AI Intern

Edge computing company Armada raises $230M at $2B valuation. The oversubscribed Series B will support growth as the company invests in modular data center production. Published on May. 20, 2026 Armada, an edge computing company providing modular AI infrastructure, secured $230 million in an oversubscribed Series B funding round co-led by Overmatch, BlackRock and 8090 Industries. The company, valued at $2 billion, also announced a framework agreement with Johnson Controls to establish production of Galleon Forge One, a modular data center, at a factory in Arizona. Galleon Forge One will span 400,000 square feet and is expected to create 500 jobs alongside various supply chain roles. Continuous production will commence in summer 2026, starting with Leviathan, Armada's megawatt-scale modular data centers for high-density AI training and inference workloads. In addition to the framework agreement, Armada secured an investment from Johnson Controls that furthers its ability to produce and deploy AI infrastructure where customers need it. "The AI race will not be won by one-off projects," Dan Wright, Armada's co-founder and CEO, said in a statement. "It will be won by the companies and countries that can manufacture, deploy and continuously improve AI infrastructure, with speed, scale and sovereignty. At Galleon Forge One, we will do what America does best: build the industrial base to win." Armada will invest its latest capital injection in supporting its growth initiatives and unlocking capacity for modular data centers.

Armada, a San Francisco, CA-based infrastructure and artificial intelligence company developing an edge computing and satellite-enabled platform, raised $230M in Series B funding

Pipeshift partners with Armada to bring open-source inference to Bridge Marketplace. Armada Bridge customers can now access Pipeshift for production open-source LLM inference May 19, 2026 Partnership Pipeshift is excited to partner with Armada.ai as part of the launch of Bridge Marketplace, Armada's new ecosystem for validated AI infrastructure software. Through the marketplace, Armada Bridge customers can access Pipeshift for production open-source LLM inference. Pipeshift brings MAGIC, its model optimization framework for serving workloads across latency, throughput, and cost constraints. The partnership gives Bridge customers a direct way to run optimized model endpoints on secure compute infrastructure provided through Armada. Teams can use Pipeshift to turn open-source LLMs into inference services for internal applications, customer-facing products, and regulated AI workloads. Once the GPU infrastructure is available, the serving path still has to be built. Armada Bridge gives businesses a way to access secure compute infrastructure for AI workloads. That includes the operational layer around multi-tenant access, scheduling, orchestration, billing, infrastructure services, and platform services. Production AI still needs model endpoints that hold latency, throughput, and cost behavior under real demand. Pipeshift works on that inference path, optimizing runtime decisions such as: * how the model is served * how requests move through the runtime * how the endpoint balances latency, throughput, and cost MAGIC tunes the serving path around each workload. MAGIC is Pipeshift's model optimization framework for production inference. After a team selects a model and compute target, the serving setup still has to match the workload. Chat traffic, batch jobs, and long-context pipelines each need different runtime behavior. Different AI workloads put pressure on different parts of the serving path: * Chat and agent endpoints: prioritize first-token latency, streaming behavior, and queue control during traffic spikes * Coding workloads: need stable decode performance across longer responses and concurrent users * Long-context and RAG workloads: put heavier pressure on prefill, memory movement, and cache reuse * Batch inference jobs: push for higher GPU utilization while keeping them away from latency-sensitive endpoints * Structured-output endpoints: need reliable formatting and tool-call behavior without adding heavy serving overhead The diagram below shows the flow from Armada Bridge compute infrastructure to a workload-specific inference endpoint. MAGIC gives each workload a serving path matched to its runtime pressure. A latency-sensitive endpoint, a long-context pipeline, and a batch job each need different runtime assumptions. Bridge customers can use Pipeshift to run open-source models as purpose-built inference endpoints on compute infrastructure provided through Armada. AI teams should be able to request model endpoints directly. For Bridge customers, Pipeshift makes secure compute infrastructure easier to use for production model serving. AI demand rarely arrives as one clean workload. A coding assistant, internal RAG system, support agent, and batch summarization job can run inside the same business with different serving requirements. Those requests may involve Qwen, Llama, DeepSeek, Mistral, Gemma, or another open-source model the team wants to run. Infrastructure teams need a repeatable serving layer for new models, workloads, and application requirements. And, the operators need a repeatable serving layer for new models, workloads, and tenant requirements. Pipeshift helps standardize that layer across open-source inference deployments. A new endpoint can start from a known serving pattern, then get tuned for the model, workload, and capacity target. Pipeshift is excited to support Armada customers building production AI infrastructure. Pipeshift is glad to support the launch of Armada Bridge Marketplace. Armada Bridge customers can find Pipeshift in the marketplace to explore production open-source LLM inference on secure compute infrastructure provided through Armada. Teams that want to discuss a specific model, workload, or deployment target can also reach out to Pipeshift directly.