OpenAI's own inference chip, Nvidia's ~$12.9B Hugging Face bid, and cheap open Qwen3.8-Flash
ksraj1001 · reddit · 2026-08-29
The author ties three stories into one theme — inference economics is now the main event:
- OpenAI's first custom inference chip 'Jalapeño' (with Broadcom, Celestica; Samsung on HBM4): claims 1.5–1.9× throughput per kW and 1.7–3.6× lower end-to-end latency vs Nvidia GB200/GB300 racks, deploying end of 2026 — vendor-reported benchmarks, so grain of salt.
- Nvidia closing in on acquiring Hugging Face for $12.9B: HF is the de facto neutral hub for open models; Nvidia owning it raises neutrality and hardware-default questions.
- Alibaba's Qwen3.8-Flash: 125B open weights, reportedly competitive with Opus 4.6 and DeepSeek V4-Flash at aggressive pricing; Qwen passed 3B downloads, ahead of Meta and Google, with revenue-sharing tests for large commercial users.
Context: this month's biggest rounds were inference infra (Fireworks AI $1.5B, Together AI $800M), while Claude had a rough uptime month. Practical advice: stop treating your model provider as fixed — benchmark cheap open models on your real workload and build fallbacks — while watching the concentration risk as chips, the open-source hub and frontier models consolidate.
More from Venture
- Bears getting euphoric as big short seller John Hempton says he's enjoying the day — toptickcrypto · 2026-10-16
- AI-driven job losses could turn $10T consumer debt into a $1T banking hole — churchkey · 2026-10-16
- I/O Fund: the 17-year bull market has room to run, but a ~20% correction comes first — Beth_Kindig · 2026-10-16
- TSMC Q3 revenue hits $46.7B, up 41% YoY; Q4 guidance of $52-54B tops estimates — Beth_Kindig · 2026-10-16
- BytePlus-backed AI Film Award offers $1M grand prize for films 70% made with Seedance — DavidmComfort · 2026-10-16
- dunkhippo33 on startup timing: most great ideas only exist because they weren't possible before — dunkhippo33 · 2026-10-16