A10 Networks has introduced A10 AI Gateway, an enterprise control plane designed to manage how employees, applications and AI agents access large language models. The product is scheduled to become available in the fourth quarter of 2026.
Rather than tying an organisation to one model provider, the gateway places a common routing and governance layer in front of multiple services. A10 says it can classify requests by complexity, send simpler work to lower-cost models and reserve more capable models for tasks that require advanced reasoning.
Routing decisions combine prompt complexity and policy
A10 describes the system as a single API endpoint through which teams can reach supported AI providers. Each request is assessed against routing policies and available budgets before it reaches a model. Administrators can define priority chains so traffic falls back to another model when a preferred option is unavailable or has exhausted its allocated budget.
The company’s product page lists upstream support for providers including OpenAI, Anthropic and Microsoft Azure. Provider credentials are stored in an OpenBao-based vault instead of being placed in application configuration files, according to A10. The company has not yet published a complete compatibility list, licensing structure or pricing.
Identity, budget and rate controls for each team
Access can be linked to an organisation’s existing identity and directory systems, allowing model permissions and routing policies to differ by user or group. A finance team, for example, could receive a different set of models and spending limits from an engineering team.
The gateway is also intended to provide live cost tracking for individual requests. Administrators can set token budgets by model, team or user, with hard limits and soft-limit alerts, while requests-per-minute and tokens-per-minute controls can be applied by key or team. These controls are meant to address a practical problem in multi-model deployments: AI bills and capacity limits are often spread across separate provider dashboards.
A10 says virtual API keys associate callers with their own budgets and rate limits. That abstraction may also reduce application changes when an organisation adds or replaces a provider, although the operational benefit will depend on the models, APIs and features supported at launch.
Designed for customer-controlled environments
A10 plans to offer the gateway as software or with integrated hardware, with deployment options for on-premises infrastructure, private clouds and air-gapped environments. Its product page says the single-tenant software can run through Helm on Oracle Kubernetes Engine without a GPU, while GPU-enhanced configurations will also be available.
That deployment approach is aimed at organisations that want routing policies, credentials and usage records to remain inside infrastructure they control. It does not by itself guarantee data sovereignty: buyers will still need to examine where each selected model processes and stores prompts, along with logging, retention, encryption and contractual terms across the full request path.
Part of A10’s broader AI security portfolio
A10 is positioning the gateway alongside TrojAI for model testing and guardrails, A10 AI Firewall for runtime protection, and ThreatX for web application and API security. The gateway itself focuses on orchestration, identity, cost and policy enforcement rather than replacing those security layers.
The combination reflects a wider shift in enterprise AI infrastructure. As organisations use more models and embed agents into business workflows, the management problem increasingly resembles an API and identity-control challenge, not simply a choice of chatbot.
A10 has not supplied independent performance comparisons or detailed launch pricing, and TTR has not tested the product. Prospective customers should evaluate routing accuracy, added latency, auditability, provider coverage and failure behaviour when production documentation becomes available.
More information is available on the official A10 AI Gateway product page and in A10’s solution brief.
Source: A10 Networks press release dated 18 September 2026 and official product materials.
