AI networking startup Upscale scores $200M to challenge Nvidia's NVSwitch
(2026/01/22)
- Reference: 1769040476
- News link: https://www.theregister.co.uk/2026/01/22/upscale_skyhammer_nvidia/
- Source link:
AI networking startup Upscale AI on Wednesday announced it has raised $200 million in Series A funding to challenge Nvidia's dominance of switches for rack-scale AI systems, putting it in competition with the likes of Cisco and AMD.
Founded last year, with backing from the likes of Intel, AMD, and Qualcomm, Upscale is developing a new class of scale-up network switches specifically designed to compete with the NVSwitch chips found in Nvidia's NVL72 racks.
NVLink is a high-speed interconnect technology that Nvidia developed to abstract memory and compute resources from multiple GPUs so they appear as a single logical resource.
[1]
The tech [2]debuted in 2024, and ever since, the likes of AMD and Cisco have tried to create alternatives. But their efforts so far, like Ultra Accelerator Link (UALink) and Ethernet for Scale-Up AI Infrastructure (ESUN) remain immature.
[3]
[4]
The first UALink-based rack systems from AMD will hit the market later this year, but they'll be tunneling the protocol over Ethernet. Purpose-built UALink switches capable of competing with Nvidia's NVSwitch simply aren't ready yet. Upscale aims to change that with its SkyHammer custom ASICS.
"Instead of retrofitting legacy systems, what we are re-imagining is what scale truly means in AI networking," Barun Kar, Upscale AI CEO told El Reg .
[5]
"The heart of this architecture is essentially to scale up. It's meant for AI workloads and nothing else."
The startup isn't ready to reveal technical details about SkyHammer, but has said it will offer its first silicon as a standalone ASIC for hyperscale integration, and as an integrated switch blade and rack.
While we don't have enough information about the chip to draw any comparisons to NVSwitch 6 or Broadcom's Tomahawk 6, Kar tells us it is using a memory semantic-based load-store network architecture, and will feature acceleration for collective communication similar to Nvidia's Sharp.
[6]
The platform will also support both UALink and the competing ESUN protocol.
To make the whole thing manageable at scale, Upscale is working to extend support for the SONiC network operating system (NOS). Originally developed by Microsoft, the open source NOS is both widely deployed and well understood by hyperscale customers.
[7]Cisco finally fixes max-severity bug under active attack for weeks
[8]AMD threatens to go medieval on Nvidia with Epyc and Instinct: What we know so far
[9]Every conference is an AI conference as Nvidia unpacks its Vera Rubin CPUs and GPUs at CES
[10]Amazon's Trainium3 is the latest to conform to Nvidia's mold
For the moment, Upscale is primarily focused on scale-up networking products, but long-term it plans to expand its offering to more traditional scale-out switches. For this, Kar tells us, the company is still evaluating its options and may lean on third-party IP from partners.
With an additional $200 million from Tiger Global, Premji Invest, and Xora Innovation's coffers [11]in hand Upscale is now looking to expand its engineering, sales, and operations teams in anticipation of shipping its first AI networking productions later this year.
"We are already partnering with hyperscalers and GPU vendors who have validated the architecture. That is complete. Now the focus with this funding is to turn innovation into deployment," Kar said. ®
Get our [12]Tech Resources
[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[2] https://www.theregister.com/2024/03/18/nvidia_turns_up_the_ai/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2026/01/15/cisco_fixes_cve_2025_20393/
[8] https://www.theregister.com/2026/01/07/mi500x_amd_ai/
[9] https://www.theregister.com/2026/01/05/ces_rubin_nvidia/
[10] https://www.theregister.com/2025/12/07/trainium3_all_nvidia_nvl72_mold/
[11] https://upscaleai.com/from-100m-seed-to-unicorn-in-months-upscale-ai-closes-oversubscribed-200m-series-a-to-build-the-first-pure-play-ai-networking-company/
[12] https://whitepapers.theregister.com/
Founded last year, with backing from the likes of Intel, AMD, and Qualcomm, Upscale is developing a new class of scale-up network switches specifically designed to compete with the NVSwitch chips found in Nvidia's NVL72 racks.
NVLink is a high-speed interconnect technology that Nvidia developed to abstract memory and compute resources from multiple GPUs so they appear as a single logical resource.
[1]
The tech [2]debuted in 2024, and ever since, the likes of AMD and Cisco have tried to create alternatives. But their efforts so far, like Ultra Accelerator Link (UALink) and Ethernet for Scale-Up AI Infrastructure (ESUN) remain immature.
[3]
[4]
The first UALink-based rack systems from AMD will hit the market later this year, but they'll be tunneling the protocol over Ethernet. Purpose-built UALink switches capable of competing with Nvidia's NVSwitch simply aren't ready yet. Upscale aims to change that with its SkyHammer custom ASICS.
"Instead of retrofitting legacy systems, what we are re-imagining is what scale truly means in AI networking," Barun Kar, Upscale AI CEO told El Reg .
[5]
"The heart of this architecture is essentially to scale up. It's meant for AI workloads and nothing else."
The startup isn't ready to reveal technical details about SkyHammer, but has said it will offer its first silicon as a standalone ASIC for hyperscale integration, and as an integrated switch blade and rack.
While we don't have enough information about the chip to draw any comparisons to NVSwitch 6 or Broadcom's Tomahawk 6, Kar tells us it is using a memory semantic-based load-store network architecture, and will feature acceleration for collective communication similar to Nvidia's Sharp.
[6]
The platform will also support both UALink and the competing ESUN protocol.
To make the whole thing manageable at scale, Upscale is working to extend support for the SONiC network operating system (NOS). Originally developed by Microsoft, the open source NOS is both widely deployed and well understood by hyperscale customers.
[7]Cisco finally fixes max-severity bug under active attack for weeks
[8]AMD threatens to go medieval on Nvidia with Epyc and Instinct: What we know so far
[9]Every conference is an AI conference as Nvidia unpacks its Vera Rubin CPUs and GPUs at CES
[10]Amazon's Trainium3 is the latest to conform to Nvidia's mold
For the moment, Upscale is primarily focused on scale-up networking products, but long-term it plans to expand its offering to more traditional scale-out switches. For this, Kar tells us, the company is still evaluating its options and may lean on third-party IP from partners.
With an additional $200 million from Tiger Global, Premji Invest, and Xora Innovation's coffers [11]in hand Upscale is now looking to expand its engineering, sales, and operations teams in anticipation of shipping its first AI networking productions later this year.
"We are already partnering with hyperscalers and GPU vendors who have validated the architecture. That is complete. Now the focus with this funding is to turn innovation into deployment," Kar said. ®
Get our [12]Tech Resources
[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[2] https://www.theregister.com/2024/03/18/nvidia_turns_up_the_ai/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/networks&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aXGu-ay3IhlD6cYrxJ7iWAAAAs4&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2026/01/15/cisco_fixes_cve_2025_20393/
[8] https://www.theregister.com/2026/01/07/mi500x_amd_ai/
[9] https://www.theregister.com/2026/01/05/ces_rubin_nvidia/
[10] https://www.theregister.com/2025/12/07/trainium3_all_nvidia_nvl72_mold/
[11] https://upscaleai.com/from-100m-seed-to-unicorn-in-months-upscale-ai-closes-oversubscribed-200m-series-a-to-build-the-first-pure-play-ai-networking-company/
[12] https://whitepapers.theregister.com/