Netris has raised $15 million in Series A funding from a16z, aiming to tackle one of the least glamorous—and most time-consuming—parts of building AI infrastructure: getting the networking layer ready quickly enough that teams can actually start deploying models and applications.
The company’s thesis is straightforward. In the rush to stand up “neocloud” environments—AI-optimized cloud stacks that blend traditional infrastructure with specialized networking, orchestration, and security—most organizations discover that the bottleneck isn’t compute procurement or even storage configuration. It’s the operational friction required to make networks behave correctly at scale: provisioning, policy enforcement, traffic steering, telemetry, and the constant churn of changes that come with real workloads. Netris positions itself as software that runs directly on network switches, providing an “in-between” layer that helps operators reduce the time it takes to bring AI infrastructure online.
For AI teams, speed matters in ways that are easy to underestimate. A model training run might be scheduled weeks in advance, but the environment it depends on—routing policies, isolation boundaries, performance tuning, monitoring hooks, and failover behavior—often gets iterated right up until the moment workloads begin. When networking readiness lags, the entire deployment timeline stretches. That delay can mean missed windows for experimentation, slower iteration cycles, and higher costs from idle compute waiting for the network to catch up.
Netris’ approach is designed to compress that timeline by shifting more of the “go-live” work closer to where the decisions ultimately happen: at the switch. Instead of treating networking as a static configuration task handled after compute is ready, the company focuses on making the network programmable and operationally consistent as neocloud environments come online.
What makes this category different from typical network automation is the emphasis on reducing time-to-ready rather than simply improving configuration management. Many networking tools help teams automate repetitive tasks—pushing templates, generating configs, or integrating with orchestration systems. Those are useful, but they don’t necessarily solve the deeper problem: the operational complexity of ensuring that the network is correct for the specific workload patterns and policy requirements of AI deployments, especially when those deployments are dynamic.
In practice, AI neoclouds tend to be “alive.” They change frequently: new clusters come online, traffic patterns evolve as training jobs scale, security boundaries tighten, and observability needs expand. Each change can require careful coordination across multiple layers—switch configuration, routing behavior, access control, monitoring, and sometimes even application-level expectations. The result is that networking readiness becomes a project of its own, often staffed by specialists and gated by change-management processes.
Netris is betting that if the networking layer can be made more self-contained and workload-aware—through software that runs on the switch—operators can move from manual, multi-step readiness workflows to faster, more repeatable bring-up processes. The company’s platform is aimed at helping neocloud operators reduce the time it takes to go live, which implies not only faster initial provisioning but also faster iteration when something inevitably needs adjustment.
Why the switch matters more than people think
Network switches have historically been treated as relatively fixed infrastructure: you configure them, you verify them, and then you hope they behave predictably. But modern AI infrastructure is pushing networks into a more dynamic role. High-bandwidth, low-latency traffic patterns—especially those associated with distributed training—require careful handling. Even when the physical connectivity is in place, the logical behavior of the network determines whether workloads get the performance and reliability they were designed for.
By running software directly on switches, Netris can influence how traffic is handled and how policies are enforced at the point of forwarding. That matters because many of the “last mile” issues that slow down go-live—misaligned policies, inconsistent telemetry, delays in applying changes, and the difficulty of validating behavior end-to-end—are rooted in the gap between what operators intend and what the network actually does under load.
A switch-based software layer can also reduce the number of external dependencies involved in bringing up a new environment. If the logic for certain behaviors lives closer to the hardware, the system can be more resilient to delays elsewhere in the stack. It can also make validation more immediate: instead of waiting for a separate orchestration workflow to confirm that configurations propagated correctly, the switch can participate in enforcing and reporting the state needed for readiness.
This is the kind of shift that doesn’t always show up in marketing language, but it’s central to why time-to-go-live improves. When the network is treated as a set of configurations pushed from afar, every new environment requires a chain of steps: generate configs, apply them, verify them, coordinate approvals, and then troubleshoot edge cases. When the network becomes a programmable substrate with built-in logic, the chain shortens.
Neoclouds and the “in-between” layer
The term “neocloud” is often used to describe next-generation cloud environments optimized for AI workloads, but the reality is that these environments are not just “cloud plus GPUs.” They typically involve a more integrated approach to networking, security, orchestration, and performance management. That integration is precisely where the in-between layer becomes critical.
