Aria Networks Raises $125 Million In Funding

Listen to this article

Aria Networks has raised $125 million in its Series A funding round to accelerate the development and deployment of its AI native networking infrastructure.

Aria Networks’ $125 million Series A round represents the company’s first institutional funding and brings total capital raised to $125 million. Investors include Sutter Hill Ventures (lead), Atreides Management, Valor Equity Partners, and Eclipse Ventures. Atreides Managing Partner and CIO Gavin Baker and Sutter Hill’s Stefan Dyckerhoff have joined the board of directors alongside the founding team.

What is Aria Networks?

Aria Networks was founded in January 2025 in Palo Alto, California, by Mansour Karam. Karam previously founded Apstra, the intent based networking company acquired by Juniper Networks in 2020. The new venture builds on that foundation but shifts focus entirely to AI workloads, designing networking from the ground up rather than retrofitting existing architectures. The company operates in stealth until the simultaneous funding announcement and platform launch, positioning itself at the intersection of high performance Ethernet switching and AI driven intelligence.

What is Aria Networks’s technology?

Aria Networks has achieved general availability of its Deep Networking platform, branded as “Networks that Think.” This end to end system integrates purpose built hardware, hardened SONiC operating system, real time microsecond-granularity telemetry collected across every layer (switches, transceivers, and host NICs), and intelligent agents that operate autonomously at each stack level. A cloud layer agent leverages large language models for natural language insights, while an exposed MCP server enables seamless integration with external job schedulers and LLM routers.

The platform adopts a path centric architecture instead of the traditional switch centric model used by incumbents. Telemetry is extracted directly from the switching ASIC via embedded code on its ARM processors, enabling adaptive tuning of Dynamic Load Balancing, DCQCN, and failover logic in microseconds, far faster than threshold based or post facto monitoring tools. Lower layer agents react instantly to events such as transceiver flaps and reroute traffic in milliseconds, while higher layer agents make strategic flow placement decisions. Continuous cloud delivered software updates occur weekly, incorporating learnings from forward deployed engineers embedded with customers.

All models support DSP, LRO, and LPO optics and deliver the density and radix required for clusters scaling to 100,000+ accelerators.

How will Aria Networks use the funds?

The $125 million will fund accelerated hardware production, software refinement, global deployment of customer clusters, and expansion of the forward deployed engineering teams. Primary goals include scaling AI factories, maximizing token efficiency (tokens produced per dollar or per unit time), and sustaining peak Model FLOPS Utilization (MFU) as clusters grow. Early customer orders are already secured, with initial deployments focused on neo cloud operators and hyperscale AI training environments.

AI workloads have exposed critical limitations in conventional data center networking: even minor link degradation (e.g., a single faulty NIC in a 10,000-XPU cluster) can drop MFU by 1.7 percentage points, translating to tens of millions in lost annual revenue at current token pricing. Training clusters typically achieve 33–45% MFU, while inference often falls below 30%. Network performance directly determines whether clusters scale linearly or suffer cascading inefficiencies.

Aria’s platform addresses these by treating the network as an active, intelligent participant rather than passive plumbing. It works with any AI accelerator (Nvidia, Google, or others) preserving customer flexibility to upgrade chips without rip and replace networking overhauls. By embedding AI reasoning at every layer and delivering microsecond responsiveness, the system minimizes idle time, reduces retransmissions, and maintains high token throughput even at extreme scale. A 3% MFU gain in a 10,000-XPU cluster equates to roughly $49.8 million in additional annual value or a 7.9% improvement in token economics.

Recommended: Doss Raises $55 Million In Series B Funding Round

Incumbent vendors rely on coarser telemetry and manual or threshold driven interventions, resulting in slower reaction times and higher operational overhead. Aria differentiates through:

• ASIC embedded, real time telemetry at microsecond resolution.
• Layered autonomous agents that prevent single points of failure and hallucinations by grounding decisions in live cluster data.
• Weekly software velocity versus industry cycles measured in months.
• Path centric intelligence that optimizes end to end flows rather than isolated switch ports.

The architecture is purpose built for AI, enabling operators to achieve and sustain peak MFU while scaling clusters without proportional increases in power or complexity.

The participation of Sutter Hill Ventures (where Karam previously served as “entrepreneur in residence”), Atreides, Valor Equity Partners, and Eclipse Ventures signals strong conviction in both the founder and the category. Board additions of Gavin Baker and Stefan Dyckerhoff bring deep expertise in AI infrastructure investing and operational scaling. The round size, among the largest early stage commitments in AI networking, underscores the strategic importance of specialized fabrics as GPU clusters move from thousands to hundreds of thousands of accelerators.

Aria Networks enters the market at a pivotal moment when AI infrastructure spend is shifting from compute to the full stack, with networking emerging as a primary bottleneck for both training and inference economics. The “Networks that Think” platform positions the company to capture share in the rapidly expanding AI factory segment, where token efficiency and predictable MFU directly drive ROI for model developers and cloud providers.

With production hardware shipping, early deployments live, and a seasoned team executing at speed, Aria is poised to become foundational infrastructure for the next generation of large scale AI systems. The funding provides runway to refine the platform through real world feedback, expand geographic reach, and solidify partnerships across the AI ecosystem. As clusters continue their exponential growth, solutions that deliver measurable gains in token output per dollar will define competitive advantage, and Aria’s path centric, microsecond intelligent approach places it at the forefront of that shift.

Please email us your feedback and news tips at hello(at)superbcrew.com