AI Inference Memory-Wall Breaker Volantis Raises $88M Series A - Wowtale
Article excerpt
Highlighted: the sentence this signal was extracted from
Money and M&A are both pouring into optical interconnects at once. Ayar Labs has grown its 2026 funding total to $650 million, and NVIDIA poured $6.5 billion into photonics in just three months . Now Volantis has joined the pack with an $88 million Series A. The problem Volantis is trying to solve is the AI "memory wall." Running giant AI models fast requires two things at once: enough memory capacity to hold the entire model, and enough memory bandwidth to keep feeding that data to the compute chip without a bottleneck. Until now, those two have traded off against each other. Chips built around on-chip SRAM get high bandwidth but limited capacity; GPUs and TPUs built around HBM get high capacity but comparatively low bandwidth. Volantis says it can break that trade-off with optics. Instead of keeping lasers external to the chip, the company built custom, integrated micro-VCSELs directly into its silicon, and it ditched traditional optical fiber for a new structure that pushes bandwidth density higher. Because its design pools memory across chips, it can lean on cheaper memory while still raising throughput, which the company says drives cost per token down too. The goal is to run models with more than 10 trillion parameters at up to 10,000 tokens per second per user - fast enough, the company says, to turn a 30-minute coding-agent task into a 30-second one. Founder and CEO...
Keep reading with a free account
The rest of this article, and every signal for Ayar Labs, is in your free account.
