Every time AI answers a question, someone pays to run it. NVIDIA’s Rubin platform, built around six chips, goes after that bill. AI use is counted and priced in small chunks of text called tokens. The company claims Rubin cuts the cost per token up to tenfold compared with Blackwell, its previous generation.
Most leaders think of AI as a training problem. Training is the expensive build phase, when a model learns from data. Answering real requests from staff and customers comes after, and that daily use is where Rubin cuts costs. Once each answer is cheap and fast, the question for a company is where to run AI.
If the cost of each answer drops far enough, teams stop rationing AI and design their work around constant help. Stronger models can run more often and across more steps of a process, without a cost fight over every feature. AI also gets cheap enough to include in standard plans and default features, instead of only in premium tiers.
Investors misread this in early 2025. When the AI company DeepSeek launched its cheaper models, they treated the launch as a direct threat to demand for NVIDIA’s chips. Reuters reported a record one-day drop in NVIDIA’s market value during the selloff. The reaction assumed the chips were bought only for training. It ignored how often models get used once they are good enough and cheap enough.
NVIDIA’s own announcement presents Rubin as a way to lower costs and speed up mainstream use, because that is where the volume lies. The volume is the stream of support replies, email drafts and meeting summaries a company produces every day. Quick responses also let AI sit inside everyday apps like email, where it feels instant.
At CES in Las Vegas in January 2026, NVIDIA chief executive Jensen Huang said Rubin was in full production. Reuters reported it was due to reach customers in late 2026. Many AI projects stall when each use costs too much or responses come too slowly, so adoption stays narrow. If that limit eases, the bottleneck moves to workflow design, data access and governance. The teams best placed are those that have already done the unglamorous work of ownership, policy and measurement.
