Novita brings H100 and L40S back to self-serve GPU instances
Novita re-added NVIDIA H100 SXM 80GB ($3.39/hr on-demand, $1.70/hr spot) and L40S 48GB ($0.55/hr) to its self-serve GPU-instance lineup, removing RTX 6000 Ada 48GB and the RTX 5090 'High frequency' variant in the same pass.
Self-serve /en/gpus lineup: RTX 4090 24GB ($0.33/$0.17), RTX 5090 32GB High frequency ($0.72/$0.36), RTX 5090 32GB ($0.73/$0.37), RTX 6000 Ada 48GB ($0.77/$0.39) — RTX-class only since 2026-06-24.
Self-serve /en/gpus lineup: RTX 4090 24GB ($0.33/$0.17), NVIDIA L40S 48GB ($0.55/—), RTX 5090 32GB ($0.73/$0.37), NVIDIA H100 SXM 80GB ($3.39/$1.70) — data-center-class GPUs return to self-serve for the first time since 2026-06-24.
Novita’s /en/gpus self-serve page had been limited to RTX-class consumer GPUs since a 2026-06-24 repricing pulled L40S and H100 off the page entirely, routing data-center capacity through dedicated endpoints and bare-metal only. The 2026-08-26 capture reverses that: H100 SXM 80GB and L40S 48GB are both back as self-serve on-demand/spot instances, while RTX 6000 Ada 48GB — itself only re-added one day earlier, on 2026-08-25 — was pulled again, and the RTX 5090 “High frequency” variant was dropped, consolidating the RTX 5090 row to a single $0.73/hr listing. Notably, self-serve H100 on-demand ($3.39/hr) is priced well above both the dedicated-endpoint rate ($1.99/hr) and bare-metal ($1.70/hr), while H100 spot ($1.70/hr) exactly matches the bare-metal on-demand rate — a buyer comparing “the same GPU” across Novita’s four products now has five price points to reconcile instead of four.
The /en/gpus self-serve lineup regained two data-center-class GPUs pulled on 2026-06-24: NVIDIA H100 SXM 80GB (new — $3.39/hr on-demand, $1.70/hr spot) and NVIDIA L40S 48GB (new — $0.55/hr on-demand, no spot rate shown). In the same capture, RTX 6000 Ada 48GB — itself only re-added the day before, 2026-08-25, at $0.77/hr — was delisted again, and the RTX 5090 'High frequency' variant ($0.72/hr) was removed, leaving a single RTX 5090 32GB listing at $0.73/hr on-demand / $0.37/hr spot. Serverless model rates, dedicated endpoints, bare-metal, and Agent Sandbox rate cards were all unchanged.