Biren's second-/third-generation GPU development advances, while domestic AI computing demand remains above supply

Overview

This is a Goldman Sachs conference-takeaways-style company research report on Biren, based primarily on the Asia Communacopia + Technology conference held in Hong Kong on 18-19 May 2026. The report focuses on Biren's second- and third-generation GPU development plans, R&D investment direction, and demand for AI computing and domestic GPUs in China.

Core views

The report's core view is constructive: Biren's first-generation GPU has already been shipped to customers; the second-generation GPU will be sampled in 2026 and is expected to launch in 2H26; and the third-generation GPU remains in the R&D stage. Management believes AI computing demand in China exceeds supply, and CSP clients are actively testing new domestic GPU/ASIC vendors. The second-generation product will enhance computing power, interconnect, and CUDA compatibility, with greater emphasis on inference capabilities to adapt to the market's expansion from training toward inference.

Analysis framework

The report is mainly based on key points from management discussions at the conference and combines these with Goldman Sachs' target price framework to explain valuation: using a target 2030E EV/EBITDA multiple as the basis for forward valuation, then discounting it back to 2027E at the cost of equity. In addition to company fundamentals, the report also discloses general research framework explanations such as Goldman Sachs' Factor Profile, M&A rank, and the Quantum database.

Methodology notes

• Valuation methodsDiscounted target EV/EBITDA multiple method

  Estimate target value using 41.0x 2030E EV/EBITDA, then discount it back to 2027E at a 12.7% cost of equity.

  This method combines forward earnings capability with the relationship between comparable companies' forward EV/EBITDA and EBITDA YoY growth to derive the 12-month target price.

• Factor ProfileGS Factor Profile

  Compares the stock with the market and industry peers across four dimensions: growth, financial returns, valuation multiples, and composite factors.

  Growth is based on forward sales, EBITDA, and EPS growth; financial returns are based on ROE, ROCE, and CROCI; valuation multiples are based on metrics such as P/E, P/B, P/D, EV/EBITDA, and EV/FCF; the composite factor is the average of growth, financial returns, and the inverse of valuation.

• M&A scenarioM&A rank

  Goldman Sachs uses scores from 1 to 3 to assess a company's probability of being acquired, where 1 represents high probability, 2 represents medium probability, and 3 represents low probability.

  For companies rated 1 or 2, Goldman Sachs may incorporate an M&A component into the target price; a rating of 3 is usually viewed as immaterial and not included in the target price.

Asset mapping & comparison

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

• Biren (6082.HK)

  The company's stock covered by the report, positioned within the domestic AI GPU supply chain and the theme of Chinese CSP computing demand.

  Strengths

  The first-generation GPU has already been shipped; the second- and third-generation product lines continue to advance; the second-generation product targets upgrades in computing power, interconnect, CUDA compatibility, and inference capability; domestic CSPs are actively willing to test domestic GPUs.

  Weaknesses

  The second-generation product has not yet been fully commercialized, and the third generation remains under development; revenue expansion depends on customer validation, mass production, and ecosystem compatibility.

  Comparison

  Management believes ASICs and GPUs will coexist, with ASICs suitable for some CSP proprietary algorithms and general-purpose GPUs better suited to the broad user demand of CSP cloud businesses.

  Risks

  AI chip demand in China may come in below expectations, market competition may be stronger than expected, and wafer supply constraints may affect GPU board shipments.

Key data

• Report date2026-05-18The cover shows Equity Research 18May2026 3:15PM HKT.
• Conference location and timeHong Kong, 18-19 May 2026The report is based on discussions at the Asia Communacopia + Technology conference.
• Second-generation GPU timelineSampled in 2026; expected launch in 2H26Management said the second-generation product will upgrade computing power, interconnect capability, and CUDA compatibility.
• Third-generation GPU statusUnder developmentThe third-generation GPU remains in the R&D stage.
• Target priceHK$61.712-month target price, based on 41.0x 2030E EV/EBITDA and discounted back to 2027E.
• Closing priceHK$58.55Closing price on 07-May-26 in the historical target price table.
• Valuation multiple41.0x 2030E EV/EBITDAThe target EV/EBITDA multiple is derived from the relationship between peers' forward EV/EBITDA and EBITDA YoY growth.
• Cost of equity12.7%Used to discount the valuation back to 2027E.

Impact & implications

If Biren launches its second-generation GPU as planned and improves inference capability, CUDA compatibility, and interconnect capability, the company may expand its customer base and increase revenue opportunities. China's shortage of AI computing supply relative to demand and rising willingness to test domestic GPUs support medium-term demand for domestic GPU vendors; however, competition, wafer supply, and fluctuations in AI chip demand remain limiting factors.

Risks

• AI chip demand in the Chinese market may come in below expectations.
• Market competition may be stronger than expected.
• Wafer supply constraints may affect GPU board shipments.
• The second-generation GPU's sampling, launch, and customer validation progress may fall short of expectations.
• Progress in domestic GPU ecosystem compatibility and CUDA adaptation may affect customer adoption.

What to watch

• Progress of second-generation GPU sampling in 2026 and the pace of its 2H26 launch.
• Actual improvements in the second-generation product's computing power, interconnect capability, CUDA compatibility, and inference performance.
• R&D progress of the third-generation GPU.
• Test results, order conversion, and deployment scale of domestic GPU/ASIC products among Chinese CSP clients.
• Whether the gap between AI token demand and available computing supply continues.
• Wafer supply and GPU board shipment constraints.