Goldman Sachs initiates coverage on Iluvatar: positive on China's AI capital expenditure cycle, assigns Buy rating

Overview

This report is Goldman Sachs' initiation of coverage on Iluvatar (9903.HK). The report believes the company is a leading Chinese fabless general-purpose graphics processor/GPGPU chip company, serving more than 300 industry customers across verticals including internet, AI foundation models, scientific research, finance, and education. Goldman Sachs is positive on continued expansion in China's AI infrastructure investment, increasing opportunities in the domestic AI chip supply chain, and the company's ability to expand products and customers through its TG training chips, ZK inference chips, and software stack.

Core views

The core view is that Iluvatar is well positioned to benefit from China's upward AI computing investment cycle. Goldman Sachs raised its forecasts for AI capital expenditure by Chinese cloud service providers by 37%/44%/55% for 2026E/2027E/2028E, respectively, believing that higher AI application adoption will continue to drive infrastructure spending. Meanwhile, the expansion of China's GPU, server, networking, foundation model, and AI application ecosystem is prompting customers to reduce reliance on single suppliers, creating opportunities for the domestic supply chain. The company's product performance, compatibility, and adaptation capabilities are seen as competitive advantages, with GPU board shipments expected to grow rapidly and drive improvements in revenue, margins, and cash flow.

Analysis framework

The report uses a combined top-down and bottom-up analytical framework: top-down tracking of China CSP capital expenditure, AI model applications, and localization trends; and bottom-up forecasts for Iluvatar's GPU board shipments, ASP, product mix, gross margin, expense ratio, net profit, cash flow, and balance sheet. Valuation uses a discounted 2030E EV/EBITDA method, cross-checked against the relationship between semiconductor peers' EV/EBITDA and EBITDA growth, while also referencing P/S, P/E, and PEG & M ranges.

Methodology notes

• Valuation methodsDiscounted 2030E EV/EBITDA

  Estimate the 12-month target price based on long-term profitability

  Goldman Sachs applies a target EV/EBITDA multiple of 26.1x to 2030E EBITDA and discounts it back to 2027E using a cost of equity of 12.7%, deriving a 12-month target price of HK$1,000.

• valuation_cross_checkPEG & M

  Use forward P/E relative to net profit growth and operating margin to test valuation reasonableness

  The target price implies 21x 2030E P/E, or about 0.3x PEG & M, within the 0.1x to 1.9x range of Chinese and global semiconductor peers.

• factor_profileGS Factor Profile

  Compare stock characteristics across growth, financial returns, valuation multiples, and composite dimensions

  Goldman Sachs Factor Profile uses analyst forecasts to compare a stock's growth, financial returns, valuation multiples, and overall attributes on a percentile basis to provide investment context.

• corporate_eventM&A framework

  Assess the probability of the company being acquired and whether that is included in the target price

  Goldman Sachs assigns Iluvatar an M&A Rank of 3, indicating a low probability of acquisition, and does not include any M&A premium in the target price.

Asset mapping & comparison

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

• Iluvatar (9903.HK)

  The covered company in the report, directly benefiting from China's AI chip demand and localization trend.

  Strengths

  More than 300 customers across multiple industries; a relatively complete product portfolio formed by TG training chips, ZK inference chips, and software stack; increasing domestic supply chain opportunities.

  Weaknesses

  Still in a high-speed ramp-up stage, with losses still expected in 2026E; earnings delivery depends on shipment growth, improved expense ratios, and product iteration.

  Comparison

  The report says global tier-one AI chip suppliers still hold more than 50% share in China in 2025, leaving growth room for domestic suppliers; the target price implies a 2030E P/S below the trading range of some global GPU suppliers.

  Risks

  Intensifying competition, weaker-than-expected demand for AI chips in China, slower-than-expected product migration, and geopolitical restrictions on advanced foundries.

• AI capital expenditure by Chinese cloud service providers

  An upstream demand driver.

  Strengths

  Goldman Sachs raised its 2026E-2028E forecasts for China CSP AI capital expenditure, reflecting strong AI investment momentum.

  Weaknesses

  Capital expenditure could slow if AI application monetization or model usage growth falls short of expectations.

