Weka, a AI data and memory infrastructure company, announced production-scale benchmarks that show how organizations can improve the economics of long-context AI inference by serving more users and tokens on the same GPU footprint.The benchmarks show that Weka’s NeuralMesh platform with Augmented Memory Grid on Oracle Cloud Infrastructure (OCI) serves 10x more concurrent users, delivers 10x higher token throughput, and produces 7x more tokens per GPU than DRAM-only configurations without adding infrastructure. The results were validated on a nine-node OCI bare-metal H100 cluster with 100,000-token context windows.

“Enterprise AI workloads are pushing context windows and GPU utilization to new limits,” said Pablo Selem, senior director, software development, Oracle Cloud Infrastructure. “These benchmarks show how Weka’s NeuralMesh platform with Augmented Memory Grid on OCI helps remove memory bottlenecks so customers can support larger, more demanding inference workloads without simply adding more GPUs.”

Three Outcomes That Change the Math on Inference
Validated at production scale on a bare-metal H100 cluster (nine nodes, 72 GPUs, 100,000-token context windows, thousands of concurrent users), NeuralMesh with Augmented Memory Grid on OCI delivered:

• 10x more concurrent users served, without adding infrastructure. NeuralMesh with Augmented Memory Grid scaled past 5,000 concurrent users vs. about 600 for DRAM-only configurations. This eliminates the failure cliff that hits when cache saturates by expanding the active cache working set from 8.64 TiB of DRAM to 287 TiB of usable NVMe. In addition, more users per GPU means the same investment stretches further
• 10x higher token throughput. More output from every GPU in the cluster. On OCI, NeuralMesh with Augmented Memory Grid reached approx. two million tokens per second, compared to under 200,000 for the DRAM-only baseline. For product teams running real-time AI features, including search, summarization, code assist, and multi-turn agents, the throughput determines the ceiling for how many users can be served, how fast features respond, and how much revenue the infrastructure can support
• 7x more tokens served. Lower cost per token at scale. NeuralMesh with Augmented Memory Grid served five billion tokens, compared to 700 million for the DRAM-only baseline, in a single one-hour, 2,400-user test. For organizations running agentic workflows, DRAM saturation quietly drains GPU capacity through constant recomputation, creating a direct hit on cost per token and ROI

“Inference is bottlenecked by how much effective memory is available to GPUs,” said Liran Zvibel, CEO, Weka. “These results prove that AI token economics aren’t solved by hardware alone; they’re solved by eliminating the memory wall that has been the real ceiling on what existing hardware can do. NeuralMesh with Augmented Memory Grid running on OCI brings orders of magnitude more tokens to customers in an extremely cost-efficient way.”

Transforming AI Economics with Context Memory Infrastructure
As inference demand grows, AI infrastructure inefficiencies compound. Every key-value (KV) cache eviction is a tax: on GPU cycles, latency, user experience, and the cost of every token served. For long-context and agentic workloads, where inputs routinely run to 100,000 tokens or more, that tax is not a rounding error. It is a direct hit on the unit economics of every organization running production AI.

Augmented Memory Grid, a capability of NeuralMesh, solves the problem at the architectural level by decoupling KV cache from local GPU memory and storing it in a high-performance token warehouse accessible across the cluster. Any host can serve any session with cache hits intact, eliminating rigid session stickiness while delivering superior performance to DRAM, improving load balancing, and enabling clean horizontal scaling as concurrency grows. The result is persistent context memory for AI agents and the cost lever that makes long-context inference economical to run at scale.

Production-Grade Proof
OCI published the full benchmark methodology, system configuration, and results on its AI & Data Science blog on May 13, 2026. The benchmarks, executed on a nine-node OCI bare-metal H100 cluster, move beyond the prior phase of validation, which demonstrated 1000x more KV cache capacity and up to 20x faster time to first token at 128,000 tokens. This latest phase tests the full economics of inference in production: concurrency density, sustained throughput, cache persistence, and service level objective (SLO) stability when demand spikes under high load.

Available on Oracle Marketplace
NeuralMesh with Augmented Memory Grid is available to Weka customers and on the Oracle Marketplace, with OCI as Weka’s exclusive cloud launch partner. Organizations running long-context inference on OCI can deploy a validated, production-ready architecture today.