SambaNova is now valued at $11 billion. The California-based startup, which focuses on specialized AI chips, has raised $1 billion in fresh funding—a strong signal that institutional investors are willing to bet on genuine alternatives to Nvidia's GPU dominance.
The essentials
- Funding round: $1 billion, led by General Atlantic; additional investors include Seligman Ventures, T. Rowe Price, Capital Group
- Valuation: $11 billion post-round
- Focus: Inference chips for on-premise deployments (data processing within the company, not in the cloud)
- Major customer: JPMorgan Chase will deploy SambaNova systems for secure enterprise-wide AI workloads
- IPO plans: CEO Rodrigo Liang is considering a public offering in 2027, likely in the U.S.
Inference, not training – a different strategy
While Nvidia dominates training massive AI models with its GPUs, SambaNova targets a different but equally lucrative field: inference—the fast, cost-efficient execution of already-trained models. This is where AI systems actually work in production, and where companies can save billions if they have efficient hardware.
The startup sells its latest chip, the SN50, as part of server units for data centers. The key difference: these systems can be deployed on-premise—directly at the company using them, not in someone else's cloud.
"Inference has broken everything open, and so what we're seeing now is that as a standalone company, you have the ability to really move fast and drive the business across a broad range of sectors," co-founder and CEO Rodrigo Liang told CNBC.
Why on-premise matters for banks and enterprises
For financial institutions like JPMorgan Chase, this is a game-changer. Sensitive data stays in-house, under their control, behind their firewall. No detour through external cloud providers, no data leakage risk.
"For banks and for other industries where data is incredibly important, bringing this infrastructure on prem, bringing this infrastructure with models that are then under your control with your private data, and having all within your firewalls is an incredibly important aspect of running AI in a very secure and private manner," Liang explained.
JPMorgan will deploy SambaNova systems for "on-prem inference in demanding enterprise AI workloads"—proof the model works even at the largest financial institutions.
Timing: A market in flux
SambaNova isn't alone. A wave of AI chip startups is trying to loosen Nvidia's grip. The semiconductor index (PHLX) has risen roughly 80 percent this year—the market generously rewards innovation in this sector.
The funding round follows an earlier investment of over $350 million earlier this year, including from Intel, with which SambaNova also announced a strategic partnership.
What this means for German enterprises
For German mid-market and large companies, this matters: if alternatives to Nvidia gain traction, inference costs drop. That makes local, data-sensitive AI deployments more economical. The pattern here is clear: specialized hardware for specific tasks (not the universal GPU model) may be how the market fragments. German companies hesitant so far because Nvidia infrastructure is expensive could soon have real alternatives—provided SambaNova and similar players deliver.
Next milestone: an IPO in 2027. By then, we'll know if the model truly scales.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
