Akamai has announced two new solutions to strengthen its artificial intelligence offering in Akamai Cloud: Akamai Inference and Akamai Agent Sandbox. Both incorporate managed model services and isolated execution environments for AI agents, respectively. With these new features, companies will be able to create and tune open weight models, owned or hosted, use their own data and run agents in controlled environments, in addition to protecting applications and associated infrastructure. All from a single cloud platform optimized for artificial intelligence workloads.
These capabilities represent a new advancement in Akamai’s strategy, focused on providing organizations with the infrastructure and services necessary to develop, deploy and secure AI applications and agents close to users and the data they use. Akamai Inference and Akamai Agent Sandbox add a new layer of managed services to an infrastructure that has already closed multi-year contracts for Akamai Cloud worth more than $14 billion.
Robust data-centric architecture
The company is committed to combining CPU and GPU resources with cloud storage, networking and Edge Computing capabilities distributed across its global infrastructure. This architecture seeks to make it easier for companies to develop, implement and run large-scale artificial intelligence workloads, without giving up control of their data or the security of services.
Moving from pilot projects to production environments requires much more than AI models. Organizations need infrastructure to provide these services, access their data, manage request routing, and control application execution. Adding to these needs is the security of autonomous agents, especially when they begin to generate and execute code on their own. Until now, many companies used different providers to cover these functions and had to manage the protection of the different components separately.
Akamai Cloud aims to address this need through an integrated technology environment that combines distributed computing resources, from large GPU and CPU clusters in data centers to more than 4,350 locations at the edge. This infrastructure is complemented by a portfolio of security solutions that protect the different layers of the service.
“We have been implementing an AI infrastructure strategy based on the idea that AI inference will represent the majority of business focus and investment. Our customers have proven that thesis and taken us further,” said Adam Karon, COO and CEO of Akamai Cloud. «They want to create, run and protect AI on the same platform. Akamai Inference and Akamai Agent Sandbox add the inference and execution layers that turn our AI infrastructure into a complete AI technology environment, and our portfolio of security solutions protects each layer. “Akamai is uniquely positioned to offer our scalability and delivery to address real-time scale inference and agentic workloads.”
Akamai Inference: Access to AI models without managing infrastructure
Akamai Inference provides managed access to custom and open-weight models using OpenAI-compatible APIs. This feature allows the service to be integrated into existing applications without having to modify their programming. Customers can choose from different deployment modalities, including dedicated options and pay-as-you-go serverless inference services.
At launch, the catalog includes 12 open weight models, with sizes ranging from 8 billion to 2.8 trillion parameters. These include NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b and Google Gemma 4, as well as integrated models for recovery augmented generation (RAG).
The platform also makes it easy to access Akamai Cloud managed databases and object storage, as well as connect to public and private cloud services. In this way, the models can work with corporate information. Akamai Inference handles orchestration, model execution, and authentication, while requests are processed across its global network. The goal is to reduce latency, improve availability, and deliver consistent performance with predictable service levels.
Akamai Agent Sandbox: Sandboxes for AI Agents
The second novelty responds to the security challenges posed by the expansion of autonomous agents. These applications are no longer limited to answering queries, but can perform actions on behalf of users and generate and execute code to complete certain tasks.
Akamai Agent Sandbox provides developers with a controlled environment to run these agents and the code they generate. Agents run on top of Akamai Functions, while the code created by them runs in a separate, isolated environment. The solution is distributed as an open source component of the Spin framework and uses WebAssembly-based isolation mechanisms to set limits on compute resources, memory, outgoing connections, and access to system capabilities.
This model allows you to control where each process runs and what resources it can access. The proposal contemplates different isolation architectures, from micro virtual machines with the ability to create snapshots and forks to containers subject to security policies and functions isolated through WebAssembly.
This way, developers can select the most appropriate model for each workload and combine it with dedicated CPU services, NVIDIA-accelerated infrastructure, managed Kubernetes and Akamai Functions for serverless execution, both at the edge and in central data centers. The company also plans to incorporate new capabilities to control agent interactions and limit both the resources they can access and the connections they can receive.
More GPU capacity with NVIDIA Blackwell Ultra and Vera Rubin
Akamai has also expanded its GPU infrastructure with the addition of NVIDIA Blackwell Ultra clusters and is working to adapt its data centers to the future NVIDIA Vera Rubin platform, with support planned for 2027.
The GPU fleet now includes NVIDIA RTX PRO 6000 Blackwell Server Edition, aimed at low-latency inference services, and NVIDIA Blackwell Ultra clusters aimed at companies that need to develop, tune and host their own models on an infrastructure they can directly control.
The use of dedicated clusters also gives companies greater control over model versions and their update cycles.
Each Blackwell Ultra node integrates eight GPUs, with 288 GB of HBM3e memory per unit, for a total of 2.3 TB. This capability enables you to work with large models and workloads that require high memory availability, such as advanced reasoning, scalability over inference time, and agents that handle large contexts.
Using dedicated clusters also gives companies greater control over model versions and their update cycles, without the frequency limitations and queues of multi-tenant environments. Additionally, these resources are integrated with the data, connectivity, and security services available in Akamai Cloud.
“Enterprises should not have to choose between speed and control when bringing AI to production,” said Jon Alexander, senior vice president of Cloud Technology at Akamai. “Some teams want to launch an AI feature this quarter on a managed platform. Others need dedicated GPUs to run frontier-class open models on their own terms. All computers that run agents need a limit around the code those agents write. Akamai Cloud now supports them all on a single platform, with built-in security and resiliency.”
Akamai Inference and Akamai Agent Sandbox, the latter via the Spin SDK, are now available in beta. The first includes serverless deployment modalities. Companies interested in accessing NVIDIA B300 GPU clusters can consult the conditions and availability of this service through the Akamai sales department.
