Upscale AI Raises $190 Million to Fix the Biggest Problem in AI Infrastructure

Nobody talks about networking at dinner parties. They talk about ChatGPT. They talk about robots and self-driving cars. But behind every one of those flashy AI demos is a data center full of chips that need to talk to each other at insane speeds. And right now, that communication layer is completely broken.

Upscale AI, a Santa Clara-based startup, just raised $190 million to fix it. The company closed a Series A-1 round, pushing total capital raised to $500 million and valuation to $2 billion. In under 18 months. Let that sink in.

What Is Upscale AI and What Does It Do?

Upscale AI is a pure-play AI networking infrastructure company. Not an AI app. Not a chatbot wrapper. Core infrastructure, the kind that makes everything else actually work.

Here is the kicker. Most AI chips today cannot talk to each other efficiently unless they come from the same manufacturer. Picture a home Wi-Fi network where your laptop only connects if every single device in the house is made by the same brand. Frustrating at home. Catastrophic at data center scale when you are running thousands of GPUs in parallel.

Upscale AI is building what it calls an open-standard AI networking fabric. A full-stack platform covering silicon, systems, and software. It connects accelerators, memory, and storage into one high-performance AI engine, purpose-built for large-scale training and inference workloads.

The company was incubated inside Auradine, a blockchain and AI compute infrastructure firm, before spinning out on its own.

Upscale AI Raises $190 Million in Series A-1 Funding

On June 22, 2026, Upscale AI officially closed a $190 million Series A-1 round. This is an extension of the earlier $200 million Series A from January 2026.

So here is the full picture. Upscale AI launched in September 2025 with a $100 million seed. Four months later, a $200 million Series A and unicorn status. Six months after that, another $190 million. The pace is not normal.

And it matters because building in the chip and networking space is brutally capital-intensive. You need money to commit to suppliers. You need money to lock in production slots, sometimes two years in advance. You need money before a single product ships. With $500 million on the balance sheet, Upscale AI can now have those conversations from a position of strength.

CEO Barun Kar said it straight: AI infrastructure is being redefined at cluster scale, and networking is one of the most critical bottlenecks. He is not wrong.

Who Invested in Upscale AI’s Latest Funding Round?

The Series A-1 was led by Premji Invest, which also led the Series A. Familiar face. Strong signal.

New investors coming in for the first time include NVIDIA, Salesforce Ventures, Seligman Ventures, and Temasek. Returning investors who wrote another check include Maverick Silicon, Mayfield, Prosperity7 Ventures, StepStone Group, and Tiger Global.

Now, stop for a second on NVIDIA.

Upscale AI is explicitly building open-standard alternatives to NVIDIA’s own proprietary networking technologies. And NVIDIA still backed the round. That is not a contradiction. That is NVIDIA placing a bet on where the industry is heading, even if it competes with parts of their own stack. Smart money does not ignore reality.

Premji Invest’s Sandesh Patnam framed it cleanly. Upscale AI delivers proprietary-grade performance while embracing open standards. That combination of technical conviction and real-world operating insight is exactly what drew continued support.

StepStone’s John Avirett added that the team is building with a long-term view of how AI infrastructure needs to evolve. And Prosperity7’s Abhishek Shukla called the decision to deepen their investment a reflection of long-term conviction in both the team and the market opportunity ahead.

Upscale AI Hits $2 Billion Valuation and $500 Million in Total Funding

Let’s be honest. A $2 billion valuation in under 18 months is the kind of number that raises eyebrows. But the funding trail tells you why investors are leaning in hard.

Seed round, September 2025: $100 million. Series A, January 2026: $200 million, valuation hits $1 billion. Series A-1, June 2026: $190 million, valuation doubles to $2 billion. Total raised: $500 million.

Going from $100 million seed to $2 billion valuation in less than two years is rare. And it is not happening because of hype. It is happening because the problem Upscale AI is solving is real, the market is enormous, and the founding team has done this before.

Executive Chairman Rajiv Khemani put it simply. The team is making rapid progress building out the product portfolio, and this investment helps execute on immediate priorities while advancing long-term ambitions.

Why AI Networking Is the Biggest Bottleneck in Data Centers Right Now

Here is something most people outside of infrastructure circles do not fully grasp. Raw compute power is only half the problem.

When you are running a massive AI training job across thousands of GPUs, those chips need to constantly share data with each other. Weight updates. Gradient calculations. Activation outputs. The speed at which that communication happens determines whether your cluster is actually running at full capacity or leaving performance on the table.

Traditional networking was never built for this. It was designed to connect endpoints, servers, laptops, storage. AI networking needs to unify an entire cluster as if it were one giant brain. That is a completely different engineering challenge.

The AI networking market is projected to exceed $20 billion. And as generative AI drives demand for ever-larger computing environments, that number is going up, not down.

Upscale AI builds its platform on open standards, including SONiC, Ultra Accelerator Link, Ultra Ethernet, and Switch Abstraction Interface. The team also actively participates in the Ultra Ethernet Consortium, the Open Compute Project, and the SONiC Foundation. Not just building products. Shaping the standards.

How Upscale AI Plans to Compete with NVIDIA’s Dominant Networking Tech

This is the most uncomfortable question in the room. And it deserves a straight answer.

NVIDIA currently owns AI data center networking through two proprietary technologies. NVLink handles chip-to-chip communication inside a server. InfiniBand handles cluster-level networking across servers. Both are powerful. Both lock customers completely into NVIDIA’s ecosystem. If you want peak performance, you buy NVIDIA chips and NVIDIA networking. Full stop.

The bet Upscale AI is making is that this model breaks down as AI infrastructure scales. Enterprises want flexibility. They want to mix chips from different vendors. They want to avoid a world where one company controls their entire compute stack.

So the goal is clear: deliver the performance of a proprietary system with the openness of a standards-based one. CEO Barun Kar put it in plain terms. Upscale AI is trying to get GPUs to talk the same language, regardless of manufacturer.

Rajiv Khemani does not frame this as a war with NVIDIA. His view is that the future AI stack will be heterogeneous, with multiple chip architectures running across the same infrastructure. That is the world Upscale AI is positioning for.

But the competitive pressure is real. Nexthop AI, a direct rival targeting the same hyperscaler and neocloud customers, raised $500 million in a Series B in March 2026 at a $4.2 billion valuation. The race to become the Cisco of the AI era is already heating up. There will not be twenty winners here.

What’s Next for Upscale AI After This Massive Fundraise?

The $190 million has a job to do.

First priority is scaling operations and accelerating delivery of AI-native networking technology. Second is locking in supply chain commitments. Custom chip design at today’s advanced manufacturing nodes costs hundreds of millions before anything ships. Suppliers want committed purchase orders, sometimes two years out. A strong balance sheet is not optional in this business. It is the price of admission.

The company has already grown to several hundred employees in under a year. That is fast. Very fast. And it signals that execution is happening, not just planning.

The long-term ambition is to be the defining open-standards networking company of the AI era. The infrastructure layer that lets the world’s AI chips work together regardless of who built them. Investors from NVIDIA to Temasek to Tiger Global are putting real money behind that ambition.

The reality is, the AI infrastructure buildout is still in early innings. Most of the compute clusters that will power the next decade of AI have not been built yet. Networking is where the next great infrastructure company gets born.

Upscale AI is making its case that it will be that company. And right now, the market is listening.

Seedcamp Raises $320M

Upscale AI’s MASSIVE $100M