Nvidia Buys Into Intel, Apple’s Modem, Corporate Jobs, and the Billion-Dollar Cable Race
A few articles I really enjoyed, round up of all my content from September, and a short quiz.
Welcome to the weekday free-edition. Today is a collection of articles I found interesting, as well content published in this newsletter this past month. If you’re new, start here!
Note: No Sunday deep-dive this week because I’m on a beach drinking Feni and spending some quality time with family. Regular programming will continue next week.
Nvidia’s 5% stake in Intel: Who else wants a piece?
The biggest news this week by far is the Nvidia and Intel partnership. In addition to the US Government owning a 10% stake in Intel, Nvidia got a 5% piece for $5B. 85% is still available. Who else wants in?
While rivals at one point, this announcement brings Nvidia’s GPUs to Intel’s x86 based CPUs. While I am no gamer, the ability to integrate Team Green’s GPUs alongside x86 chips in a single SoC via NVLink fusion unlocks a new realm of performance that is really compelling. This makes Intel+Nvidia a strong contender in the AI PC game, putting a new challenger on the arena in which AMD and Qualcomm are playing. This also puts price pressure on MediaTek’s processors that were intended for use in AI PCs with Nvidia GPUs.
This announcement also has big implications for the AI server market. Providing customers with access to x86-based AI compute is good diversification away from the Arm-based servers in deployment today.
discusses the implications of this partnership with a broad lens; I recommend reading his take on this.
Apple Modems and Qualcomm
I’m not going to add my take to this conversation for obvious reasons, and so that legal teams don’t come after me with pitchforks. But the new iPhone has an Apple modem which is functional, and people seem happy with the performance. It lacks a lot of the features Qualcomm modems provide, but the question of whether they are really necessary is still something a lot of people are debating.
as usual has an entertaining and highly informative piece on Qualcomm, which I’m just going to leave below for your enjoyment.
Active Electrical Cables Are a Billion Dollar Industry
We’ve spoken quite a bit about interconnects this month on this newsletter, and for good reason too: cables are a big deal right now. Credo (CRDO 0.00%↑) has seen its stock skyrocket because of their vertically integrated DSP-enabled copper interconnect that extends reach up to 7 meters. A lot of other players also make DSP chips that implement forward error correction (FEC), equalization, and amplification such as Marvell, Astera Labs, Point2 Technologies, and Broadcom, to name a few. But Credo tapes out it own silicon, and ships the actual cable too. Today, they are providers to every major hyperscaler: Meta, Amazon, Microsoft and xAI.
, in his casual, almost vocal style of writing, ties the tech and business of active electrical cables nicely in the post below.
Corporate Employment: Hiring and Pretenses
The recent H1B immigration changes puts a lot of jobs in jeopardy when companies have to pay $100,000 as visa fees, and changes the hiring equation significantly. From time to time, I get emails asking my thoughts about hiring patterns in general from young readers looking to get into the industry. In most cases, I am lost for words. The landscape is so dynamic and different from what I went through after graduating from college that I can’t pretend to understand what people are going through today.
On the other hand, I read essays like the one below from
and find myself agreeing with many parts the post. While there are plenty of capable engineers who truly move the needle of technology in an admirable fashion, a corporate structure does have a lot of roles that can be done away with. Many people constantly live in this pretense; always appearing busy, and doing busywork without deep meaning. Everyone knows they’re playing the same corporate game, but no one says it to each other.
This is my cognitive dissonance: hiring for new grads is terrible at the moment, but what are they all fighting for? Is it all worth it when they feel the corporate pretense? I’ve written about this tangentially too as “The Game of Life: MegaCorp Edition.”
Content this Month on Vik’s Newsletter
In case you’ve missed it, here is all the content published in the month of September.
Free Posts
Why Documentation is the Missing Link Between AI and Chip Design
My 100th post on this newsletter, more about writing and my YouTube channel
Paid Posts
YouTube Videos
Short Quiz
Here are 5 short quiz questions, if you’re down to have some fun and play along! Answers at the end.
The Transformer architecture started with a landmark paper titled “Attention is All You Need” focusing on a notoriously complex natural language task: translation, but was later actually capable of generating meaningful language, with capabilities way beyond simple translation. English is one of the language choices in the translation exercise. What was the other language?
Proposed in 1971 by Leon Chua as the “missing fourth element” of circuit theory, this two-terminal device retains a memory of its resistance state. Touted as the future of in-memory and analog AI compute — especially after HP’s 2008 announcement that it had been built in practice — it has remained perpetually “just around the corner.” What is it?
An MIT-trained computer architect who pioneered stream processing and parallel computing, he co-founded Velio Communications and Stream Processors Inc. before becoming chief scientist at the GPU giant driving today’s AI boom. Beyond academia and silicon, he once survived a dramatic crash landing while piloting his own plane. Who is this?
It doesn’t host a top-five foundry (for now), yet its chemical industry quietly underpins advanced lithography. In fact, over 80% of the world’s ArF immersion photoresists originate here — a reminder that chipmaking power isn’t only about fabs. Name the country.
In Transformer decoders, attention may grab the headlines, but the real parameter hog lies in the block that sits between attention and output. Expanding the hidden state by multiple factors before squeezing it back down, this “one-way only” network often holds the majority of a model’s weights. What is it called?
Answers:
German
Memristor
Japan, particularly through companies like JSR Corporation, Tokyo Ohka Kogyo (TOK), Shin-Etsu Chemical, and Sumitomo Chemical.
Feedforward layer (FFN) or Multilayer Perceptron (MLP)
Thanks for playing! 🍻
Somehow, me holding on to a couple of INTC shares hasn't received the same attention as Nvidia's or Softbank's investments 😜.