[Compressed] Institutional knowledge: DeepSeek

TRAJECTORY: Our understanding of DeepSeek evolved from a quiet research lab to a rapidly commercializing entity with surging media attention, marked by a $7.4B Series A, abandonment of its no-funding stance, and a lifecycle shift to 'established' phase, while its custom ASIC design and speculative decoding innovations position it as a key player in reducing GPU dependency and inference latency. KEY FACTS: - DeepSeek is in 'established' lifecycle phase (1 mentions/3d, 8/14d, 66 total mentions). - DeepSeek experienced a velocity spike from 1 to 3 mentions in 3 days. - DeepSeek is designing custom ASICs to reduce GPU dependency, with China's domestic chip budget share hitting 46%. - DeepSeek's DSpark unifies parallel generation with adaptive verification for speculative decoding, though no benchmarks disclosed. - DeepSeek appears in the same emerging cluster as Nvidia and Google but is unconnected to Claude Code, Anthropic, or OpenAI. - Huawei's AI infrastructure play creates a parallel ecosystem decoupled from Nvidia, potentially affecting DeepSeek's hardware options. CONFIRMED: - DeepSeek's transition from quiet research lab to commercial entity is confirmed by its $7.4B Series A and lifecycle shift to 'established'. - DeepSeek's focus on custom ASICs to reduce GPU dependency is confirmed by research observations. - DeepSeek's velocity spike in mentions is confirmed by observation data. REFUTED: - DeepSeek is NOT directly connected to Claude Code, Anthropic, or OpenAI, as per Huawei's parallel ecosystem analysis. - DeepSeek's no-funding stance was abandoned, proven wrong by the $7.4B Series A. - The hypothesis that DeepSeek's next major move will be constrained by Nvidia hardware is refuted by evidence of custom ASIC development and Huawei's parallel ecosystem, suggesting alternative pathways. OPEN QUESTIONS: - Will DeepSeek announce MCP protocol support within 60 days, driven by competitive pressure from Tencent and Alibaba? - Will at least one Chinese lab (DeepSeek or Alibaba) announce MCP support within 30 days, contradicting current 0 co-occurrence pattern? - How will DeepSeek's custom ASICs compete with Nvidia's hardware in performance and cost? - What are the benchmarks for DeepSeek's DSpark speculative decoding technology? - Will ZTE and Tencent announce a joint AI cloud venture competing with DeepSeek in the Chinese market?