AI Startup Upscale Raises $200M to Take on NVIDIA's Networking Dominance
The New Challenger in AI Networking
In the high-stakes world of AI hardware, a fresh battle is brewing over the crucial networking technology that connects GPUs together. Upscale AI, a relatively new player founded just last year, has made waves by securing $200 million in Series A funding from investors including Tiger Global and Premji Invest.
Targeting NVIDIA's Crown Jewel
The startup isn't going after NVIDIA's GPUs directly - instead, it's taking aim at the networking technology that ties them together in large AI systems. Upscale's secret weapon? A custom-designed chip called "SkyHammer" that promises to outperform NVIDIA's proprietary NVSwitch technology used in their powerful NVL72 racks.
"What makes SkyHammer different is we're designing specifically for modern AI workloads," explained CEO Barun Kar. "We're not just copying existing approaches - we're rethinking how data should move between processors."
Breaking Down Barriers
Upscale faces formidable competition beyond just NVIDIA. Tech giants like Cisco and AMD are also developing alternatives through initiatives like UALink. But Upscale believes its approach offers several advantages:
- Flexible Compatibility: Supporting both UALink and ESUN protocols right out of the gate
- Open Ecosystem: Built-in support for SONiC, an open-source network operating system
- Industry Backing: Strategic partnerships with Intel, AMD and Qualcomm
The company claims these features will significantly reduce complexity for large-scale AI deployments while avoiding vendor lock-in.
The Road Ahead
With fresh funding secured, Upscale plans an aggressive product rollout timeline:
- Initial SkyHammer samples available Q2 2026
- Full production expected by Q4 2026 The timing couldn't be more critical as demand for AI infrastructure continues its explosive growth.
Key Points:
💰 Major Funding Boost: $200M investment signals strong confidence in Upscale's approach 🚀 Custom Silicon Advantage: SkyHammer chip designed specifically for AI workloads 🌐 Open Ecosystem Play: Supporting multiple protocols challenges NVIDIA's closed NVLink system
