Subscribe

Hardware

Chips, accelerators and the physical substrate AI runs on.

  • Samsung Exynos 2600

    Samsung’s first 2nm phone chip, which it says more than doubles generative AI performance over the Exynos 2500.

  • Samsung Exynos 2400

    Samsung’s 4nm flagship phone chip, with an NPU it rates 14.7x faster at AI than the Exynos 2200.

  • Samsung Exynos 2500

    Samsung’s first 3nm GAA phone chip, with an NPU rated at up to 59 TOPS.

  • Microsoft Maia 100

    Microsoft’s first in-house AI accelerator, built for its own Azure AI services.

  • Microsoft Maia 200

    Moves to 3nm and HBM3e and is the first Maia with a published FP4 and FP8 figure.

  • Microsoft Maia 300

    Microsoft’s third Maia moves to 2nm-class chiplets and HBM4 for multi-gigawatt Azure deployments from 2027.

  • AWS Trainium2

    Up to 4x faster training and 3x the memory of first-generation Trainium.

  • AWS Trainium3

    AWS’s first 3nm chip, with up to 2x the MXFP8 throughput of Trainium2 per chip.

  • AWS Inferentia2

    Moves from DDR4 to HBM, with up to 4x the throughput and 10x lower latency than Inferentia.

  • AWS Trainium

    AWS’s first in-house training chip, sold through EC2 Trn1 instances of 16 chips.