks

  Policy implementation falling short, CSP capex volatility, changes in overseas access to advanced chips, and issues related to domestic chip yields and software migration costs.

Key data

• Target price adjustmentCut from Rmb1,528 to Rmb1,408Mainly reflects downward revisions to 2026-2028 EPS forecasts.
• RatingOverweightThe report maintains its Overweight rating, believing the long-term structural growth thesis remains intact.
• 2Q26 revenue forecastRmb3.8bnExpected to grow 31% sequentially, but 2% below the previous forecast.
• 2Q26 EPS forecastRmb2.09Quarterly EPS forecast in the earnings preview.
• 2026/27/28E EPS reduction6% / 5% / 8%Reflects slower deliveries and supply-chain uncertainty.
• 2026/27/28E revenue forecast reduction2% / 5% / 8%Mainly due to slower-than-expected MLU580/590 deliveries.
• China AI chip TAM forecastUS$91bn in 2030ERaised 36% from the previous US$67bn estimate, implying a 23% CAGR from 2025 to 2030.
• China AI chip self-sufficiency forecastRising from 42% in 2025 to 70% in 2030EDriven by expansion of advanced-node capacity and continued improvements in chip performance.
• Current valuationApproximately 98x 2026e P/E; approximately 31x 2026e P/SThe report considers absolute valuation high but supported by strong growth and strategic positioning.
• Base/bull/bear case valueRmb1,408 / Rmb2,660 / Rmb709The bull and bear cases correspond to approximately 72x and 19x 2026e P/S, respectively.

Impact & implications

For investors, the report signals that “near-term delivery risk is rising, but the long-term domestic AI computing thesis remains intact.” Lower target price and earnings forecasts may limit near-term share-price upside, but if MLU580 deliveries improve, MLU690 progresses as scheduled and CSP orders continue to ramp, Cambricon could still benefit from China’s AI infrastructure investment, rising inference demand and domestic substitution. Conversely, if capacity, yields, system-level delivery or software ecosystem progress falls short of expectations, the high valuation would amplify execution risk.

Risks

• MLU580/590 deliveries are slower than expected, and order-to-revenue conversion is below expectations.
• Production ramp-up, memory availability, system-level delivery and supply-chain maturity remain near-term constraints.
• New products from peers in 2H26 may intensify competition, testing Cambricon’s relative performance and cost-performance positioning.
• Valuation is elevated, creating significant compression risk if growth falls short of expectations.
• Customer concentration is high; if the company cannot continue to secure large orders from leading customers, revenue and profit growth may fall below expectations.
• Yield, capacity and domestic supply-chain migration progress may fall short of expectations.
• Domestic AI capex, LLM commercialization or sovereign AI investment may progress more slowly than expected.
• Price competition, rising wafer manufacturing costs and insufficient capacity utilization may pressure gross margins.

What to watch

• Actual MLU580/590 delivery pace and order-to-revenue conversion in 2H26 and 2027.
• Whether MLU690 progresses according to the 4Q26 roadmap, as well as its performance improvement and customer validation.
• Changes in AI infrastructure capex by major CSP customers, operators, SOEs and local governments.
• Whether China’s AI chip TAM and domestic self-sufficiency rate deliver in line with the report’s upward forecast path.
• Whether Cambricon’s gross margin can remain within the forecast ranges of approximately 51%, 50% and 49% in 2026-2028.
• The impact of peers’ new product launches in 2H26 on Cambricon’s performance, cost-performance and market share.
• Whether capacity, yields, memory supply and system-level delivery capabilities improve.
• The coverage and retention effects of the equity incentive plan, as well as differences between its revenue and profit targets and market expectations.