MA
What is Google even doing?Matthew Berman7 Aug 2026
Google's own AI chips, now at roughly the eighth generation. The creator cites them as a competitive advantage, since Google doesn't depend on Nvidia GPUs alone.
Google's custom AI accelerators, offered through Google Cloud for training and inference of machine learning models.
Updated EU status checked Last shown
Google TPU at a glance
- What is Google TPU?
- Google TPU is a hardware from Google. Google's custom AI accelerators, offered through Google Cloud for training and inference of machine learning models.
- Who makes Google TPU?
- Google TPU is made by Google.
- Does Google TPU work in the EU?
- Available in the EU. The Cloud TPU pricing page lists EU regions (europe-west4 Amsterdam, europe-west1 Belgium, europe-west2 London) with Trillium, v5e and Ironwood pricing. Newest generations may not be offered in every region, and the eighth-generation 8t is listed as coming soon. Last verified 14 Nov 2026.
- How much does Google TPU cost?
- Paid. Billed per chip-hour in USD. Examples: Ironwood on-demand $12.00 (us-central1) and $13.20 (europe-west2 London); Trillium $2.70 (us-east1) and $2.97 (europe-west4 Amsterdam). Spot, DWS Flex-start/Calendar and 1-/3-year commitments are discounted. New Google Cloud customers get $300 in credits. Licence: Proprietary.
- Where was Google TPU shown?
- In 1 creator video by Matthew Berman. Most recently by Matthew Berman on 7 Aug 2026, at 11:20 in “What is Google even doing?”.
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
About
Tensor Processing Units (TPUs) are custom accelerators designed by Google for machine learning workloads. They are used internally for Gemini and products such as Search, Photos and Maps, and are rentable to customers as Cloud TPU on Google Cloud. They run as part of Google's AI Hypercomputer architecture.
The current portfolio on the official page spans several generations: Trillium (6th gen), Ironwood (7th gen, generally available, 9,216 liquid-cooled chips per pod), and the eighth generation, which includes TPU 8t (large-scale training, listed as coming soon), TPU 8i (post-training and low-latency inference), TPU 8tx (secure and confidential inference) and TPU 8i Max (more compute, SRAM and HBM bandwidth than 8i). Google states clusters can reach up to 1 million chips.
The software stack includes JAX, PyTorch via TorchTPU, OpenXLA, MaxText, Tunix and vLLM, with deployment management via Google Kubernetes Engine. TPUs are accessed as a cloud service billed per chip-hour, with on-demand, Spot, DWS Flex-start/Calendar and 1- or 3-year commitment options. They are not sold as standalone hardware on the pages provided, and availability of specific generations varies by region.
Key features
- Eighth-generation systems: TPU 8t, 8i, 8tx and 8i Max
- Ironwood (7th gen) generally available, 9,216 chips per pod
- Native JAX and PyTorch (TorchTPU) support, plus vLLM inference
- Scales to up to 1 million chips in a single cluster
- Per chip-hour pricing with on-demand, Spot, and committed-use options
- Confidential inference option (TPU 8tx) for regulated workloads
1 moment
As seen on YouTube
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Keep exploring
Similar AI tools
Closest neighbours by meaning, not just by category.
T8
TPU 8i
Hardware · Google
Google's eighth-generation Cloud TPU for post-training and low-latency inference of large MoE models.
Limited in EUHardware & Devices
Apr 2026Paid
T8
TPU 8t
Hardware · Google
Google's eighth-generation TPU for large-scale AI pre-training, scaling to 9,600 chips per superpod.
Limited in EUHardware & Devices
Apr 2026Paid
PS
Project Suncatcher
Hardware · Google
Google research initiative to put TPU-equipped satellites in orbit as a prototype space-based AI data center.
UnverifiedHardware & Devices
Nov 2025
OI
OpenAI inference chip
Hardware · OpenAI
Custom AI inference chip announced by OpenAI and Broadcom, built to serve ChatGPT-style workloads faster and cheaper.
UnverifiedHardware & Devices
Release date n/a
OC
OpenAI custom AI chip
Hardware · OpenAI
OpenAI's in-house AI accelerator chip, developed to lower the cost of training and running its models; not publicly available.
UnverifiedHardware & Devices
Release date n/a
GR
Gemini Robotics 2
Model · Google DeepMind
Google DeepMind family of robot models: a VLA, ER 2 for reasoning and planning, and On-Device 2 for offline use.
UnverifiedRobotics
Release date n/a