Fabricated Knowledge
@fabknowledge
Simplifying the world of semiconductor investing in the age of AI. Part of the @semianalysis_ gang. all views are personal
906 Following 34.2K Followers
Incredible that I've been writing this and Tao just... tweeted it out
1/ we partnered with @stripe to add convenient payments to Muse with Stripe Link. Muse can make purchases for you; and it asks for your approval before spending your money
astra is kinda undercooked posttraining wise
AI is the new Oil.
Before there was "AI psychosis" one of the most popular phrases during the early oil boom was "Oil on the brain" referring to people who got swept away obsessing about oil
Show more
Gemini 3.8 Flash and Muse Spark 1.3 are two of the most clearly benchmaxxed models we've seen yet. Despite being comparable to both GPT-6 and Fable 5.1 on Terminal Bench 2.1, their Terminal Bench 4.0 performance is markedly worse. (1/5)🧵
Show more
RT @SemiAnalysis_: TPU Inference Externalization Full Steam Ahead
InferenceX, Up to 50% Better Performance per Dollar
Rapid Externalization…
@fabknowledge says this is a honeypot, anyone want to curl for us so I can win a bet?
The REAL bottleneck bottleneck, the bottleneck that no one knows about or saw coming, the neck that is bottled beyond bottleneck, the TRUE bottleneck upstream of all other bottlenecks, the bottleneck’s bottleneck, bottleneck bottleneck bottleneck bottlebeck bottlewrek bottlewept
Show more
gotta go fast
is it memory-bound or compute-bound? you have to call it
Nvidia denies the pause:
"The new business model we introduced in July that opens up compute access to the fast-growing AI ecosystem is still in place and continues to evolve due to high demand." $NVDA
Show more
And there we go
*NVIDIA PAUSES REVENUE-SHARING DEALS WITH AI COMPANIES: WSJ
*NVIDIA STEPPED BACK FROM REVENUE-SHARING PROGRAM LAST WEEK: WSJ
SCOOP: Nvidia paused some deals in a new financing initiative that offered credit support to AI cloud providers in exchange for a share of revenue.
- Nvidia employees have expressed concern the program could draw antitrust scrutiny
- There are sensitivities around the extent to which the chip giant can dictate how its customers do business
details in @WSJ:
Show more
NVDA is going to break all time highs.
For the NVDA funding shorts, they’ll cover quickly. For the underweight long only managers, that’ll take a bit longer.
Announcing the expansion of NVIDIA NVLink Fusion with NVHBM, a next-generation high-bandwidth memory technology that brings higher memory performance and efficiency to XPUs.
Amazon's @AnnapurnaLabs will be the first to work with us on NVHBM, combining @awscloud custom silicon with our memory technology and the NVLink scale-up architecture to enhance performance and efficiency for AI workloads.
Learn how we're helping hyperscalers and AI innovators build the next generation of AI infrastructure:
Show more
$NVDA is guiding on its call for revenue to grow ~70% in FY28 (ends Jan. '28). Consensus is only at 44.5%. (transcript via Quartr)
$NVDA sharing a bunch of details about its purchase commitments and guarantees in its CFO commentary. Among them: Supply/capacity commitments rose by $160B Q/Q to $279B, "primarily" due to memory.
Show more
Getting my kit together for Jensen to talk to me like I'm 5 years old for an hour. This is it, earnings super bowl! Pumped! $NVDA
Sources: Moonshot AI is in early talks over revenue-sharing agreements with Microsoft, Amazon, and Google to host Kimi K3, and is seeking up to a 30% share (Reuters)
(Visit Techmeme dot com for the link and full context!)
Show more
Jalapeno beats VR200 in A0 silicon with a vibe-ported, non-specdec implementation of DeepSeek 🌶️ 🌶️
And with much faster program execution than VR200, and with a B0 update landing imminently
Congrats to the Jalapeno team!!!
Show more
One of the most interesting parts of this is how the Jalapeno team (allegedly, and I think this is reasonable) used AI to accelerate the bringup of the chip on frontier(ish) models
They can literally just profile the runs and hand it off to Codex for infinite iteration
This simply would not have been possible to do so efficiently 2 yrs ago
Show more
