Upscale AI

If proven, widely deployed, multi-year production references are required

• The vendor launched publicly in late 2025 and there is limited public end-user evidence for production deployments in the sources accessed.

If transparent security certifications and compliance artifacts are required up front

• Public documentation of compliance artifacts was limited in the materials retrieved, which can slow procurement.

If turnkey procurement simplicity and global support on day one is required

• Large incumbents typically have mature global support organizations, field engineering presence, and channel partnerships.

If the environment is tightly standardized on a single ecosystem (e.g., a full Nvidia stack)

• Switching costs and compatibility risks may outweigh potential benefits unless performance and TCO gains are demonstrated for the specific workload.

Current limitations acknowledged by the vendor

• No explicit public "known issues" page was found in the sources accessed.

Features on the roadmap

• The vendor indicates plans to expand from scale-up to scale-out and references partnerships; specific dates and deliverables were not enumerated in the materials retrieved.

Scale limitations

• Not publicly documented in accessible sources.

Security or compliance gaps

• Public documentation of certifications and attestations (SOC 2/ISO) was limited in the sources accessed.

API limitations or integration constraints

• Not publicly documented in the sources accessed.

• The vendor promotes alignment with SONiC/SAI and participation in several industry groups, presented as a differentiator versus proprietary approaches.

• If executed as planned, a single architecture optimized across layers could be a differentiator; however, this approach also raises execution, supply chain, and qualification risks.

• Vendor materials and third-party event coverage describe an architecture designed for AI networking demands; independent validation and production references are limited in the sources accessed.

• Test topology and network diagrams.
• Model/workload specifics and GPU/xPU type.
• Traffic patterns, congestion scenarios, and collective operation performance.
• Failure scenarios, upgrade procedures, and rollback behavior.
• Power, thermal, and cost model assumptions used in comparisons.

• Heads of Infrastructure, VPs of Data Center Operations, AI Platform Engineering leaders, CTOs / Chief Architects for AI infrastructure, and procurement/risk-management groups evaluating hardware and software vendors.

• Network engineers responsible for data-center fabrics, SREs and platform engineers operating clusters, and AI performance engineers optimizing collectives and utilization.