Salesforce Made AI 8x Cheaper and Outage-Proof
AI that runs your business can now cost less and survive data center failures.
Salesforce builds AI agents, called Agentforce, that many businesses rely on for customer support, sales, and internal tasks. Running these agents takes powerful and expensive computer chips called GPUs. Salesforce discovered they could cut costs by 8 times, because multiple AI models could share the same chips instead of each owning dedicated ones.
That cost saving came with a catch. When models share chips, they could all land in the same data center, or even on the same server. If anything failed there — a crashed server or a local power outage — the AI would suddenly go dark. For companies that depend on this AI, that could mean lost sales, frustrated customers, or broken workflows.
So Salesforce worked with Amazon Web Services to fix it. AWS released a new setting that lets customers control exactly where AI copies get placed. You can tell the system to spread copies evenly across data centers, so no single failure takes down the whole service. You can also choose to pack copies tightly to save money, or spread them apart for safety.
The key lesson: for critical AI, always keep at least two copies in different data centers. That small extra cost is insurance against a much bigger outage bill. Other companies can now use the same trick, making business AI both cheaper and more dependable.
- Salesforce cut AI infrastructure costs by 8 times by letting multiple models share the same GPUs.
- A new AWS feature lets AI copies be spread across separate data centers, keeping services online during outages.
- For critical AI, you need at least two copies in different locations — one copy means a single point of failure.
Why It Matters
Businesses get cheaper, more reliable AI. Fewer outages means fewer lost customers and lower operating costs.