The AI hardware battle is officially expanding from training to inference.
SambaNova Systems has raised $350 million in a new funding round to capitalize on the surging demand for AI inference chips—the hardware responsible for running AI models and making real-time decisions. The round brings together an unlikely alliance: Private Equity heavyweight Vista Equity Partners and legacy semiconductor giant Intel.
💰 THE DEAL DYNAMICS:
- The Lead Investors: Co-led by Vista Equity Partners and Cambium Capital, alongside Intel Capital.
- The Rare PE Play: This marks a highly unusual foray into hardware for Vista, a firm historically laser-focused on enterprise software. It signals that AI infrastructure is becoming too critical for traditional software investors to ignore.
- The Backup Plan: The funding round and multi-year strategic partnership follow stalled acquisition talks between SambaNova and Intel, where Intel had previously weighed a $1.6 billion buyout of the startup.
⚙️ THE TECH & TRACTION:
- The Hardware: Proceeds will fund the expansion of SambaNova’s new SN50 AI chip and scale its SambaCloud platform.
- The First Customer: SoftBank Corp will be the first major client to deploy the SN50 chip within its AI data centers in Japan.
- The Intel Synergy: Intel and SambaNova will jointly deliver cost-effective AI inference solutions, complementing Intel’s existing data center GPU lineup while giving SambaNova massive distribution leverage.
💡 ANALYST TAKEAWAY: The era of “Nvidia or Nothing” is facing its first real test in the inference market. As enterprises move from building AI models to actually deploying them, cost-efficiency and speed become paramount. Vista’s rare hardware investment proves that software margins are now fundamentally tied to compute costs. Meanwhile, Intel is hedging its bets—if it couldn’t buy SambaNova outright, a deep strategic partnership ensures it doesn’t get left behind in the enterprise inference race.
👇 Semiconductor Investors: Will specialized inference chips from startups like SambaNova successfully commoditize the deployment layer, or will Nvidia’s CUDA ecosystem maintain its iron grip?