Full Investment Committee Memo: Spatial Walk (北京空间步态科技)

Date: 2026-03-22 Project Code: SPW-2026-A Status: NOT RECOMMENDED (High Valuation, Low Moat, Significant Financial Risk)

---

I. Executive Summary

Spatial Walk is a Beijing-based startup building an interactive platform for 3D digital humans based on 3D Gaussian Splatting (3DGS). Founded by top-tier researchers behind AD-NeRF and SadTalker, the company has raised significantly at a $120M valuation. However, the core rendering technology is being commoditized by Meta and Qualcomm SDKs. With a 100x P/S ratio and only 5-7 months of runway, the company faces an imminent liquidity crisis unless a pivot to high-margin interactive services is achieved immediately.

II. Company Overview

• Founded: Late 2023 / Early 2024
• Headquarters: Beijing, China
• Focus: Real-time interactive 3DGS digital humans and mobile-end spatial computing engines.
• Key Products: 3DGS "Interactive AI Agent" platform (Beta), mobile optimization SDKs.

III. Team

• Founders: Yudong Guo (USTC Professor) and Xiaodong Cun (GBU). Known for AD-NeRF and SadTalker [Tier S research influence].
• Analysis: High technical alpha, but severe "academic risk." Founders' dual roles create IP entanglement and commitment concerns.
• Confidence Level: 🟡 Medium (Strong CVs, questionable FTE status).

IV. Market Opportunity

• Sector: Spatial Computing & AI Digital Humans.
• Problem: Existing 3DGS models are static or computationally expensive for low-end mobile devices.
• Opportunity: Low-latency interaction (lipsync/animation) on consumer-grade hardware.
• Threat: Rapid commoditization. As spatial OS (Apple AVP, Meta Horizon) integrates 3DGS natively, the market for standalone engines shrinks.
• Confidence Level: 🔴 Low (Market is being eaten by platform owners).

V. Product & Technology

• Technology: 3D Gaussian Splatting (3DGS) with a focus on AD-NeRF-inspired facial animation.
• Claims: Real-time performance on low-end chips (RK3576).
• Moat: Diminishing. The "rendering engine" is no longer a sustainable moat; only the "interaction/animation" layer remains a potential differentiator.
• Tier Assessment: [Tier B] - Claims are documented but unverified against latest 2025 platform SDKs.

VI. Competitive Landscape

• Big Tech: Meta (Presence Platform), Qualcomm (Snapdragon Spaces), Apple (RealityKit). All offer free or low-cost system-level 3DGS support.
• Startups: Luma AI, Postshot, and HeyGen (2D-gen competitor).
• Positioning: Spatial Walk is trapped between high-quality 2D-video competitors and "free" 3D system-level tools.

VII. Traction & Financial Overview

• Revenue: ~$1.2M (Fiscal Year 2025). High likelihood of being project-based/custom services rather than recurring SaaS.
• Valuation: $120M (100x P/S). This is a "Pricing for Perfection" scenario in a "Down-Market" climate.
• Burn Rate: ~$700k/month.
• Runway: 5-7 months (Critical liquidity gap).
• Confidence Level: 🟢 High (Financial data suggests a cliff).

VIII. Investment Merits (Bull Case)

1. IP Acquisition: If the company can be acquired for its interactive animation IP (AD-NeRF delta) by a hardware player.
2. Mobile Optimization: Superior performance on low-end ARM chips could win specific B2B IoT/Display contracts.

IX. Risk Factors & Mitigants

• Risk: Commoditization by SDKs. (Mitigant: Pivot to interaction/agentic layer).
• Risk: Funding Cliff. (Mitigant: Aggressive cost-cutting and bridge round).
• Risk: Academic IP. (Mitigant: Formal audit and university licensing agreements).

X. Valuation & Returns Analysis

At a 100x P/S multiple, there is no margin for error. A standard VC exit of 10x return would require $120M in revenue (assuming a 10x P/S exit). Current growth trajectory does not support this without a radical shift in business model.

XI. Recommendation & Next Steps

Verdict: PASS.

Rationale:

1. Financial Insustainability: $700k/mo burn on $1.2M annual revenue with <7 months runway is lethal in the current market.
2. Moat Erosion: The company’s original core value (rendering) is now a feature of the OS.
3. Valuation: The $120M entry price is disconnected from current market realities for "SDK Wrapper" businesses.

Next Steps (If reconsidering):

1. Audit the IP ownership of AD-NeRF/SadTalker improvements.
2. Verify FTE status of founders.
3. Perform a rigorous benchmark of Spatial Walk vs. Meta Presence Platform (v72+) rendering quality/speed.

---

Sources:

• [S] AD-NeRF/SadTalker Academic Papers (arXiv)
• [A] Internal analysis of 2024/2025 Qualcomm/Meta SDK releases
• [B] Spatial Walk website/financial data summaries
• [S] CAC (Cyberspace Administration of China) AI regulations (GB 45438-2025)