SenseTime: Native multimodal AI model and proprietary compute create differentiation

Overview

This report is Goldman Sachs' meeting note on SenseTime released after the Asia Communacopia + Technology conference in Hong Kong. The core discussion centers on the company's growth strategy, generative AI products, technology capabilities, and progress in cost reduction and efficiency improvement. The report believes SenseTime's native multimodal AI model, proprietary compute resources, and '1+X' strategy focused on core businesses are important sources of differentiation versus peers.

Core views

Goldman Sachs' main views are: First, SenseTime is restructuring its business into a '1+X' structure, making generative AI and computer vision core businesses to improve resource concentration; Second, the company is boosting efficiency through lower operating expenses and workforce integration; Third, SenseTime's model natively supports deep fusion of vision and text and can deliver strong results at relatively small parameter scales; Fourth, its self-owned data centers and roughly 40,000 PFLOPS of compute both support internal R&D and can generate revenue through third-party leasing.

Analysis framework

The report combines meeting notes with fundamental research, organizing management's comments on strategy, products, compute, customer structure, and expense control, and uses a two-stage DCF valuation framework to derive a 12-month target price. Key analytical dimensions include generative AI customer expansion, model differentiation, domestic GPU adaptation, proprietary compute, cost efficiency, and the competitive environment.

Methodology notes

• Valuation methodsTwo-stage DCF

  Derive target price using a two-stage discounted cash flow model

  Goldman Sachs derives SenseTime's 12-month target price of HK$3.55 using a two-stage DCF, assuming a WACC of 10.9% and a terminal growth rate of 2%.

• Factor AnalysisGS Factor Profile

  Compare stock attributes across growth, financial returns, valuation multiples, and composite factors

  Goldman Sachs' factor framework compares covered stocks with industry peers through normalized rankings, and the appendix explains how growth, financial returns, valuation multiples, and composite percentiles are calculated.

• M&A FrameworkM&A Rank

  Assess the probability that a company becomes an acquisition target and decide whether to include it in the target price

  Goldman Sachs classifies covered companies into levels 1 to 3 by M&A probability, where level 1 indicates a relatively high probability and level 3 is usually excluded from the target price.

Asset mapping & comparison

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

• SenseTime 0020.HK

  Coverage target

  Strengths

  Native multimodal model, deep integration of vision and text, 40,000 PFLOPS of proprietary compute, domestic GPU adaptation, focus on core AI businesses, and expense reductions.

  Weaknesses

  The scale-up of generative AI customers still needs validation, early-stage diversified businesses have uneven maturity, and commercialization timing remains uncertain.

  Comparison

  The report emphasizes that SenseTime's model differs from mainstream approaches that separate language and vision models, and says it can achieve solid performance with fewer parameters.

  Risks

  Slower-than-expected generative AI customer ramp, lower-than-expected customer spending, and stronger-than-expected market competition.

Key data

• 12-month target priceHK$3.55Derived from a two-stage DCF.
• Reported current priceHK$1.84Current price disclosed in the chart.
• Implied upside92.9%Calculated from the target price relative to the current price.
• WACC10.9%DCF valuation assumption.
• Terminal growth rate2%DCF valuation assumption.
• Proprietary compute40,000 PFLOPSData center compute resources disclosed by management.
• Model parameter scale8B/13BThe company says it can deliver solid model performance at this parameter scale.

Impact & implications

If SenseTime can continue expanding generative AI customers, improve training and inference efficiency on domestic GPUs, and convert its proprietary compute into internal R&D advantages and external leasing revenue, growth in its AI business and margin recovery may be supported. The Buy rating and relatively high implied upside suggest Goldman Sachs believes the market has not fully reflected the potential of its native multimodal model and expense improvement.

Risks

• The ramp-up of generative AI customers is slower than expected.
• Customer spending is lower than expected.
• Market competition is more intense than expected.
• There is uncertainty around AI model commercialization and the conversion of compute leasing revenue.

What to watch

• The rollout pace for generative AI enterprise customers, especially clients in sensitive-data industries such as banking and insurance.
• Whether the native multimodal model's differentiation in vision-text reasoning and generation tasks can translate into orders and revenue.
• Domestic GPU adaptation capability and the actual performance of 8B/13B parameter-scale models.
• Utilization of the 40,000 PFLOPS proprietary compute base, its contribution to internal R&D, and third-party leasing revenue.
• Margin improvement after operating expense reduction, workforce integration, and focus on core businesses.