Compute provisioning is increasingly automated. Storage provisioning is also improving. But networking readiness remains a complex operational discipline because it sits at the intersection of performance engineering and security policy. It’s also where many organizations have accumulated legacy processes: change windows, manual verification steps, and specialized knowledge that doesn’t scale linearly with demand.
Netris’ positioning suggests it wants to become part of the standard workflow for neocloud bring-up. Rather than treating networking as a late-stage dependency, it aims to make the network a first-class component of the environment lifecycle. That means operators can bring up AI infrastructure faster without sacrificing correctness.
The company’s focus on “the in-between layer of cloud readiness” is a subtle but important framing. It implies that Netris isn’t trying to replace the entire neocloud stack—compute orchestration, storage systems, or higher-level AI platform tooling. Instead, it targets the transitional phase where the network must be ready for the rest of the system to function reliably.
That transitional phase is where many teams lose time. It’s also where failures are expensive. A misconfiguration can lead to degraded performance, intermittent connectivity issues, or security gaps that are difficult to detect until workloads are already running. By emphasizing readiness time, Netris is implicitly targeting both speed and risk reduction: faster go-live should come with fewer surprises.
What the Series A signals
A $15 million Series A from a16z is not just a vote of confidence in a product; it’s also a signal that investors see a durable market for infrastructure software that reduces operational friction in AI deployments. The AI infrastructure boom has created a wave of startups focused on model development, data pipelines, and developer tooling. But as organizations move from experimentation to production, the bottlenecks shift toward the systems that keep workloads running reliably.
Networking is one of those systems. It’s also one of the hardest to “bolt on” later. If a company builds its AI platform around assumptions that the network will be configured in a certain way, then any delay or inconsistency in networking readiness can ripple outward—affecting scheduling, scaling, observability, and incident response.
By backing Netris, a16z is effectively endorsing the idea that switch-based networking software can become a core component of AI infrastructure operations. That’s a meaningful bet because it requires trust from operators: they need to believe that the software can integrate cleanly with existing network environments, deliver measurable improvements, and maintain reliability at scale.
While the announcement emphasizes time-to-go-live, the underlying value proposition likely includes operational consistency. In infrastructure, consistency is often what enables speed. If every environment is brought up using a slightly different process, then even automation won’t fully eliminate delays. A platform that standardizes the readiness workflow—especially at the switch layer—can reduce variance between environments, which in turn reduces troubleshooting time.
How this could change the go-live playbook
To understand why Netris’ approach could be impactful, it helps to consider what “go-live” usually entails for AI neocloud operators. Typically, it includes:
1) Provisioning connectivity and verifying baseline reachability
2) Applying segmentation and access control policies
3) Ensuring routing behavior matches workload expectations
4) Enabling telemetry so operators can observe performance and diagnose issues
5) Validating that changes propagate correctly and that failure modes behave as expected
6) Coordinating changes across teams and change-management processes
Even when each step is automated, the overall workflow can still be slow because it requires coordination, verification, and iterative fixes. The switch-based layer Netris is building suggests it can streamline parts of this workflow by embedding certain behaviors and checks closer to the forwarding plane.
If the platform can reduce the number of manual steps required to validate readiness, then operators can shorten the time between “environment provisioned” and “environment ready for workloads.” That’s the difference between a neocloud that exists on paper and one that can actually support production-grade AI runs.
There’s also a second-order effect: faster go-live can enable more frequent environment creation. Teams that can spin up neocloud environments quickly are more likely to run experiments in parallel, test new configurations sooner, and iterate on performance tuning without waiting for long networking change cycles. Over time, that can improve the organization’s ability to learn from production-like conditions.
A unique angle: moving beyond configuration to operational readiness
Many networking products focus on configuration management. Netris’ framing is different: it’s about operational readiness. That distinction matters because operational readiness includes not only “is the config correct?” but also “does the system behave correctly under real workload conditions?” and “can we observe and manage it effectively once it’s live?”
Running software on switches can support this by enabling more direct enforcement and more immediate visibility. Instead of relying solely on external monitoring systems that infer behavior after the fact, the switch layer can participate in the readiness process—helping ensure that the network is in the expected state before workloads