Axelera's Titania runs FP8 and FP4 math inside memory arrays, betting air-cooled racks can beat GPUs

Axelera Titania Brings Native FP8 and FP4 to In-Memory Compute

Axelera's Titania runs FP8 and FP4 math inside memory arrays, betting air-cooled racks can beat GPUs

Axelera AI's Titania chiplet performs native FP8 and FP4 math inside its digital in-memory compute arrays, with RISC-V cores wrapping every matrix engine. An 8-package, 4U air-cooled server is claimed to deliver 20 PFLOPS of FP8 and 1 TB of LPDDR6 at around 6 kW. Axelera projects up to 7x throughput per watt and 20x better TCO, backed by $300 million in commitments ahead of Q3 2027 production — though all figures come from simulations of a chip that has not taped out.

We don't ask anyone to trust a simulator. The architecture behind Titania already turns 88 percent of its memory bandwidth into tokens on Europa silicon that is shipping today.

More from this day

2026-11-19