Tomorrow
← Segments

Chips & Compute

AI accelerators, inference silicon and the compute platforms everything else runs on.

18 tools · 69 changes / 30d

AMD Instinct MI355X

AMD

1

High-memory GPU alternative

Version
MI355X
Cost
Enterprise GPU hardware pricing not published
Model
CDNA 4
Apple Silicon Neural Engine

Apple

On-device inference across Mac, iPhone and iPad.

Version
M5
Cost
Bundled with hardware
Model
Apple Foundation Models
AWS Trainium2

AWS

5

Cloud-native training silicon

Version
Trainium2
Cost
Pricing not published; available via AWS EC2 instance pricing.
Model
Neuron compiler
Broadcom custom accelerators

Broadcom

Co-designed XPUs behind most hyperscaler silicon.

Version
2025
Cost
Enterprise supply agreements
Model
powers Google TPU
Cerebras WSE-3

Cerebras

9

Wafer-scale inference and training

Version
WSE-3
Cost
Free tier / developer tier pricing available (contact/partner integrations)
Model
Codex-Spark
d-Matrix Corsair

d-Matrix

Digital in-memory compute for inference efficiency.

Version
Corsair
Cost
System sales
Model
n/a
Etched Sohu

Etched

Transformer-only ASIC betting on architecture lock-in.

Version
Sohu
Cost
System sales
Model
transformer-only
FuriosaAI RNGD

FuriosaAIAsia

Korean inference silicon with strong perf-per-watt.

Version
RNGD
Cost
System sales
Model
n/a
Google TPU Ironwood

Google

12

Inference-first TPU pods

Version
Ironwood (7th generation)
Cost
from $5.4/unit
Model
Gemini
Graphcore IPU

SoftBankEU

IPU accelerators now inside SoftBank.

Version
Bow IPU
Cost
System sales
Model
n/a
Groq LPU

Groq

4

Deterministic low-latency inference

Version
LPU
Cost
Pricing not published. Contact sales or check account console for usage-based rates.
Model
canopylabs/orpheus-arabic-saudi
Huawei Ascend 910C

HuaweiChina

3

China domestic AI accelerator

Version
Ascend 910C
Cost
Pricing not public
Model
Ascend 910C
Intel Gaudi 3

Intel

5

Cost-focused training and inference accelerator.

Version
1.24.1
Cost
Pricing not published. Available via Intel Tiber AI Cloud or OEM platforms.
Model
Accelerator
Lightmatter

Lightmatter

Photonic interconnect for scaling accelerator fabrics.

Version
Passage M1000
Cost
System sales
Model
n/a
NVIDIA Blackwell

NVIDIA

9

GB200/B200 rack-scale AI systems

Version
Blackwell
Cost
Pricing is not publicly listed (enterprise hardware architecture/system)
Model
Blackwell
Qualcomm AI200

Qualcomm

Rack-scale inference accelerators entering the datacentre.

Version
AI200
Cost
System sales
Model
n/a
SambaNova SN40L

SambaNova

5

Dataflow chips for model serving

Version
SN40L
Cost
Free tier available. Paid plans include Developer (Pay-as-you-go) and Enterprise (Subscription-based).
Model
DeepSeek-V3.1
Tenstorrent Blackhole

Tenstorrent

16

Open RISC-V AI hardware

Version
Blackhole
Cost
from $999/unit
Model
Tensix cores

Recent changes in Chips & Compute

  • version

    Google TPU Ironwood — Google TPU Ironwood moved to Ironwood (7th generation)

    7th-generation (Ironwood)Ironwood (7th generation)

    source
  • version

    NVIDIA Blackwell — NVIDIA Blackwell moved to Blackwell

    GB200 NVL72Blackwell

    source
  • version

    Intel Gaudi 3 — Intel Gaudi 3 moved to 1.24.1

    1.24.01.24.1

    source
  • version

    Tenstorrent Blackhole — Tenstorrent Blackhole moved to Blackhole

    v0.79.0-dev20260913Blackhole

    source
  • capability

    Tenstorrent Blackhole — New capabilities: Liquid-cooled option, High-level Python API (TT-NN)

