Google TPU

What hackers pick instead

• Taalas HC1Model-in-silicon chip praised for very fast, low-power inference; the main caveat is that the model cannot be updated without replacing the chip.59recommend
• Taalas hardwired LLM chipPeople recommend it for very fast LLM inference from a model etched into silicon. Critics note the chip is locked to one model and updates are limited.45recommend
• Cerebras wafer-scale chipWafer-scale chip that people say runs LLM inference far faster than GPUs. Critics agree it is costly and struggles with large models.29recommend
• Cerebras CS-4People recommend it for very fast AI inference, above 1,000 tokens per second on large models. Critics say it is scarce, hard to access, and not affordable for individuals.12recommend
• Groq LPURecommended for fast, efficient AI inference, using SRAM instead of HBM. Critics point to small per-chip SRAM and hard deployment requirements.9recommend

Every quote

The 40 most detailed of 102 comments. Each links to the original.

1. even if google TPUs aren't competitive in a pure performance sense they are surely competitive from a pricing perspective

   Mixed · Jul 2026 · in “Grok 4.5”

2. TPUs are the economically-conscious inference choice if you've already separated your training/inference workflows.

   Recommends · Apr 2026 · in “Our eighth generation TPUs: two chips for the agentic era”

3. Google’s TPU is going to be a massive advantage that’s almost impossible for Amazon or Microsoft to replace

   Recommends · Apr 2026 · in “Google controls ~25% of global AI compute, with ~3.8M TPUs and 1.3M GPUs”

4. Google is crushing them on inference. By TPUv9, they could be 4x more energy efficient and cheaper overall

   Recommends · Feb 2026 · in “GPT‑5.3‑Codex‑Spark”

5. They are using that hardware to wipe the floor with everyone if you look at the price per million tokens.

   Recommends · Nov 2025 · in “TPUs vs. GPUs and why Google is positioned to win AI race in the long term”

6. Having the TPU (and models purpose built for TPU) allows for best cost in serving at hyperscale.

   Recommends · Feb 2026 · in “Nvidia and OpenAI abandon unfinished $100B deal in favour of $30B investment”

7. It can't compete with NVidia's best offerings for sure but it gives them a path to independence.

   Mixed · Oct 2026 · in “Anthropic to receive high 'junk' rating from credit agencies following IPO”

8. Was a solid bet on energy efficient dense matrix multiplication and has stood the test of time.

   Recommends · Jun 2026 · in “OpenAI unveils its first custom chip, built by Broadcom”

9. Google's TPUs are much more efficient for inference because they're mostly tensor cores by area

   Recommends · Apr 2026 · in “Are the Costs of AI Agents Also Rising Exponentially? (2025)”

10. they've had some initial success with TPUs which could be a major differentiator in the future

    Mixed · Jun 2026 · in “Noam Shazeer Joins OpenAI”

11. Owning your hardware and your entire stack is huge, especially these days with so much demand.

    Recommends · Apr 2026 · in “Our eighth generation TPUs: two chips for the agentic era”

12. until they can match that interconnect scale, they simply don't compete in this weight class

    Recommends · Nov 2025 · in “TPUs vs. GPUs and why Google is positioned to win AI race in the long term”

13. Nobody wanted to mess with an immature crashing compiler and very steep performance cliffs

    Mixed · Apr 2026 · in “Microsoft to Stop Sharing Revenue with Main AI Partner OpenAI”

14. their latest-gen TPUs win on compute-per-Joule for LLM-type workloads by quite a margin

    Recommends · Aug 2026 · in “Nvidia's Risky Business”

15. They are more efficient than nvidia GPUs, a huge amount of inference is done with them

    Recommends · Jan 2026 · in “Furiosa: 3.5x efficiency over H100s”

16. the most bang-for-your-buck (on hyperscalers) ML infrastructure by far (because TPUs)

    Recommends · Oct 2026 · in “Google Search is now less than half of Alphabet’s revenue”

17. last we tried, so much of our kernels needs rework that it’s not worth the effort

    Mixed · Apr 2026 · in “Google to invest up to $40B in Anthropic in cash and compute”