  Comparison

  The report mentions a sample including Bytedance, Tencent, Alibaba, and Baidu.

  Risks

  Macro environment, capital spending discipline of cloud vendors, and changes in the pace of AI application adoption.

• Domestic AI chip supply chain

  The industry chain and competitive set to which Iluvatar belongs.

  Strengths

  Customers are reducing reliance on single suppliers, while domestic GPU, server, networking, foundation model, and application ecosystems are expanding.

  Weaknesses

  Competition among domestic suppliers is intense, and differences in product performance, software ecosystem, and delivery capability will affect market share.

  Comparison

  According to IDC, Iluvatar ranked among the top 7 domestic suppliers by shipments in 2025.

  Risks

  Price competition, lagging technology iteration, and supply chain or foundry constraints.

Key data

• 12-month target priceHK$1,000.00Initiated with a Buy rating.
• Current priceHK$495.20Price as of the close on July 17, 2026.
• Implied upside101.9%Calculated based on the target price and current price.
• Market capitalizationHK$128.7bn / US$16.4bnListed in Key Data of the report.
• Enterprise valueHK$119.1bn / US$15.2bnListed in Key Data of the report.
• 3-month average daily trading valueHK$979.2mn / US$124.9mnListed in Key Data of the report.
• Upgrade to China CSP AI capital expenditure forecasts2026E +37%; 2027E +44%; 2028E +55%Reflects Goldman Sachs' more positive view on China's AI investment momentum.
• Net profit CAGR+119% CAGR in 2027E-2029EThe report expects net profit to grow rapidly after turning positive in 2027E.
• Revenue CAGR+108% CAGR in 2025E-2030EDriven by ramp-up in GPU board shipments and product upgrades.
• GPU board shipment CAGR+92% CAGR in 2025E-2030EExpected to exceed 1 million units by 2030E, versus about 45k units in 2025E.
• 2030E net profitRmb10bnSupported by growth in AI computing board shipments.
• Target EV/EBITDA multiple26.1x 2030E EV/EBITDAUsed for target price valuation.
• Cost of equity12.7%Composed of Beta 1.5, risk-free rate 3.0%, and market risk premium 6.5%.
• 2030E gross margin expectationabove 51%Although slightly lower due to a higher contribution from inference chips, it is still expected to remain at a high level.
• 2030E ROE expectation40%ROE improvement is driven by larger revenue scale and improved asset turnover.

Impact & implications

If Goldman Sachs' view plays out, Iluvatar's investment case would shift from being merely a domestic AI chip concept to one backed by quantifiable delivery in shipments, revenue, profit, and cash flow. Rising AI capital expenditure in China and customer supply chain diversification could expand the addressable market for domestic chipmakers; iteration of the TG/ZK series and software stack compatibility will determine whether the company can capture greater share in training, inference, and integrated solution scenarios. The key to valuation is whether high growth can continue to support forward EV/EBITDA and P/S multiples.

Risks

• AI chip demand in the Chinese market may be weaker than expected, which could pressure revenue growth and shipment forecasts.
• Market competition may be more intense than expected, potentially dragging on GPU board shipment growth and margins.
• Product migration or next-generation chip launches may be slower than expected, which could weaken competitiveness.
• Geopolitical tensions may restrict the company's access to global tier-one foundry resources.
• If China's AI infrastructure spending is lower than expected, the core growth thesis would weaken.

What to watch

• Whether China CSP AI capital expenditure continues to rise, especially trends related to major customer groups such as Bytedance, Tencent, Alibaba, and Baidu.
• Whether products such as TG Gen 4, TG Gen 5, ZK Gen 2, and ZK Gen 3 are launched on schedule and adopted by customers.
• Whether GPU board shipments can ramp from about 45k units in 2025E to more than 1 million units by 2030E.
• Whether the forecasts for positive net profit, positive ROE, and positive FCF in 2027E can be delivered as expected.
• Whether gross margin can remain above 51% by 2030E despite a higher proportion of inference chips.
• The competitive landscape among domestic AI chip suppliers, pricing pressure, and progress in customer supplier diversification.
• The impact of geopolitics and advanced foundry availability on product delivery and upgrades.