Upscale AI
Upscale AI builds open-standard AI networking infrastructure for hyperscalers.
Upscale AI develops high-performance AI networking infrastructure combining custom silicon (SkyHammer ASICs), systems, and software to enable ultra-low-latency connections between GPUs, AI accelerators, and storage in large-scale AI clusters. The company competes directly with Nvidia's NVSwitch by building open-standards networking solutions designed for hyperscalers and AI infrastructure operators seeking scalable alternatives to proprietary architectures. Founded by serial entrepreneurs from Palo Alto Networks, Innovium, and Cavium, Upscale is backed by $300M+ in funding and positioning open standards as superior to closed ecosystems for AI infrastructure.
Problem solved
Hyperscalers and AI operators lack open-standard, vendor-neutral networking solutions optimized for ultra-low-latency AI workloads, forcing reliance on Nvidia's proprietary NVSwitch technology.
Target customer
Hyperscalers, cloud providers, and AI infrastructure operators building large-scale GPU clusters and seeking alternatives to Nvidia's proprietary networking solutions.
Founders
B
Barun Kar
CEO
Former VP of Engineering at Palo Alto Networks leading large-scale security platform development; co-founder at Auradine building blockchain and AI compute infrastructure.
R
Rajiv Khemani
Executive Chairman
Serial entrepreneur with 25 years experience; co-founder and CEO of Innovium (acquired by Marvell 2021), former CEO of Auradine, COO at Cavium (acquired by Marvell); helped grow Cavium from startup to $500M+ annual design wins.
Funding history
Seed
$100M
September 2025
Led by Mayfield, Maverick Silicon
· StepStone Group, Celesta Capital, Xora, Qualcomm Ventures, Cota Capital, MVP Ventures, Stanford University
Series A
$200M
January 21, 2026
Led by Tiger Global, Premji Invest, Xora Innovation
· Maverick Silicon, StepStone Group, Mayfield, Prosperity7 Ventures, Intel Capital, Qualcomm Ventures
Total raised:
$300M+
Pricing
Not publicly available. Company is in early product development with initial products slated to ship in 2026.
Integrations
Based on open technologies: SONiC, Ultra Accelerator Link (UAL), Ultra Ethernet (UE), Switch Abstraction Interface (SAI)
Tech stack
jQuery Migrate (JavaScript libraries)
jQuery (JavaScript libraries)
ZURB Foundation (UI frameworks)
Bootstrap (UI frameworks)
WordPress (Blogs)
Google Analytics (Analytics)
HSTS (Security)
Nginx (Reverse proxies)
Varnish (Caching)
PHP (Programming languages)
Fastly (CDN)
Unpkg (CDN)
MySQL (Databases)
MariaDB (Databases)
DoubleClick Floodlight (Advertising)
Google Tag Manager (Tag managers)
Yoast SEO (SEO)
Pantheon (PaaS)
CookieYes (Cookie compliance)
UserWay (Accessibility)
Gravity Forms (WordPress plugins)
GoDaddy (Hosting)
Website
Competitors
Nvidia NVSwitch
Proprietary closed-ecosystem switch technology; Upscale aims to provide open-standards alternative with comparable performance.
Cisco
General-purpose data center networking; Upscale is purpose-built specifically for AI training and inference workloads.
AMD
Broader semiconductor and networking portfolio; Upscale focuses narrowly on AI cluster interconnects with open standards.
Why this matters: Upscale AI is attacking one of the most defensible parts of Nvidia's moat—proprietary networking infrastructure for AI clusters—by betting that open standards can compete with Nvidia's integrated approach. With $300M+ in funding from top-tier VCs and leadership from proven infrastructure entrepreneurs, the company represents a significant attempt to break Nvidia's lock on the AI infrastructure stack.
Best for: Hyperscalers and AI infrastructure operators building next-generation GPU clusters who want to avoid vendor lock-in and leverage open-standards networking architectures.
Use cases
Large-Scale AI Training Clusters
Hyperscalers deploying thousands of GPUs for foundation model training can use Upscale's ultra-low-latency networking fabric to maintain GPU synchronization and reduce training times compared to general-purpose data center networks.
Multi-Vendor AI Infrastructure
Cloud providers building heterogeneous AI infrastructure with multiple GPU types and accelerators can use Upscale's open-standards approach to avoid proprietary lock-in and integrate different hardware vendors into unified clusters.
Edge-to-Cloud AI Deployments
Organizations running distributed AI inference across edge and cloud environments can leverage Upscale's rack-scale networking platforms to maintain consistent low-latency performance across hybrid infrastructure.
Alternatives
Nvidia NVSwitch
Proprietary, tightly integrated with Nvidia GPUs; choose Upscale for open-standards flexibility and vendor-neutral approach.
Broadcom BCM Switches
General-purpose data center switching without AI-specific optimization; choose Upscale for ultra-low-latency AI workloads.
Intel Networking Products
Broad networking portfolio with legacy-focused architecture; choose Upscale for modern AI cluster optimization.
FAQ
What does Upscale AI do? +
Upscale AI develops high-performance AI networking infrastructure combining custom silicon (SkyHammer ASICs), systems, and software designed to enable ultra-low-latency connections between GPUs and AI accelerators in large-scale clusters. The company builds open-standards alternatives to Nvidia's proprietary NVSwitch technology, enabling hyperscalers to avoid vendor lock-in while maintaining breakthrough performance for AI training and inference workloads.
How much does Upscale AI cost? +
Pricing is not publicly available. Upscale AI is in early product development with initial solutions slated to ship in 2026. Contact the company directly for enterprise pricing.
Who founded Upscale AI? +
Upscale AI was founded by Barun Kar (CEO) and Rajiv Khemani (Executive Chairman), serial entrepreneurs with deep experience at Palo Alto Networks, Innovium, Cavium, and other infrastructure companies. The company was incubated within Auradine, Kar's previous AI infrastructure startup.
Who uses Upscale AI? +
Target customers are hyperscalers, cloud providers, and AI infrastructure operators building large-scale GPU clusters. No named customers have been publicly disclosed yet, though the company reports strong early traction.
How does Upscale AI compare to Nvidia NVSwitch? +
Nvidia NVSwitch is proprietary technology tightly integrated with Nvidia GPUs, creating vendor lock-in. Upscale AI offers open-standards networking with comparable performance, enabling customers to build multi-vendor AI clusters and avoid dependency on a single supplier.
Tags
AI networking
GPU interconnects
data center infrastructure
open standards
semiconductor
cloud infrastructure
hyperscaler