Tekedia Capital says OpenAI’s Jalapeño inference chip outperforms Nvidia GB200, GB300 on power efficiency, latency
AVGO•Tekedia Capital analysis on OpenAI’s Jalapeño inference chip
- Tekedia Capital analysis flagged OpenAI’s push into custom AI inference chips as a potential catalyst for a more fragmented semiconductor market.
- OpenAI’s Jalapeño chip, developed with Broadcom, was positioned as optimized for large language model serving rather than training.
- Benchmark comparisons versus NVIDIA GB200 and GB300 highlighted 1.5-1.9x higher throughput per kilowatt, with lower response latency.
- Deployment timeline pointed to initial rollout later in 2026, with scaled use targeted during 2027.
- NVIDIA’s position was framed as durable due to its software ecosystem, with custom silicon expected to reduce inference costs.
