[HN Gopher] AWS Trainium3 Deep Dive - A Potential Challenger App...
       ___________________________________________________________________
        
       AWS Trainium3 Deep Dive - A Potential Challenger Approaching
        
       Author : Symmetry
       Score  : 58 points
       Date   : 2025-12-04 19:19 UTC (5 days ago)
        
 (HTM) web link (newsletter.semianalysis.com)
 (TXT) w3m dump (newsletter.semianalysis.com)
        
       | klysm wrote:
       | This won't materialize into a legitimate threat on the NVIDIA/TPU
       | landscape without enormous software investment. That's why NVIDIA
       | won in the first place. This requires executives to see past the
       | hardware and make riskier investments and we will see if this
       | actually materializes under AWS management or not.
        
         | stogot wrote:
         | This is addressed in the article.
         | 
         | > In fact, they are conducting a massive, multi-phase shift in
         | software strategy. Phase 1 is releasing and open sourcing a new
         | native PyTorch backend. They will also be open sourcing the
         | compiler for their kernel language called "NKI" (Neuron Kernal
         | Interface) and their kernel and communication libraries matmul
         | and ML ops (analogous to NCCL, cuBLAS, cuDNN, Aten Ops). Phase
         | 2 consists of open sourcing their XLA graph compiler and JAX
         | software stack.
         | 
         | > By open sourcing most of their software stack, AWS will help
         | broaden adoption and kick-start an open developer ecosystem. We
         | believe the CUDA Moat isn't constructed by the Nvidia engineers
         | that built the castle, but by the millions of external
         | developers that dig the moat around that castle by contributing
         | to the CUDA ecosystem. AWS has internalized this and is
         | pursuing the exact same strategy.
        
           | coredog64 wrote:
           | I wish AWS all the best, but I will say that their developer-
           | facing software doesn't have the best track record. Munger-
           | esque "incentive defines the outcome" and all that, but I
           | don't think they're well positioned to collect actionable
           | insight from open GitHub repos.
        
           | almostgotcaught wrote:
           | This isn't an "enormous software investment", this is table
           | stakes which lose out heads up against Nvidia. See AMD.
        
         | ivape wrote:
         | In terms of their seriousness, word on the street is they are
         | moving from custom chips they could be getting from Marvell
         | over to some company I've never heard of it. So, they are
         | making decisions that appear serious in this direction:
         | 
         |  _With Alchip, Amazon is working on "more economical design,
         | foundry and backend support" for its upcoming chip programs,
         | according to Acree._
         | 
         | https://www.morningstar.com/news/marketwatch/20251208112/mar...
        
         | mrlongroots wrote:
         | Hyperscalers do not need to achieve parity with Nvidia. There's
         | a (let's say) 50% headroom in terms of profit margins, and
         | plenty of headroom in terms of the complexity custom chip
         | efforts need to implement: they don't need the complexity or
         | generality of Nvidia's chips. If a simple architecture allows
         | them to do inference at 50% of the TCO and 1/5th the complexity
         | and reduce their Nvidia bill by 70% that's a solid win. I'm
         | being fast and loose with numbers and Trainium clearly seems to
         | have ambitions beyond inference, but given the hundreds of
         | billions each cloud vendor is investing in the AI buildout, a
         | couple billion on IP that you will own afterwards is a no
         | brainer. Nvidia has good products and a solid head start but
         | they're not unassailable or anything.
        
         | bri3d wrote:
         | IMO the value of COTS software stack compatibility is becoming
         | overstated: academics, small research groups, hobbyists, and
         | some enterprises will rely on commodity software stacks working
         | well out of the box, but large pure/"frontier"-AI inference-
         | and-training companies are already hand optimizing things
         | anyway and a lot of less dedicated enterprise customers are
         | happy to use provided engines (like Bedrock) and operate at
         | only the higher level.
         | 
         | I do think AWS need to improve their software to capture more
         | downmarket traction, but my understanding is that even
         | Trainium2 with virtually no public support was financially
         | successful for Anthropic as well as for scaling AWS Bedrock
         | workloads.
         | 
         | Ease of optimization at the architecture level is what matters
         | at the bleeding edge; a pure-AI organization will have teams of
         | optimization and compiler engineers who will be mining for
         | tricks to optimize the hardware.
        
         | trueismywork wrote:
         | And data hosting rules
        
         | willahmad wrote:
         | Don't underestimate AWS.
         | 
         | AWS can make it seamless, so you can run open source models on
         | their hardware.
         | 
         | See their ARM based instances, you rarely notice you are
         | running on ARM, when using Lambda, k8s, fargate and others
        
       | jauntywundrkind wrote:
       | > _they will go with three different scale-up switch solutions
       | over the lifecycle of Trainium3, starting with a 160 lane, 20
       | port PCIe switch for fast time to market due to the limited
       | availability today of high lane & port count PCIe switches, later
       | switching to 320 Lane PCIe switches and ultimately a larger
       | UALink to pivot towards best performance._
       | 
       | It doesn't have a lot of ports and certainly not enough NTB to be
       | useful as a switch, but man, wild to me than an AMD Epyc core has
       | 128 lanes of PCIe and that switch chips are struggling to match
       | even a basic server's worth of net bandwidth.
        
       | artur44 wrote:
       | The hardware story is interesting, but I'm curious how much of
       | the real-world adoption will depend on the maturity of the
       | compiler stack. Trainium2 already showed that good silicon isn't
       | enough if the software layer lags behind.
       | 
       | If AWS really delivers on open-sourcing more of the toolchain,
       | that could be a much bigger signal for adoption than raw specs
       | alone.
        
       | thecopy wrote:
       | I have seen links to semianalysis before, i just am scared of the
       | length of this content. Is anyone reading these start to finish?
       | Why?
        
         | esafak wrote:
         | I think they're for investors.
        
         | mlmonkey wrote:
         | Ask Gemini to summarize it? Or maybe NotebookLM to turn it into
         | a 10-minute podcast? :-)
        
         | hobo_mark wrote:
         | I don't read them, but I listen to them on my commute (with a
         | saas I made).
        
           | ijidak wrote:
           | What is the saas? I've been looking for something like this.
        
         | mNovak wrote:
         | I do, just for fun. It's become sort of a hobby, learning more
         | depth/detail behind the current AI arms race. It certainly cuts
         | through the shallow takes that get thrown around constantly.
        
       | cmiles8 wrote:
       | Chips without an ecosystem and software (CUDA) does not a serious
       | challenger make. Thats where Amazon has, and continues to,
       | struggle.
        
       | t1234s wrote:
       | What does this mean for a company like Coreweave?
        
         | Analemma_ wrote:
         | CoreWeave already had to issue more convertible debt earlier
         | this week after a big dip in their share price. It seems like
         | the market suspects the end is near.
        
       ___________________________________________________________________
       (page generated 2025-12-09 23:01 UTC)