Replit — quick pilot play
Web notes recommend offering a short pilot (4–8 H100s or a DGX Station) to newly‑funded Replit teams to accelerate codegen training and iteration cycles — a time‑to‑value sales entry. (x.com)
Replit closed a $400 million Series D at a $9 billion valuation on March 11, 2026, with Georgian Partners leading the round. (techcrunch.com) The financing coincided with Replit’s launch of Agent 4, which ships parallel AI agents plus an “infinite canvas” and Replit’s claim of up to 10x faster app development. (blog.replit.com) NVIDIA’s H100 GPU includes a dedicated Transformer Engine designed for large-model training, and NVIDIA’s DGX H100 system is built around eight H100 GPUs in a single appliance. (cdn.prod.website-files.com) Market guides put H100 street pricing around ~$25,000 per card, which implies 4–8 H100s would represent roughly $100k–$200k of GPU hardware alone, while a fully configured DGX H100 appliance is commonly quoted in the ~$300k–$500k range. (docs.jarvislabs.ai) NVIDIA’s newly announced DGX Station (GB300/Blackwell) advertises about 20 petaFLOPS of AI compute and ~748 GB of coherent memory and is now being made available through OEM partners and preorders. (venturebeat.com) For teams that prefer cloud capacity over buying kit, recent pricing surveys show H100-based cloud instances varying roughly from $2.99/hour up to about $10/hour depending on provider and configuration. (docs.jarvislabs.ai) Replit’s prior disclosure noted annualized revenue expanding from $2.8 million to $150 million in under a year (Sept. 10, 2025), a growth figure the company cited when announcing earlier funding and product expansion. (prnewswire.com)
