MaxLinear, Inc., a provider of high‑performance connectivity and data movement solutions, showcased Panther V, the latest generation of its storage accelerator platform, at Dell Technologies World 2026.Panther V addresses one of the most critical constraints emerging in large-scale AI inference data centers: the cost, latency, and inefficiency of data movement across storage, memory, and compute. As AI workloads transition from experimental pilots to persistent, production-scale inference, system performance is increasingly constrained by how efficiently data is staged, prepared, and activated for inference.

Optimized for AI Inference and Time‑to‑First‑Token (TTFT)
Panther V reduces end‑to‑end latency and improves responsiveness and throughput for modern AI inference by tightly coupling CPU, accelerator, and GPU resources to keep data moving efficiently through the system. Inline execution of data transformation, compression, encryption, and integrity operations eliminates unnecessary CPU involvement and memory round‑trips, reducing GPU idle time and accelerating time‑to‑first‑token while freeing up host CPUs to focus on model execution and coordination.

As agentic inference grows and workloads become increasingly latency‑sensitive, Panther‑based accelerators enable the same CPUs and GPUs to support many more simultaneous inference agents. This improves utilization, scalability, and overall system efficiency for interactive, real‑time AI services.

Built for Today’s Inference‑Dominated Workloads
As AI inference becomes the primary driver of production AI deployments, Panther V is purpose‑built to support the most demanding inference scenarios, including:

• Low‑latency inference, where fast TTFT is essential for conversational AI and real‑time applications
• Retrieval‑Augmented Generation (RAG), accelerating document retrieval and preparation from enterprise data stores
• KV‑cache‑intensive inference, enabling reuse of pre-fill‑stage key‑value data across users and agents without impacting GPU hot‑path performance

By accelerating compression, decompression, encryption, and integrity validation in silicon, Panther V enables smaller, verified data to move faster through storage, memory, and network fabrics, which improves inference economics without increasing power or infrastructure cost.

Key Panther V Capabilities
Panther V combines scalable performance, deep CPU offload, and advanced security and integrity acceleration to enable efficient, high‑concurrency AI inference at scale.

• Scalable Performance: Supports system architectures exceeding 6Tbps, delivering up to 450Gbps per accelerator
• CPU Offload: Dedicated hardware engines perform single‑pass compression, encryption, and checksum processing entirely in silicon, avoiding multiple PCIe pass‑throughs
• Advanced Accelerations: GZIP, Zlib, Deflate, XP10, AES encryption (ECB, CBC, CTR, XTS, GCM), and SHA‑1/2 hashing and checksums
• Data Integrity: Real‑time, end‑to‑end verification with CRC validation and NVMe T10 DIF/DIX support
• Software Flexibility: SDK supporting synchronous and asynchronous APIs, kernel and user space, NUMA‑aware queues, and peer‑to‑peer DMA
• ZFlush for OpenZFS: A hardware‑accelerated OpenZFS implementation that integrates seamlessly with Panther V to improve file‑system performance
• Industry‑Standard Form Factors: Available in PCIe and OCP NIC 3.0 configurations

Powering the Economics of AI Data Centers
The AI inference market is expanding rapidly, with sustained double‑digit growth projected into the early 2030s. As inference becomes persistent and monetized, infrastructure buyers are prioritizing system efficiency, power optimization, and time‑to‑value over peak compute metrics alone. Panther V enables data center operators to scale inference concurrency, support longer context windows, and deliver faster user experiences without linear increases in cost or power consumption.

“AI inference is rapidly becoming a real‑time, revenue‑generating workload, and data movement, not compute, is emerging as the primary system bottleneck,” said Vikas Choudhary, SVP & GM, connectivity and storage business, MaxLinear. “By accelerating faster node bring‑up, growing context sizes, and KV‑cache compression, Panther V enables more efficient and low latency inference pipelines along with scalable AI inference economics. We believe that the size of the serviceable market for purpose-built silicon accelerator solutions, such as Panther V, is approximately $5 billion.”