Context Platform Engineering to Reduce Token Anxiety — Val Bercovici and Callan Fox, WEKA
AI Engineer Code 2025 · 23:52
AI data storage and memory infrastructure
WEKA builds data storage and memory infrastructure for enterprises, AI cloud providers, and AI builders running training, inference, and agent workloads. Its NeuralMesh platform lets applications access shared datasets through file and object interfaces across on-premises and cloud environments. Augmented Memory Grid stores inference key-value caches beyond local GPU memory, allowing different hosts to reuse a session’s cached context. WEKA also offers WEKApod storage appliances, announcing a third generation in 2026.
Founded in 2013 by Liran Zvibel, Maor Ben-Dayan, and Omri Palmon, WEKA is led by CEO Zvibel. Its engineering approach distributes both data and metadata across servers instead of assigning metadata to dedicated machines. NeuralMesh’s user-space data path bypasses operating-system kernel storage and networking stacks to reduce processing overhead. Its shared namespace also lets teams ingest data through S3 and process it through POSIX without duplicating datasets between those workflows.
In 2026, the company reported adoption by 30% of the Fortune 50. Zvibel reported nine-digit annual recurring revenue in 2024. That year, WEKA completed a $140 million Series E transaction led by Valor Equity Partners at a $1.6 billion post-money valuation, comprising $100 million for the company and $40 million in secondary sales for veteran employees.
AI Engineer Code 2025 · 23:52
Affiliations reflect their AIE appearances, not necessarily current employment.
Callan Fox examines context-window saturation and repeated prefills alongside token costs, then discusses GPU pressure and memory-tier design, including pooled DRAM and persistent KV-cache infrastructure.
Affiliations reflect each recorded session, not necessarily current employment.