Product Development Briefing: AI supply crunch, naming shifts, humanoids, and compliance signals
Signals this week point to three priorities for product leaders: secure AI supply chains, design for enterprise-ready automation, and align early with fast-moving compliance frameworks. Below are the key moves and what to do next.
AI devices: Quark glasses hit supply limits
Quark's AI glasses demand blew past forecasts. Core suppliers added a new production line, with capacity expected to open up starting next week. Internally, Quark is pushing to fully unlock output by January to meet the Spring Festival spike.
- Action for hardware teams: Lock alternate EMS/ODM capacity, add second-source components, and pre-book January logistics. If you ship wearables, run a 12-week rolling S&OP with weekly capacity reviews.
- DFX checks: Validate assembly takt times on the new line, tighten outgoing QC on optics, batteries, and thermal reliability as volume scales.
Enterprise AI: DingTalk rolls out auto-replies and smarter recording
DingTalk 8.1.10 adds AI Q&A for audio, an upgraded recording card, and a Smart Reply that can auto-respond in chats. This is a clear nudge toward AI-in-the-loop workflows across orgs.
- Action: Set policy guardrails before rollout: approved intents, data retention windows, and human-in-the-loop thresholds. Publish "reply on your behalf" rules for legal, HR, and customer-facing teams.
- Build: Add prompt templates, message classification, and citation trails. Instrument metrics: response accuracy, time saved, and escalation rate.
Model branding: Tencent Hunyuan becomes "Tencent HY"
The English brand simplifies to Tencent HY, starting with HY 2.0 Think and HY 2.0 Instruct. Clarity and pronunciation beat cleverness when you're selling globally.
- Action: If your model or platform name confuses sales or partners, re-evaluate now. Short, searchable, pronounceable. Keep upgrade paths explicit (capabilities by tier/version).
Pharma tech: "AI Kongming" debuts for end-to-end drug design
GHDDI launched "AI Kongming," an AI platform covering the full drug design workflow. For life-science product teams, the takeaway is workflow orchestration and verifiable outputs across stages.
- Action: Map your pipeline for auditability: dataset lineage, model versioning, and protocol-linked outputs. Prepare for AI-assisted review requirements.
Policy tailwinds: Central Economic Work Conference
Policy direction emphasizes boosting domestic demand, linking scientific and industrial innovation, deepening reforms and opening-up, and improving people's well-being while managing key risks.
- Action: Align 2025 roadmaps to "new-quality productive forces" and domestic-demand expansion. Prioritize products that convert directly to enterprise cost down or revenue up.
Regulation: NMPA to explore AI-assisted review
China's NMPA will push "AI + drug regulation," optimize registration processes, and explore AI-assisted review for drugs and devices. Expect higher scrutiny on data quality and model explainability.
- Action: For regulated products, implement model cards, dataset documentation, clinical or performance validation plans, and post-market monitoring hooks.
Humanoids and components: early-stage, high-upside
Gaoce says it's advancing reducers for humanoid robots with North American customers. Progress is early and risk remains high, but the stack is maturing: reducers, tendon ropes, and precision screw grinding.
- Action: If you pilot humanoids in tour-guide, retail, or education scenarios, define success in terms of uptime, path accuracy, and interaction completion rates. Lock spare parts SLAs now.
Scenario launch: fully autonomous humanoid guided tours
Beijing Humanoid Robot Innovation Center announced a fully autonomous tour-guide solution built on a general embodied intelligence platform with multi-robot scheduling and IoT linkage. Think repeatable service routes and measurable conversion.
- Action: Start with narrow scripts (exhibitions, shopping guides). Instrument dwell time, Q&A success, and staff deflection. Keep a manual override.
Large UAV platform: "Jiutian" maiden flight
The "Jiutian" drone completed its first flight. Highlights: 16-ton MTOW, 6,000 kg payload, 12-hour endurance, modular payload design. It's a signal for modularity-first architectures.
- Action: For platform teams, decouple core flight/control systems from mission payloads. Standardize interfaces and certification paths early.
Content ecosystems: Kuaishou funds animated dramas
Kuaishou is offering cash incentives up to 20,000 yuan for agencies and 10,000 yuan for individuals, plus traffic rebates for high-quality premieres. Revenue share is ramping, with most revenue returned to partners.
- Action: If you build creator tools, integrate publishing to Kuaishou and automate eligibility checks for incentives. Surface ROI dashboards per project.
Retail signals: Yonghui warns on speculation
After three limit-ups, Yonghui flagged overheated sentiment and higher speculation risk. Operations remain normal. For retail-tech vendors, budget volatility is a real constraint this quarter.
- Action: Time proposals around confirmed CAPEX cycles. Offer SaaS with outcome-based pricing to de-risk adoption.
Bioprocess capacity: CoQ10 scale-up
KingDeeWay says its CoQ10 renovation and expansion is progressing, targeting 920 tons/year at full run, using fermentation plus integrated extraction for efficiency.
- Action: If your product relies on bio-based inputs, connect demand planning to upstream batch yields and cycle times. Validate cost-down assumptions against energy and feedstock prices.
Trust and UX: Bilibili denies "site-wide membership" rumor
Bilibili stated the circulating "all videos require membership" claim is false and will pursue rumor sources. Crisis playbook 101: fast response and clear messaging.
- Action: Ship an in-product "what changed" feed and a rumor-kill micro-site pattern you can reuse within an hour.
Funding to watch
- Youlian Intelligence: $5M angel led by SenseTime Guoxiang Capital, focusing on embodied intelligence brain control, motion control, and 800V AI data center power systems.
- Jizhao Energy Storage: Angel+ in the tens of millions of yuan for liquid metal battery systems aimed at MW-class deployments.
- Action: Robotics teams: assess Youlian's control stack for pilot integrations. Energy-heavy AI workloads: model TCO with liquid metal storage for peak shaving and uptime.
Big-picture view: Zhu Xiaohu on AI leadership
Zhu Xiaohu expects China to lead the US in AI within ten years, citing US data center power constraints as a ceiling. Whether you agree or not, the operational point stands: compute and power availability will decide roadmaps.
- Action: Build a compute strategy now: multi-cloud plus domestic providers, GPU/ASIC diversity, and energy contracts tied to growth. Track PUE and inference cost per user task weekly.
This week's checklist
- Run a 90-day capacity and logistics review if you ship AI hardware or peripherals.
- Publish AI auto-reply usage policies and roll out a pilot in one internal function.
- Create model documentation and audit trails if you touch regulated workflows.
- Pick one humanoid or service-robot scenario and define hard metrics before a pilot.
- Model compute and energy cost per feature; set thresholds for feature gating.
Level up your team
If you're aligning 2025 product plans to AI-heavy roadmaps, upskilling your PMs, designers, and engineers pays off fast. Here's a curated path by role:
Your membership also unlocks:
