Novita AI cuts self-serve GPU prices ~48% and trims its instance lineup to RTX-class
Novita on-demand RTX 4090 GPU instances drop to $0.35/hr ($0.18 spot) from $0.67/$0.34; L40S and H100 instances removed, H100 moves to dedicated/bare-metal only.
RTX 4090 24GB on-demand $0.67/hr ($0.34 spot); instance lineup included L40S 48GB ($0.55/hr) and H100 SXM 80GB ($2.59/hr).
RTX 4090 24GB on-demand $0.35/hr ($0.18 spot); lineup narrowed to RTX-class (RTX 4090, RTX 4090 HF $0.69, RTX 5090 HF $0.72, new RTX 6000 Ada 48GB $0.77); L40S and H100 instances removed.
On 2026-06-24 Novita AI repriced its self-serve GPU-instance page (/en/gpus). The on-demand RTX 4090 24GB rate fell roughly 48% to $0.35/hr (spot $0.18, down from $0.34), and the instance catalog was narrowed to RTX-class GPUs only - RTX 4090, RTX 4090 (High frequency) at $0.69, RTX 5090 (High frequency) at $0.72, and a newly added RTX 6000 Ada 48GB at $0.77. The previously listed L40S 48GB ($0.55/hr) and H100 SXM 80GB ($2.59/hr) instances are gone; H100 capacity is now sold only via dedicated endpoints ($1.99/GPU-hr) and 8-GPU bare-metal nodes ($1.70/GPU/hr).
Two adjacent moves landed in the same capture: dedicated endpoints added an RTX-5090 SKU at $0.73/GPU-hr, and the Agent Sandbox began publishing its underlying per-unit rate card ($0.0000098/vCPU-second, $0.0000032/GiB-second, $0.00009/GB-hour with the first 60 GB included) alongside a new $100 / 90-day free-credit offer.
Self-serve GPU instances (/en/gpus) cut and re-lined-up to RTX-class only: RTX 4090 24GB drops to $0.35/hr on-demand ($0.18 spot) from $0.67/$0.34, with RTX 4090 HF $0.69, RTX 5090 HF $0.72, and a new RTX 6000 Ada 48GB at $0.77 — L40S and H100 SXM instances removed (H100 now sold via dedicated endpoints $1.99/GPU-hr and bare-metal $1.70/GPU/hr). Dedicated endpoints add RTX-5090 at $0.73/GPU-hr. Agent Sandbox now publishes per-unit rates ($0.0000098/vCPU-second, $0.0000032/GiB-second, $0.00009/GB-hour with first 60 GB free) and a $100 / 90-day free-credit offer.