Andreessen Horowitz is putting $15 million behind Netris because AI clouds are running into a problem GPUs alone can't solve: the network has to keep up.
You can buy the GPUs and still lose time in the plumbing. That is the point behind Andreessen Horowitz's June 25 Series A for Netris, a network automation startup building software for the switches, routers, and interconnects that sit underneath GPU clusters. The round is $15 million, led by a16z, and the firm says partner Guido Appenzeller is joining the board.
The fashionable part of AI infrastructure is the chip order. Netris is working on the part most customers never see, which is exactly why the deal is worth your attention. GPU-native cloud providers such as CoreWeave, Lambda Labs, Crusoe, and Nebius are trying to sell AI capacity without the internal network tooling that AWS, Azure, and Google built over many years. When a new cluster still needs engineers touching individual switches, writing provisioning scripts, and manually separating tenants, the bottleneck isn't theory. It's the week you lose before a customer can run a workload.
According to Netris, its software can help operators get a GPU cloud business running in weeks rather than years, with tenant isolation configured automatically. That's a big claim, and it rests on a plain fact about this market: the new AI clouds want hyperscaler behavior without hyperscaler headcount.
That is the gap a16z is backing. Hyperscalers built their own internal network automation systems because they had the scale, the budget, and the engineers to do it. They have no reason to hand those systems to the smaller clouds now competing for AI workloads. Netris is trying to sell that missing layer as a product, not as a consulting project stitched together one deployment at a time.
Appenzeller's role is not incidental. The a16z infrastructure partner has argued that each major platform shift in computing produces new networking companies, and AI is making the case in a very practical way. Thousands of GPUs have to communicate constantly during training and inference. If the network fabric is slow to configure, hard to isolate, or painful to fill efficiently, the expensive chips sit inside a system that can't earn its keep fast enough.
Netris calls its product NAAM, short for Network Automation, Abstraction, and Multi-Tenancy. The name is clunky, but the job is specific. The company is positioning NAAM as a newer answer to the SDN and intent-based networking tools that shaped enterprise networking over the last decade, because those tools were not built for the multi-layer fabrics now used in GPU clusters. Netris says it has at least 35 live deployments across neoclouds, sovereign AI operators, and AI factories, and claims that is more than all other network automation vendors combined. It also says ARR grew 800% in the past twelve months.
Those numbers are the sharpest part of the story. A startup that was not widely known six months ago is telling investors it already sits inside dozens of AI infrastructure deployments. If those deployments keep expanding, the company is not just selling convenience. It is selling time back to operators that are already spending heavily on GPUs, power, real estate, and customer acquisition.
The hardware ties are also worth watching. Netris has been extending its platform to run on NVIDIA BlueField DPUs and integrating with NVIDIA DSX Air across Asia-Pacific infrastructure. Red Hat has partnered with the company on multi-tenant networking for sovereign AI clouds. The planned Singapore office later this year points in the same direction: governments and regional cloud operators are trying to build domestic AI capacity, and they don't want every network change to depend on a small group of senior engineers.
The $15 million will go toward hiring engineers and sales staff, adding support for more hardware vendors, and improving the algorithm that drives provisioning decisions. Don't skip that last piece. Configuration management gets you part of the way. The harder work is deciding how to reallocate capacity across tenants as workloads move, demand spikes, and expensive clusters need to stay full.
Frankly, this is the cleaner infrastructure bet than another round of noise about who can get the most GPUs first. A cloud provider can raise money, sign a hardware contract, and still disappoint customers if its operations team can't bring capacity online quickly. Netris is betting that the next AI infrastructure fight happens inside the network fabric. A16z is betting the same.
Also read: DeepSeek is doubling its headcount and the real question is whether its famous efficiency survives the growth • Adobe is buying Topaz Labs because building on-device AI from scratch would take too long • AI revenue has finally outpaced the cost of building the infrastructure behind it