18. GOOGLE has more experience with datacenter clustered TPUs and tensor acceleration

    Recommends · Nov 2025 · in “Nvidia says its GPUs are a 'generation ahead' of Google's AI chips”

19. Google is already using their own TPUs at scale for both training and inference.

    Mixed · Nov 2025 · in “HipKittens: Fast and furious AMD kernels”

20. the TPU is a pretty big accomplishment that successfully competes for inference

    Recommends · Jun 2026 · in “Noam Shazeer is joining OpenAI”

21. TPUs can be more efficient, but are quite difficult to program for efficiently

    Mixed · Jan 2026 · in “Furiosa: 3.5x efficiency over H100s”

22. TPUs don't have enough memory either, but they have really great interconnects

    Mixed · Feb 2026 · in “GPT‑5.3‑Codex‑Spark”

23. They get to make pickaxes that are made exactly for the mine they are working.

    Recommends · Nov 2025 · in “TPUs vs. GPUs and why Google is positioned to win AI race in the long term”

24. they were first and have far more experience operating these systems at scale

    Recommends · Feb 2026 · in “Google goes from laggard to leader, pulls ahead of OpenAI with stellar AI growth”

25. I could imagine Google winning merely because their chips are more efficient.

    Recommends · Apr 2026 · in “The eighth-generation TPU: An architecture deep dive”

26. Google has a much better grasp on their stack than people building on NVidia

    Recommends · Apr 2026 · in “Our eighth generation TPUs: two chips for the agentic era”

27. If 'this is faster than a CPU for massively parallel AI training' is the best you can say about it, that's not very impressive.

    Criticizes · Nov 2025 · in “Gemini 3 Pro Model Card [pdf]”

28. Google had nigh on a decade to boost TPU production, and they're still somehow behind the curve.

    Criticizes · Mar 2026 · in “Apple discontinues the Mac Pro with no plans for future hardware”

29. I often don't prefer the TPU either because the documentation for the AI isn't made for it

    Criticizes · Nov 2025 · in “TPUs vs. GPUs and why Google is positioned to win AI race in the long term”

30. Compiler bugs in the tooling around the platform caused most of their quality issues

    Criticizes · Nov 2025 · in “OpenAI probably can't make ends meet. That's where you come in”

31. TPU has often turned out to be over-specialized and Nvidia are pointing that out.

    Criticizes · Nov 2025 · in “TPUs vs. GPUs and why Google is positioned to win AI race in the long term”

32. tell me this is competitive with Nvidia/Oracle/any neocloud with a straight face

    Criticizes · Nov 2025 · in “Michael Burry a.k.a. "Big Short",discloses $1.1B bet against Nvidia&Palantir”

33. Google's TPU and Groq's LPU are the only real, commercially viable ways to provide all the compute required for the inference people want.

    Recommends · Jun 2026 · in “Apple reveals new AI architecture built around Google Gemini models”

34. Counting those against the other available datacenter hardware used by others, it puts google at a huge advantage

    Recommends · May 2026 · in “Gemini 3.5 Flash”

35. It was just in the past couple years that it really became worthwhile to use TPU's if you're on GCP

    Mixed · Dec 2025 · in “Amazon launches Trainium3”

36. They've already done this with their TPUs but need to be able to compete with NVidia offerings.

    Recommends · Aug 2026 · in “Google DeepMind CEO Demis Hassabis is stepping down”

37. their own hardware designed specifically for AI, and designed to scale better to larger models

    Recommends · Apr 2026 · in “Sam Altman's response to Molotov cocktail incident”

38. If they move on to competing on having to do inference the cheepest then the TPUs will shine.

    Recommends · Nov 2025 · in “TPUs vs. GPUs and why Google is positioned to win AI race in the long term”

39. There is an increasingly more competition on Nvidia itself whether its from Google TPU's

    Mixed · Jun 2026 · in “LongCat-2.0, a large-scale MoE model with 1.6T total and 48B Active”

40. Google has shown with their TPu chips that there is great potential for stuff like this

    Recommends · Jun 2026 · in “SpaceX plans $55B investment to make A.I. chips”