    104 tracked106 tracked · +Liquid-cooled option, High-level Python API (TT-NN)

    source
  • model

    SambaNova SN40L — SambaNova SN40L added to its model stack

    DeepSeek-V3.1, Llama / DeepSeek open weights, Llama 3.1, Llama 3.1 (8B, 70B, 405B), Llama 3.1 405B, Llama 3.1 70B, Llama 3.1 8B, MiniMax M2.7, RDU dataflow, SN40L RDU, SambaRack SN40-16, gpt-oss-120bDeepSeek-V3.1, Llama / DeepSeek open weights, Llama 3.1, Llama 3.1 (8B, 70B, 405B), Llama 3.1 405B, Llama 3.1 70B, Llama 3.1 8B, MiniMax M2.7, RDU dataflow, SN40L RDU, SambaRack SN40-16, gpt-oss-120b

    source
  • capability

    Groq LPU — New capabilities: Orpheus OCR and Image Recognition

    31 tracked32 tracked · +Orpheus OCR and Image Recognition

    source
  • version

    AWS Trainium2 — AWS Trainium2 moved to Trainium2

    AWS Trainium2Trainium2

    source
  • version

    AMD Instinct MI355X — AMD Instinct MI355X moved to MI355X

    ROCm 10.0.0MI355X

    source
  • version

    NVIDIA Blackwell — NVIDIA Blackwell moved to GB200 NVL72

    BlackwellGB200 NVL72

    source
  • capability

    NVIDIA Blackwell — New capabilities: Liquid Cooled

    55 tracked56 tracked · +Liquid Cooled

    source
  • capability

    Cerebras WSE-3 — New capabilities: On-Premise Deployment, Cloud API

    96 tracked98 tracked · +On-Premise Deployment, Cloud API

    source
  • version

    Intel Gaudi 3 — Intel Gaudi 3 moved to 1.24.0

    1.24.11.24.0

    source
  • version

    Tenstorrent Blackhole — Tenstorrent Blackhole moved to v0.79.0-dev20260913

    Blackholev0.79.0-dev20260913

    source
  • capability

    Tenstorrent Blackhole — New capabilities: Enterprise-Class AI Server

    103 tracked104 tracked · +Enterprise-Class AI Server

    source
  • version

    AWS Trainium2 — AWS Trainium2 moved to AWS Trainium2

    Trainium2AWS Trainium2

    source
  • version

    Google TPU Ironwood — Google TPU Ironwood moved to 7th-generation (Ironwood)

    7th-generation7th-generation (Ironwood)

    source
  • version

    NVIDIA Blackwell — NVIDIA Blackwell moved to Blackwell

    GB200 NVL72Blackwell

    source
  • pricing

    Cerebras WSE-3 — Entry price decreased to $0/mo

    $10$0

    source
  • capability

    Cerebras WSE-3 — New capabilities: OpenAI API compatible, Custom kernels

    94 tracked96 tracked · +OpenAI API compatible, Custom kernels

    source
  • version

    Tenstorrent Blackhole — Tenstorrent Blackhole moved to Blackhole

    v0.79.0-dev20260910Blackhole

    source
  • capability

    Tenstorrent Blackhole — New capabilities: Servers, Liquid Cooled

    101 tracked103 tracked · +Servers, Liquid Cooled

    source
  • capability

    Huawei Ascend 910C — New capabilities: AI hardware

    46 tracked47 tracked · +AI hardware

    source
  • model

    SambaNova SN40L — SambaNova SN40L added to its model stack

    DeepSeek-V3.1, Llama / DeepSeek open weights, Llama 3.1, Llama 3.1 (8B, 70B, 405B), Llama 3.1 405B, Llama 3.1 70B, Llama 3.1 8B, MiniMax M2.7, RDU dataflow, SN40L RDU, SambaRack SN40-16, gpt-oss-120bDeepSeek-V3.1, Llama / DeepSeek open weights, Llama 3.1, Llama 3.1 (8B, 70B, 405B), Llama 3.1 405B, Llama 3.1 70B, Llama 3.1 8B, MiniMax M2.7, RDU dataflow, SN40L RDU, SambaRack SN40-16, gpt-oss-120b

    source
  • capability

    Groq LPU — New capabilities: Fast Inference

    30 tracked31 tracked · +Fast Inference

    source