Goldman Sachs initiates coverage on Iluvatar: AI capex upcycle as the key driver, with a Buy rating

Overview

This report is Goldman Sachs' initiation of coverage on Iluvatar (9903.HK). The report believes Iluvatar is a leading Chinese GPGPU fabless chip company, benefiting from rising AI capital spending by Chinese cloud service providers, expansion of the domestic AI ecosystem, and continued upgrades in training and inference chip specifications. Goldman Sachs assigns a Buy rating with a 12-month target price of HK$1,000.

Core views

The core views include three points: first, rising AI application adoption in China will continue to drive AI infrastructure investment, and Goldman Sachs has raised its forecast for China CSP AI capex; second, the expansion of domestic ecosystems in GPUs, servers, networking, foundation models, and AI applications is prompting customers to reduce reliance on a single vendor and increasing opportunities for domestic supply chains; third, Iluvatar is expanding coverage of more AI scenarios through upgrades to its TG training series, ZK inference series, and software stack, which is expected to drive rapid growth in shipments, revenue, and profit.

Analysis framework

The report combines a top-down view on China's AI capital expenditure cycle with bottom-up forecasts for the company's shipments, customer expansion, product roadmap, income statement, and cash flow; valuation uses 2030E EV/EBITDA discounted back to 2027E, with cross-checks using relative valuation methods including P/E, P/S, and PEG&M.

Methodology notes

• Valuation methods2030E EV/EBITDA discounted method

  Estimate enterprise value based on long-term profitability, then discount it to the target year.

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

• Valuation cross-checkPEG&M / P/S cross-check

  Use forward P/E, revenue growth, operating margin, and P/S multiples to test whether the target price is overly aggressive.

  The target price implies 21x 2030E P/E, 0.3x PEG&M, 32x 2027E P/S, and 5x 2030E P/S; the report believes these are within the valuation ranges of Chinese and global semiconductor peers.

• M&A scenarioGoldman Sachs M&A framework

  Assess the probability of the company becoming an acquisition target through qualitative and quantitative factors, and determine whether to include it in the target price.

  The report assigns Iluvatar an M&A rank of 3, indicating a low probability of being acquired, and therefore no M&A component is included in the target price.

• Factor profileGS Factor Profile

  Compare the stock with covered stocks and industry peers across growth, financial returns, valuation multiples, and composite factors.

  The report discloses the GS Factor Profile framework to provide relative investment context for the stock across growth, return, and valuation dimensions.

Asset mapping & comparison

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

• 9903.HK / Iluvatar

  Core covered name

  Strengths

  A leading Chinese GPGPU fabless chip company with more than 300 customers; TG training chips, ZK inference chips, and the software stack together cover training, inference, and integrated AI computing solutions; expansion of the domestic ecosystem and customers' multi-vendor strategies provide growth opportunities.

  Weaknesses

  The company is still in a phase of heavy investment and commercialization ramp-up, and is expected to remain loss-making in 2025E and 2026E; inventory days and working capital usage are relatively high, and earnings and cash flow improvement depend on continued shipment scale-up.

  Comparison

  The target price implies 5x 2030E P/S, below the trading range of global GPU suppliers mentioned in the report; NVIDIA is currently at 8.0x 2027E P/S and AMD at 10.3x 2027E P/S.

  Risks

  China AI chip demand comes in below expectations, market competition intensifies, product migration is slower than expected, or geopolitical restrictions limit access to leading global foundries.

• Global GPU suppliers such as NVIDIA / AMD

  Valuation and industry reference

  Strengths

  As global GPU suppliers, their P/S multiples provide a reference range for Iluvatar's long-term valuation.

  Weaknesses

  The report does not use them as direct operating forecast targets, only for relative valuation comparison.

  Comparison

  NVIDIA is at 8.0x 2027E P/S and AMD at 10.3x 2027E P/S; Iluvatar's target price implies 5x 2030E P/S.

  Risks

  Changes in competition and supply dynamics among global GPU suppliers may affect the pace at which Chinese domestic AI chip suppliers gain share.

Key data

• RatingBuyInitiation of coverage, with the rating effective from 2026-07-19.
• 12-month target priceHK$1,000.00Based on 2030E EV/EBITDA discounted valuation.
• Current priceHK$495.20Price disclosed on the report cover page.
• Implied upside101.9%Calculated from the target price relative to the current price.
• Market capitalizationHK$128.7bn / $16.4bnDisclosed in the report Key Data.
• Enterprise valueHK$119.1bn / $15.2bnDisclosed in the report Key Data.
• 2025E-30E GPU accelerator card shipment CAGR+92%Shipments are expected to exceed 1 million units by 2030E.
• 2025E-30E revenue CAGR+108%Driven by rising GPU accelerator card shipments, customer expansion, and the upcycle in AI capital spending.
• 2027E net profitRmb1,136.1mnGoldman Sachs expects the company to turn profitable in 2027E.
• 2030E net profitRmb10bnThe report body discloses the 2030E net profit target.
• 2030E gross marginabove 51%The report expects a slightly lower gross margin due to a higher mix of inference chips, but still remaining at a high level.
• COE12.7%Valuation discount rate, including Beta of 1.5, risk-free rate of 3.0%, and market risk premium of 6.5%.

Impact & implications

If the report's assumptions materialize, Iluvatar could become a high-beta beneficiary of China's AI compute localization cycle, with revenue, earnings, and free cash flow inflection points all concentrated after 2027E. The investment implication is that the market may be underestimating the compound growth driven by upward revisions to China AI capex, rising domestic supply chain share, and product iteration, although the high valuation also makes demand, competition, and product migration pace key variables to monitor.

Risks

• AI chip demand in China comes in below expectations, potentially reducing revenue and shipment forecasts.
• Competition among domestic AI chip suppliers is more intense than expected, potentially weighing on GPU accelerator card shipment growth and margins.
• Product migration is slower than expected, potentially weakening market competitiveness and subsequent shipment growth.
• Geopolitical tensions may limit the company's ability to use leading global foundries.
• Slower-than-expected investment in China's AI infrastructure would weaken the report's core growth assumptions.

What to watch

• Whether China CSP AI capital spending continues to be revised upward, especially investment momentum in 2026E-2028E.
• Progress in scaling GPU accelerator card shipments from 45k units in 2025E to more than 1 million units by 2030E.
• Launch and customer adoption timing for products such as TG Gen 4, TG Gen 5, ZK Gen 2, and ZK Gen 3.
• Changes in customer count and large-customer mix, especially procurement scale from local CSP customers.
• Whether gross margin remains at a high level, and whether the opex ratio declines as projected in the report from 156% to 23% by 2030E.
• Whether 2027E net profit and free cash flow turn positive on schedule.