CScale, a Silicon Valley startup, has raised $145 million in Series C funding as it comes out of stealth mode. The company is building optical interconnect technology for large AI systems. Its goal is to help thousands of AI chips work together as one powerful computer.
The funding is notable because Nvidia and Intel Capital have joined the round as CScale’s first strategic investors. The round was co-led by Atreides Management, Valor Equity Partners, and Premji Invest. Existing backers Sutter Hill Ventures and Maverick Silicon also took part.
With this new capital, CScale has now raised $188 million in total. The company was founded in 2023 and has its headquarters in Palo Alto, California. Its chief technology officer and co-founder is Sanjai Kohli. Martin Lund serves as chief executive.
The company says the new money will help it speed up product development and commercial work. CScale is focused on a part of AI infrastructure that may become more important as AI data centers grow much larger.
The Problem Is Bigger Than AI Chips
The AI industry has spent huge amounts of money on faster chips. Nvidia and other chip companies have built powerful accelerators that can process AI workloads at very high speeds.
But faster chips alone do not solve every problem.
When a large AI system has thousands of chips, those chips must exchange huge amounts of data. They need to share information quickly so the full system can work as one unit.
This creates a major networking challenge.
Imagine a large group of workers who need to share information before they can finish a task. If the workers have fast computers but a slow way to communicate, the whole team can still lose time.
The same basic idea applies to AI chips.
A chip may finish its own work very quickly. But if it has to wait for data from another chip, some of that computing power can sit unused.
CScale wants to address this problem with optical connections that use light to move data between AI chips and other parts of a large computing system.
Why Optical Connections Matter
Traditional electrical connections have limits when data must travel across very large AI systems.
Optical technology uses light to move information. This approach has become more important as AI systems have grown larger and demand for data movement has increased.
CScale is focused on what the industry calls AI scale-up networking. This refers to the high-speed links that connect accelerators inside a large AI computing system.
The purpose is not simply to connect separate data centers. The goal is to allow many accelerators across multiple racks to act like parts of one large computer.
CScale says future AI systems could contain thousands of tightly connected accelerators across dozens of racks. At that scale, the network between the chips becomes a critical part of the overall system.
This is why CScale is focused on the connection layer rather than on the AI chips themselves.
CScale Has a Different Focus
CScale is not trying to build another AI accelerator.
Instead, it wants to build the infrastructure that allows accelerators to communicate.
This is an important difference.
An AI accelerator performs calculations. A network connection moves information between those accelerators. If the connection is too slow or unreliable, the full system may not get the value that the chips can provide.
CScale describes its product as an optical interconnect for AI scale-up systems. The company says it is designing the technology for high bandwidth and very low latency. These qualities are important when thousands of chips must exchange information at very high speed.
The startup has not released full technical details about its product. That means there is still much that is not public about its exact design, performance figures, power use, or commercial product plans.
The Key Idea Is Reliability
Speed is only one part of CScale’s plan.
The company is also focused on what happens when part of an optical network fails.
CScale says its interconnect architecture can contain optical failures so that a failure does not stop the wider computing system.
This is a major issue for very large AI clusters.
A small failure may be rare at the level of one connection. But when a system contains a huge number of connections, failures can become a regular part of normal operation.
CScale’s basic idea is simple: if one optical component stops work, the rest of the AI system should not have to stop with it.
The company has used the phrase “Lasers will fail. Compute shouldn’t.” Chief executive Martin Lund has said that reliable and predictable communication is essential for the economics of large AI systems.
Why Nvidia Is Part of the Deal
Nvidia’s role makes this funding round especially notable.
Nvidia is one of the world’s largest suppliers of AI accelerators and related data center technology. Its chips are used in many of the large AI systems that power modern models.
CScale is not a chip maker. Its work sits at a different layer of the AI infrastructure stack.
Nvidia joined the CScale round as a strategic investor. Intel Capital also joined in the same role. They are CScale’s first strategic investors, according to the company.
The investment does not, by itself, establish a supply agreement, product partnership, or future technology plan between the companies. The available reports describe the two companies as investors in CScale.
Still, their participation puts more attention on the problem that CScale is trying to solve.
Intel Also Sees Value in Optical Networking
Intel Capital’s participation adds another major semiconductor name to the CScale story.
Intel has long worked across processors, data center technology, networking, and related hardware. Its investment gives CScale another major technology company as a strategic backer.
The two investments also show that interest in AI infrastructure extends beyond the main AI accelerator market.
The next stage of AI growth may depend not only on better processors but also on memory, power systems, cooling, networking, optical technology, and data center design.
CScale sits within that wider infrastructure market.
The $145 Million Round
The $145 million Series C was co-led by Atreides Management, Valor Equity Partners, and Premji Invest.
Sutter Hill Ventures, which has backed CScale since its start, also took part. Maverick Silicon was another existing investor that joined the round.
The total funding now stands at $188 million.
