Celero Raises $275 Million After Validating 2nm Optical DSP for AI Data Centers
September 9, 2026
Celero Communications has raised $275 million in Series C funding after validating what it describes as the industry’s first coherent digital signal processor built on a 2-nanometer process. The company says the technology creates a path from 1.6-terabit optical connections to next-generation 3.2-terabit links for large artificial intelligence data centers.
The financing values Celero at more than $3 billion and brings its total capital raised to $415 million. Atreides Management, Valor Equity Partners and Alphabet’s CapitalG led the round, with Sutter Hill Ventures and Maverick Silicon also participating.
Celero’s $275 million Series C: key facts
The round values Celero at more than $3 billion.
Total funding has reached $415 million.
The company validated a 2nm coherent DSP with advanced analog technology.
Celero is targeting 1.6T and future 3.2T AI optical interconnects.
The capital will fund research, product development and production readiness.
Why AI data centers need faster networking
Training and serving advanced AI models requires thousands of accelerators to operate as one system. Each processor must exchange parameters, activations and results with other devices. When the network cannot move data quickly enough, expensive chips wait idle and the entire cluster becomes less efficient.
This is why connectivity is becoming as strategically important as compute. More accelerators increase the number of links, bandwidth demand and power consumption. The physical distance between racks and buildings also grows as clusters expand. Data centers need optical systems that move more information through fewer fibers with lower energy use.
What a coherent DSP does
A digital signal processor encodes, transmits and recovers information moving across an optical connection. Coherent technology uses properties of light such as amplitude, phase and polarization to carry more data and correct distortion. It has long been important in telecommunications networks and is moving closer to data-center systems.
Celero combines digital processing with advanced analog components. The company says its architecture can scale from 1.6T to 3.2T connections while improving bandwidth density and fiber efficiency. That could help AI infrastructure expand without increasing cables and power at the same rate as compute.
Why the 2nm process matters
Moving a complex networking chip to a leading manufacturing process can improve performance and energy efficiency, but it also raises design cost and execution risk. Celero says its 2nm validation demonstrates that the core processing and analog technology work together on the new node.
Validation is not the same as mass production. The company must still complete product development, qualify manufacturing, build supply relationships and prove reliability in customer systems. The Series C gives Celero more resources to move from technical milestone to commercial deployment.
From PAM4 toward coherent optics
Many current data-center links use PAM4 signaling, which encodes data through four amplitude levels. The approach has supported rapid bandwidth growth, but Celero argues that traditional architectures are approaching practical limits in density and fiber scalability as AI clusters become larger.
Coherent interconnects can carry more information per fiber and travel farther, making them attractive for scale-up connections within systems, scale-out links across clusters and scale-across networks between facilities. The transition will depend on cost, power and whether vendors can integrate the technology without excessive complexity.
The investors are backing an infrastructure bottleneck
Atreides, Valor and CapitalG are financing Celero at a time when spending on AI data centers is accelerating. The company cites an industry estimate that global data-center investment could approach $7 trillion by 2030, with roughly $5.2 trillion connected to AI workloads.
Celero’s opportunity does not depend on one accelerator brand or networking protocol. Every large cluster needs connectivity. If coherent optics become a standard component across multiple architectures, a supplier with efficient DSP silicon could sell into a broad and growing market.
Why Celero’s team matters
The company was founded in 2024 by engineers with experience at Marvell, Inphi, Broadcom and ClariPhy. Optical networking combines semiconductor design, analog engineering, digital signal processing and systems knowledge. Investors often place a premium on teams that have already delivered products through long qualification cycles.
Celero is headquartered in Irvine, California, with design centers in Canada and Argentina. The distributed footprint can provide access to specialized engineering talent, but coordinating a complex chip program across locations requires strong execution and supply-chain management.
Competition in AI connectivity
Networking companies, chipmakers and optical-component vendors are all pursuing the AI opportunity. Hyperscalers may also develop custom designs or negotiate aggressively with suppliers. Celero must prove that its architecture offers better total economics than established products and emerging alternatives.
Performance alone will not decide the market. Customers will evaluate power per bit, reach, reliability, packaging, software support and compatibility with their network designs. A technically impressive chip can still fail commercially if deployment requires expensive changes elsewhere in the system.
Production risk is the next challenge
Advanced-node semiconductors are expensive to manufacture and test. Celero needs foundry capacity, packaging partners and a dependable supply chain. Yield problems can increase cost, while long qualification cycles can delay revenue. The funding provides a buffer, but it also creates pressure to reach production at the scale implied by the valuation.
Customers will expect samples, interoperability testing and a credible roadmap. The shift from 1.6T to 3.2T must occur on a timeline that matches new accelerators and switches. Arriving too early increases carrying cost; arriving too late allows competitors to become the default choice.
Energy efficiency could become the decisive metric
AI infrastructure already places enormous pressure on electricity supply and cooling. A network link that carries more data with less power can reduce operating expense and help a facility stay within its energy limit. As clusters grow, small improvements per connection become significant across thousands of links.
Celero should provide independently verifiable measurements for power, bandwidth density and reach under realistic conditions. Buyers will compare the complete optical module and system, not only the DSP. Efficiency claims must survive integration with lasers, packaging and error correction.
What data-center operators should evaluate
Potential customers should test throughput, latency, error rates and power across expected distances and temperatures. They should examine interoperability with switches, accelerators and network-management software. Reliability and failure recovery are critical because a single weak link can reduce the performance of an expensive cluster.
Operators should also consider supply diversity. AI infrastructure depends on a concentrated semiconductor ecosystem, and a new vendor can reduce dependence only if it can manufacture at predictable volume. Long-term support and a transparent roadmap are essential.
What to watch next
The next milestones are product samples, customer qualification and production announcements. Partnerships with optical-module makers, switch vendors or hyperscale operators would provide evidence that Celero’s technology fits real deployment plans. Revenue and order visibility will matter as the company moves beyond engineering validation.
The broader signal is that venture capital is moving deeper into the AI hardware stack. Investors are not only funding models and applications; they are backing the networking, power and optical technologies required to make larger systems possible.
The bottom line
Celero’s $275 million Series C is a major wager on connectivity becoming the next AI infrastructure constraint. Its 2nm coherent DSP milestone is technically significant, but the company now has to convert silicon validation into reliable, economical products. If coherent optics become standard inside large AI clusters, Celero could occupy a valuable position between accelerators and the fiber that connects them.
Comments