A10 launches AI Gateway for enterprise model control

A10 Networks has launched A10 AI Gateway, with availability scheduled for the fourth quarter of 2026.

The software is designed to give organisations a single system for managing artificial intelligence models, applications and agents across their operations. It combines routing, cost management and governance in one platform as companies expand their use of generative AI tools and services.

A10 is targeting a problem many businesses face as AI use spreads across departments. Companies often rely on several model providers at once, making it harder to track which tools are in use, how much they cost, and whether access controls and security rules are being applied consistently.

The gateway is intended to sit at the centre of that activity. According to A10, it can direct requests to different models based on the task and any usage policies set by the organisation, while giving managers a central view of spending and model use.

One feature provides identity-based access controls through links with local directory systems. This allows an organisation to assign different models and policies to different teams, such as engineering or finance, based on user or group identity.

Another element is what A10 calls smart routing. Instead of relying only on fixed rules, the gateway assesses the type and complexity of each request, sending simpler work to lower-cost models while reserving more advanced tasks for models suited to reasoning-heavy workloads.

The product also provides per-request cost tracking in dollar terms, budget controls by model and team, and limits or alerts tied to token use. These controls are intended to give businesses closer oversight of spending as AI costs rise across a growing mix of internal and external services.

Security focus

A10 said the gateway is designed to work alongside its existing AI security products, including TrojAI, A10 AI Firewall and ThreatX. The company presents the new software as part of a broader set of tools covering testing before deployment, protections during runtime, and controls at network and API layers.

The deployment model is also central to the launch. A10 said the gateway can run as standalone software or with integrated hardware, and can be deployed within a customer's own environment, including on-premises systems, private cloud settings and air-gapped installations.

That approach reflects demand from organisations that want tighter control over where AI workloads run and where related data is stored. Data sovereignty, internal compliance rules and security requirements have become more prominent as businesses move AI from trials into day-to-day operations.

A10, which is listed in New York and says it serves more than 7,000 customers worldwide, has long focused on application delivery, network security and infrastructure software. The launch places the company more directly in the growing market for operational tools that manage AI use rather than only protect the networks around it.

Vendors across the technology sector are trying to define this layer of the market as companies seek systems that can impose policy, provide visibility and prevent uncontrolled spending on model access. Demand has grown alongside wider adoption of large language models and AI agents in software development, customer service, finance and internal knowledge tools.

Dhrupad Trivedi, President and Chief Executive Officer of A10 Networks, outlined the company's view in a statement.

"Enterprises are being asked to move fast on AI and stay in control at the same time. AI use is spreading faster than teams can discover, govern, or secure it, driving up cost and leaving no single place to set the rules. The A10 AI Gateway gives organizations one intelligent control plane to see how AI is being used, to govern who can access which models, and match every request to the right model at the right cost. Our integrated portfolio gives customers a complete, sovereign way to put AI to work at scale. That is how enterprises adopt AI with confidence instead of trading speed for control," said Trivedi.