Upscale AI, the Santa Clara-based AI networking infrastructure company, secured a $190 million Series A-1 extension, pushing its total funding to about $500 million and achieving a $2 billion post money valuation.
Upscale AI, a Santa Clara, CA-based AI networking infrastructure startup, announced a $190 million Series A-1 extension. This brings its total funding to approximately $500 million in under 18 months since launch, with a post money valuation of $2 billion.
Funding History and Structure
- Seed round (September 2025): Over $100 million, establishing the company with a focus on open-standard AI networking.
- Series A (January 2026): $200 million oversubscribed round at over $1 billion valuation (unicorn status), led by Tiger Global, Premji Invest, and Xora Innovation, with participation from Maverick Silicon, StepStone, Mayfield, Prosperity7, Intel Capital, and Qualcomm Ventures.
- Series A-1 extension (June 2026): $190 million led by Premji Invest. New investors include Nvidia, Salesforce Ventures, Temasek, and Seligman Ventures; returning investors include Mayfield, Tiger Global, StepStone, Maverick Silicon, and Prosperity7.
This rapid capital accumulation reflects intense investor confidence in the AI infrastructure boom, particularly as hyperscalers and AI labs scale training and inference clusters. The extension strengthens the balance sheet for custom ASIC development, long lead time manufacturing commitments (e.g., with TSMC), and team expansion to several hundred employees.
What is Upscale AI?
Upscale AI positions itself as the first pure-play AI networking company, building full stack solutions (silicon, systems, software) optimized for AI workloads rather than retrofitting legacy cloud or enterprise networking. Its core thesis addresses a key bottleneck: inefficient communication between heterogeneous AI accelerators (GPUs, XPUs from Nvidia, AMD, etc.), memory, and storage in large scale clusters.
Key product/architecture elements:
- SkyHammer™: A purpose built scale-up ASIC and architecture for rack scale performance. It uses synchronized clocking for deterministic, nanosecond-level low latency communication, minimizing “stragglers” (slow packets that stall entire clusters) and enabling unified rack designs where accelerators share large memory pools. Targeted for synchronized AI training/inference; expected commercial availability in late 2026.
- Scale-out fabrics: Open Ethernet systems leveraging NVIDIA Spectrum-X switch silicon and SONiC based OS for heterogeneous, interoperable clusters. Supports open standards including UltraEthernet (UEC), UALink (for high radix, high bandwidth interconnects competing with Nvidia’s NVLink in multi vendor environments), OCP SAI/ESUN, etc.
- Overall approach: Open architecture for flexibility across compute vendors, purpose built economics to reduce bottlenecks and improve efficiency, and operational scale from racks to cloud scale deployments. Partnerships (e.g., with Nvidia for Ethernet) emphasize coexistence rather than direct replacement of proprietary solutions.
The company aims to democratize high performance AI networking by advancing open standards, reducing vendor lock-in (especially Nvidia’s dominance via NVLink/InfiniBand), and enabling better utilization in multi vendor environments critical for AGI-scale models.
Founders and Team
- Barun Kar (CEO): Founding team member at Palo Alto Networks; previously oversaw Juniper Networks’ Ethernet portfolio. Strong background in networking and security.
- Rajiv Khemani (Executive Chairman): Serial entrepreneur with successes in networking silicon. Co-founded Innovium (acquired by Marvell for ~$1.1B in 2021); prior roles at Intel (network processing) and Cavium. Also involved in other AI/hardware ventures like Auradine.
The team includes over 100 technologists from Marvell, Broadcom, Intel, Cisco, AWS, Microsoft, Google, etc., providing deep expertise across silicon, systems, and software for major computing shifts.
Recommended: InCharge Energy Raises $46 Million In Funding Round
AI data center networking is projected to exceed $100 billion annually by 2030, driven by massive capex from hyperscalers (Microsoft, Google, Meta, Amazon) estimated at $660–690 billion in 2026 alone. Legacy networks struggle with AI’s demands for ultra low latency, high synchronization, and massive scale in heterogeneous setups.
Upscale targets this by offering open, high performance alternatives to proprietary ecosystems. It aspires to be the “next Cisco” for the AI era, building the foundational plumbing as compute paradigms shift. Strengths include founder pedigree (multiple prior wins), open standards focus (backed by AMD, Intel, Google, Meta, Microsoft, etc.), and early traction signals via rapid funding and partnerships.
Faces Nvidia (dominant but proprietary), Broadcom, and rivals like Nexthop AI (raised $500M at $4.2B valuation). Challenges include proving open standards match proprietary performance, high development costs for advanced nodes, and supply chain commitments.
Strategic Implications of the Funding
- Acceleration: Funds product scaling (SkyHammer commercialization), R&D, go to market, and supply chain security.
- Investor validation: Participation from Nvidia and others signals ecosystem buy-in for hybrid/open approaches. Premji Invest’s repeat involvement (tied to prior exits) underscores execution confidence.
- Path forward: Positions Upscale for potential IPO or larger role in AI infrastructure. The $2B valuation in ~18 months highlights hype around AI infra but also execution risk in a capital intensive space.
This round cements Upscale AI as a high profile player in the AI networking surge, leveraging experienced founders, open innovation, and substantial capital to challenge incumbents and address critical bottlenecks in next generation AI clusters. Its success will hinge on timely delivery of competitive, interoperable solutions amid fierce competition and technical complexity.
Please email us your feedback and news tips at hello(at)superbcrew.com
Activate Social Media:
