Vik's Newsletter
Subscribe
Sign in
Home
Podcast
Notes
Chat
Semi Doped Podcast
SemiExponent
Archive
About
Latest
Top
Discussions
How Coherent Can Earn Investor Confidence
Bridging The Gap Between Market Skepticism And Industry Reality, And What To Watch For Next.
Aug 16
•
Vikram Sekar
27
1
OCP APAC: Copper vs. Optics Contradictions
Everybody is scrambling for ways to hook up stuff.
Aug 13
•
Vikram Sekar
54
5
[Post-$] Astera Labs is a Switch Company in Connectivity Clothing
ALAB is a copper play with high growth switches, CXL opportunities and optics dreams.
Aug 6
•
Vikram Sekar
27
Astera Labs: From the Tray to the Rack, and What Comes After
PCIe Switches, AWS Trainium, NVLink Fusion, and the UALink Future
Aug 4
•
Vikram Sekar
24
2
July 2026
The AI Trade Burned Down: Let Round 2 Begin
With more sanity this time...
Jul 31
•
Vikram Sekar
37
1
4
HBM is the Ugly Baby: What Now?
Innovations in memory architecture will save the day.
Jul 27
•
Vikram Sekar
39
1
2
Lasers for CPO/NPO: Part 2 – Lumentum’s Technology and Moat
How JDSU/Lumentum engineered their UHP laser, where the competition stands, and how it's all about "is it good enough?"
Jul 22
•
Vikram Sekar
34
1
Lasers for CPO/NPO: Part 1 – The InP DFB Laser
How an InP DFB laser works, and the four physical walls that make high power hard.
Jul 15
•
Vikram Sekar
53
5
Kyber Midplane Design, Making NVL72x2 Work, and All Roads Lead to CPO
An engineering-first read of the supposed Kyber delay, why NVL72x2 was the "odd" duck nobody wanted, how to actually deploy it, and why CPO is the…
Jul 7
•
Vikram Sekar
39
Qualcomm's High Bandwidth Compute and the Packaging Problem It Moved
HBC stacking DRAM on the compute die increases memory bandwidth and skips CoWoS. The packaging problem does not go away, it shifts to 3D.
Jul 1
•
Vikram Sekar
48
1
2
June 2026
🍪 TWiC: Jalapeno, Modular, and Gatekeeping of Frontier Models
Its all about inference these days, by a long shot.
Jun 27
•
Vikram Sekar
23
4
2
What AI Inference Actually Demands From a NAND SSD
KV cache storage is a distinct workload that requires SSDs designed for it. Number of cell levels, block sizes, and data placements all matter, and some…
Jun 25
•
Vikram Sekar
28
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts
