Sell-side research notes from Goldman Sachs Global Investment Research shared into the hub (a firm source, like WSJ — the analyst team is named on each dated page). Ratings/price targets are the bank's, archived here for the thesis and the argument, not the label.
GS Asia's Buy (TP HK$860) on China AI models: MiniMax stands out on cost efficiency, M3 sits in the favored ARR quadrant (cheap pricing × huge token volumes), 60–70% overseas revenue, ~13x P/2026E ARR vs peers several times higher.
Buy (TP HK$68): Kuaishou's Kling video model (US$18bn implied valuation) rides the multi-modal segment where — unlike text — pricing and gross margins stay healthy and demand outpaces compute.
Buy (TP HK$40, SOTP): covered as a mega-cap China AI player in GS's Competitive Positioning framework, profiled mainly on net-cash financial strength alongside Alibaba/Tencent.
Buy (SOTP TP US$186): China's closest US-hyperscaler analog — Qwen model family (open-source pioneer, Qwen-Max now closed for monetization at ~$1/1M tokens), Bailian MaaS distribution, Qoder harness, and mega-cap balance sheet for the compute race.
Buy (GS Americas Software); 2Q26 preview — PT cut to $145 (from $163) on lower comps (27x from 30x), EPS trimmed, but setup into the 7/22 print favorable. The re-rating driver is proving relevance in the enterprise-AI stack, evidenced by stabilizing organic revenue + upward WholeCo revisions.
Buy (SOTP TP HK$700): mega-cap AI player with the financial strength to outlast the low-end model price war; Workbuddy agentic/coding harness is a 2H26 signpost for capturing real-life coding data.
Initiated Neutral at US$110bn: Zhipu (Knowledge Atlas) is China's top enterprise/coding AI model — GLM5.2 is the 'intelligence moment,' strongest-positioned foundation model with DeepSeek; open-source hosting means its $1bn ARR target understates real deployment multiple-fold.
Case study, not a rating call here: Meituan's LongCat 2.0 is China's first 1.6T-param open-source MoE trained entirely on a 50,000-card domestic (Huawei) cluster — proof Chinese silicon can handle frontier-class training, a self-sufficiency milestone.
Referenced: AWS Bedrock hosts Chinese AI models fully managed in the US cloud; open-weight revenue-share on hyperscaler platforms is flagged as a future high-margin monetization path for Chinese model companies.
GS's multi-modal leader: closed-source Seed + Doubao give it ~70% of China's domestic AI token share alone; Seedance video runs ~70% gross margins at a $2bn+ ARR run-rate (per LatePost/36Kr).
Named by GS as a reference customer establishing a use case for ServiceNow's ITSM Level-1 automation — evidence of NOW's improving AI product-market fit, not a GS stance on CVS.
Strongest-positioned China foundation model with Zhipu (GS framework): 1.6T-param V4 Pro at a fraction of US SOTA size/cost, DSpark decoding +57–85% serving speed, V4 launch mid-July with 2x peak-hour pricing — demand now exceeds compute.
Cited by GS alongside NVIDIA and CVS Health as an established customer use case for ServiceNow's ITSM Level-1 automation offering — a NOW proof point, not a GS view on FDX.
Referenced: Gemini Enterprise Agent Platform (Model Garden) already hosts DeepSeek/MiniMax/GLM/Qwen fully managed in the US cloud — a key distribution channel in GS's China-models-go-global thesis.
Referenced: CEO's WSJ remarks that Microsoft may host DeepSeek on Copilot as a cheap optional model run inside Azure — legitimizing Chinese open-source models inside US enterprise software.
Named by GS as a reference customer for ServiceNow's ITSM Level-1 automation use case (with the <100-day ROI guarantee) — a NOW AI proof point, not a GS stance on NVDA here.
Private identity-governance asset acquired by ServiceNow (with Pyramid, <50bps combined); GS frames it as capturing identity-governance opportunity via the AI control tower.
In one line: a firm source — Goldman Sachs Global Investment Research notes shared into the hub; each dated page carries the analyst team's own thesis, rating and price target, so there is no single running house view.
Americas Software (Borges / Hager): constructive on ServiceNow (NOW) — Buy, though the 12-month target was cut to $145 on lower software comps. The re-rating debate is whether NOW proves its relevance in the enterprise-AI stack; stabilizing organic revenue + upward whole-company revisions are the proof points to watch.
Asia tech (Keung / Sheridan + team): China's open-source AI models have hit a "critical point of intelligence" — good enough for agentic/coding work at 10–25% of US SOTA pricing — with adoption feeding a data flywheel. Framework winners: Zhipu (2513.HK, initiated Neutral at $110bn) and DeepSeek in foundation models, ByteDance in multi-modal; MiniMax (0100.HK) is the Buy-rated expression at ~13x P/ARR. Swing risks: Western market access, frontier-model export limits, high-end-compute access.
Notes
One dated page per research note — each has its key points and a link to the saved note (PDF). Newest first.