That is a large amount of capital for a company founded only three years ago. It also gives CScale more resources to move its technology from development toward commercial use.
Premji Invest has said it was impressed by the depth of CScale’s team. Its comments point to experience across high-speed electronics, photonics, packaging, software, communications, and computing systems.
These areas all matter for optical interconnect products because such systems require more than just a good optical component.
CScale’s Team Has Deep Hardware Experience
CScale was founded in 2023 by Sanjai Kohli, who serves as the company’s chief technology officer.
The company says its wider team has experience across several parts of advanced computing and communications technology.
That background is important because optical networking is a complex field.
A commercial product must work at very high speeds and must also meet strict requirements for reliability, power, size, cost, and system integration.
CScale is therefore trying to solve both a technology problem and a product design problem.
Its success will depend on whether it can make optical interconnects practical for very large AI systems.
AI Data Centers Are Getting Much Larger
AI has changed the design of data centers.
Older data centers often focused on general computing, web services, databases, and storage. Modern AI centers need huge amounts of compute power in one location.
Large AI models can require thousands of accelerators.
As these systems grow, more chips must communicate with one another. That puts pressure on the connections between them.
CScale expects future gigawatt-class AI data centers to contain large groups of tightly connected accelerators. One scale-up domain could span thousands of accelerators across dozens of racks.
This creates a market for companies that can improve the way those chips communicate.
A Failure Can Become a Bigger Problem
Large AI systems are made of many parts. Even if each part has a low chance of failure, a very large system can still face regular faults.
For example, if a system has a small number of optical links, a single failure may be unusual. If the system has a huge number of links, the chance that at least one part needs service becomes much higher.
That creates a challenge for operators.
A failed component should not force a full AI cluster to stop. If it does, expensive computing hardware may sit idle while engineers repair the problem.
CScale’s architecture aims to contain such optical faults.
The company says this approach can allow the larger computing system to continue work even when an optical part fails. This claim comes from CScale, and the company has not yet shared enough public technical data for an independent assessment of its performance.
CScale Has Not Shared Every Detail
CScale has spent much of its short life in stealth mode.
That means the company kept many details about its technology away from the public while it worked on its product.
Now that it has exited stealth, it has shared its main goal and funding details. But several important technical details remain private.
Public reports do not provide complete figures for bandwidth, latency, power use, port count, or manufacturing technology.
The company has also not publicly named a customer or given a firm product shipment date in the available reports.
This is important context when looking at the $145 million deal. The funding shows strong investor support, but it does not prove that the technology has already reached large-scale commercial use.
Why This Matters for AI
The AI industry has focused heavily on compute.
Companies have spent billions of dollars on chips and data centers to train and run larger AI models.
But as the number of chips grows, the connections between those chips become more important.
A powerful AI cluster needs both fast compute and fast communication.
If the chips cannot exchange data at the required speed, part of their potential performance can be lost.
CScale is targeting this gap.
Its idea is that the next stage of AI infrastructure needs networks that can support large groups of accelerators without creating major delays or full-system disruption.
A Wider Shift in AI Infrastructure
CScale’s funding also reflects a wider change in the startup market.
Investors have poured large sums into AI model companies and software startups. Now, more attention is also moving toward the physical systems that support AI.
That includes chips, memory, networking, optical parts, power systems, cooling, and data centers.
CScale belongs to this hardware and infrastructure group.
Its $145 million round shows that investors see potential in technologies that can help AI systems grow to a much larger scale.
The company does not need to build an AI model itself. It wants to provide one of the basic systems that lets AI models run across many chips.
What Comes Next for CScale
CScale now has a large amount of capital and backing from several major investment firms and technology companies.
The next step is to turn its technology into a commercial product.
That will require more product development, tests, system integration, manufacturing work, and customer adoption.
The company says the new funds will help speed development and commercialization of its optical interconnect technology.
The challenge is significant. CScale must show that its design can work at the scale required by future AI systems while also meeting practical requirements for cost, reliability, power, and maintenance.
For now, the company has put its main idea into the public view.
A Bet on the Next AI Bottleneck
CScale’s $145 million Series C is more than another large AI startup funding round. It points to a problem that may become more important as AI systems grow.
The world has built faster AI chips. The next challenge is making thousands of those chips work together with speed and reliability.
CScale believes optical interconnect technology can help solve that problem.
The company now has $188 million in total funding, backing from Nvidia and Intel Capital, and support from Atreides Management, Valor Equity Partners, Premji Invest, Sutter Hill Ventures, and Maverick Silicon.
Its technology is still under development, and many technical details remain private. But its goal is clear: build a better connection layer for very large AI systems.
As AI data centers grow from large clusters into even bigger computing systems, the links between chips may become just as important as the chips themselves. CScale is now betting that this will create a major new market for optical networking.
Also Read – B2B Startup Sales: Founder-Led Sales to Repeatable Pipeline