Weka launches AI storage platform to cut expensive GPU demand
Weka's new NeuralMesh 6 platform uses cheap flash memory to cache AI computations, offering enterprises a way to lower inference costs and defer expensive GPU purchases.
Weka has launched NeuralMesh 6, a software platform paired with its first proprietary hardware line, Wekapod 3, designed to reduce the strain on scarce GPU memory in artificial intelligence operations. The system uses cheaper NAND flash to act as an extension of GPU memory, caching pre-calculated data to prevent redundant processing.
GPU memory remains the most expensive and constrained resource in production AI. As enterprises deploy internal copilots and customer service agents with long context windows, models are forced to repeatedly recompute past interactions in multi-turn conversations.
"If you have 10 turns, you may overcalculate 100 times because you're redoing all of them," said Weka co-founder and CEO Liran Zvibel. "You can put two orders of magnitude more NAND than you could afford in shared memory, and we can cache 100% of the pre-calculated tokens, so you never need to redo it."
For companies running AI at scale, the financial implication is straightforward: better utilization of existing GPU investments and lower inference costs without waiting months for additional capacity. "What we're seeing now with customers is they're chasing availability of compute, and once they get new allocation from anyone, they want to be able to grab it and start running right away," Zvibel said.
The release enters a crowded market where legacy vendors are rapidly repositioning. "The storage world is shifting its focus from serving bits to enterprise workloads to managing data at the speed of AI," said Steve McDowell, chief analyst at NAND Research. He noted that while Dell, NetApp, and Pure Storage have pivoted toward AI infrastructure over the past 18 months, Weka and VAST are "the true AI-native data companies."
Beyond caching, Weka is targeting operational bottlenecks for large-scale AI operators. A metadata-first replication feature allows users to provision new GPU environments and begin running workloads within an hour, compared to previous waits of days or weeks.
McDowell singled out Weka's Augmented Memory Grid as its primary competitive advantage. "Weka continues to have the most technically capable KV cache implementation on the market," he said. "This is critical for AI inference, as it enables a level of GPU efficiency that, without question, saves money on GPUs and memory."
The company is also attempting to distinguish itself through commercial terms rather than just technical specifications. "One flying a little under the radar: Weka is putting its money where its mouth is with its contractual guarantees for its data reduction promises," McDowell said.
