Smartbird and WEKA have announced a partnership that puts WEKA NeuralMesh inside Smartbird’s dedicated AI clusters, sold as a managed service for enterprise AI workloads. Smartbird designs, builds, and operates each cluster for a single customer’s workloads, on-premises or in a dedicated colocation facility, and the companies describe the result as sovereign by design, with customer data staying in a customer-controlled environment and production-ready from day one. The agreement is the first in a planned network of Smartbird infrastructure partnerships.
Smartbird is the company formerly known as Allbirds. Allbirds sold its footwear brand and assets, changed its name to Smartbird, Inc. in June, kept the NASDAQ listing under the BIRD ticker, and named Dr. Nadia Carlsten, previously CEO of GPU compute provider DCAI, as president and CEO. Its first clusters were still being designed at the time of the rename, with data center sites under development in the U.S. and Europe, which makes the WEKA deal the first look at what the stack will contain.
Data Throughput and Inference Claims
NeuralMesh supplies the data and memory layer. The companies say GPUs stay continuously supplied with data at throughput reaching hundreds of gigabytes per second, but the companies have not identified the hardware configurations, software versions, or test conditions behind that figure. For inference and continuously running agentic applications, the companies say the combined platform expands GPU memory capacity, shortens time to first token, and reduces recomputation, which they tie to lower token costs.
WEKA’s answer to those three in NeuralMesh is Erweitertes Speichergitter, the KV cache tier that extends GPU memory into NeuralMesh storage, and WEKA’s figures for it are up to 1000x more GPU memory capacity, up to 20x faster time to first token, and 10x more concurrent users on the same GPUs, which it attributes to production benchmarks. Those are WEKA’s numbers for NeuralMesh in general; the Smartbird announcement does not say which NeuralMesh components the clusters use and carries no measured GPU utilization, power, or cost-per-token results for them.
“Enterprises don’t want to assemble AI infrastructure from individual parts. They want a complete solution that works, operated by people who have done this before,” said Dr. Nadia Carlsten, president and CEO of Smartbird. “Every Smartbird AI cluster is designed around the customer’s specific requirements, whether they’re a bank, a hospital, a manufacturer, or a government agency. By incorporating WEKA NeuralMesh, Smartbird gives customers a data and memory layer proven to perform at scale.”
“Agentic AI is unforgiving: idle GPUs and stalled pipelines land straight on the P&L statement. Enterprises have been told the fix is a tradeoff, performance or sovereignty, but that was always a failure of infrastructure, not a law of physics,” said Liran Zvibel, co-founder and CEO at WEKA. “WEKA NeuralMesh ends the tradeoff. Smartbird delivers it as dedicated clusters customers don’t have to build or operate. Together, we’re giving regulated industries production AI that runs at full speed, on infrastructure only they touch, with more output per GPU, per watt and per dollar.”
Dedicated Infrastructure and Managed Operations
Smartbird operates the full stack, covering facilities, compute, networking, data infrastructure, and data orchestration, so the customer has one accountable partner and does not integrate vendors or track multiple hardware release cycles. Each cluster is mapped one-to-one to its customer, and the companies say the fully dedicated footprint provides the isolation, encryption, and auditability that government, healthcare, and financial services workloads require. The companies have not named deployment jurisdictions, encryption implementation, or compliance certifications.
Although the companies describe the platform as production-ready from day one, there is no general availability date, no named customer, and Smartbird’s sites were still under development as of the June rename. NeuralMesh itself is the shipping product in the pair: WEKA took it to general availability in March and moved it to multi-tenant NeuralMesh 6 in July, with a parallel deal that puts it under Andromeda’s 50-plus GPU providers.
Amazon