1. X
  2. SemiAnalysis
Log inSign up
SemiAnalysis
2,807 posts
Image
user avatar
SemiAnalysis
@SemiAnalysis_
semianalysis.com
Joined January 2024
28
Following
141.2K
Followers
AffiliatesAffiliatesRepliesRepliesMediaMedia

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • user avatar
    SemiAnalysis
    @SemiAnalysis_
    13h
    ANTHROPIC WILL BE AN AMD CUSTOMER, ACCORDING TO THE PUBLIC GITHUB OF AMD’S SENIOR DIRECTOR OF AI 🚨🚨 We explain the GitHub code and nuances below👇️ 1/4🧵
    Image
    235K0235K
    user avatar
    SemiAnalysis
    @SemiAnalysis_
    13h
    Replying to @SemiAnalysis_
    Note that Anthropic is still in the evaluation phase, and if AMD doesn’t announce Anthropic at its upcoming Advancing AI conference, that means @AnushElangovan’s FDE team has yet to address all of Anthropic’s concerns regarding software quality, and further improvement will be
    25K025K
    user avatar
    SemiAnalysis
    @SemiAnalysis_
    13h
    One general concern from the ML community is that, at the Kubernetes orchestration layer, AMD has yet to implement sufficient automated CI testing for its first-party Pollara NIC. Kubernetes is the layer used by most inference deployments worldwide. This isn’t because engineering
    Image
    21K021K
  • user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 19
    What's your favorite kind of rack?
    GPU43.6%
    Squat20.6%
    Lamb25.7%
    Bike10.1%
    1,873 votesFinal results
    81K081K
  • user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    Similar to DeepSeek in January 2025, Panicans may think that the AI networking switch TAM will massively shrink because Kimi K3 uses KDA Attention, which reduces KV-transfer networking bandwidth by up to 10x. But the opposite is true, as we explain below. 👇️ 1/8🧵
    Image
    304K0304K
    user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    Replying to @SemiAnalysis_
    Furthermore, with optimizations like incremental KV-cache transfers, prefill only needs to transfer the portions cached by the non-decode instance, so KV transfer does not take up much networking bandwidth relative to WideEP, even before KDA. 7/8🧵
    22K022K
    user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    More efficient attention will further push context lengths from 1M to 5M+, with less context rot. Jevons’ Paradox means that making attention more efficient will lead to wider AI adoption, which will require more networking. 8/8🧵
    21K021K
  • user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    A year ago, the big three was OpenAI, Anthropic, and Google. Things have changed. Moonshot's Kimi K3 sits above Gemini on every composite benchmark, and it's open source in 10 days. New episode: what K3 reveals about frontier margins, model sizes, and who's actually still in
    Image
    00:00
    117K0117K
  • user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    MASSIVE DELAY ALERT TO ORACLE’S STARGATE SITE AND BLOOM ENERGY🚨🚨 Oracle’s Project Jupiter behind-the-meter datacenter project in New Mexico that plans to use Bloom Energy is at risk of a 1-2 year delay due to permitting and pipeline building blockers. (1/8)🧵
    Image
    729K0729K
    user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    Replying to @SemiAnalysis_
    And irrespective of the pipeline: the power generation plan itself still isn't approved. In April, Oracle and BorderPlex Digital Assets announced the switch from gas turbines to up to 2.45 GW of Bloom Energy solid oxide fuel cells — one of the world's largest independent
    35K035K
    user avatar
    SemiAnalysis
    @SemiAnalysis_
    Jul 18
    However, the fuel cell Air Permit application is still pending as of mid-July. The New Mexico Environmental Department ordering a public hearing, but without a date set for it.  Project Jupiter is the most delayed of the Stargate projects. Read more about the real delays to the
    Image
    Stop Saying Half of 2026 US Datacenter Capacity Is Canceled
    From newsletter.semianalysis.com
    34K034K
Advertisement
Advertisement