Why stacked 3D-DRAM is the right future for AI inference accelerators

An analysis of performance, power, and architecture

Sep 23 • Subbu

40

6

HBM's Bandwidth Roadmap and Who Is Positioned to Win It

How HBM gets faster from HBM3e to HBM5, what comes after, and why the base die decides who wins. A field report from Hot Chips 2026.

Sep 12 • Subbu

17

1

August 2026

Hot Chips 2026 Preview: 3D-DRAM, Rubin, TPUv8, and OpenAI’s Chip

Plus MAIA 200, NVIDIA’s LPU, Rubin & Vera, the memory roadmap, HBF, and the rest of what I’m attending

Aug 20 • Subbu

25

2

3

Understanding Texas Instruments: 271% cash flow surge, $60B fab build-out, and is it worth the price?

A first-principles look at the analog giant. What TI makes, how it makes money, and what you're paying for it.

Aug 2 • Subbu

9

1

July 2026

Qualcomm's AI Roadmap: The Bull and Bear Case

Why the Modular acquisition matters, how Qualcomm read the market, and whether it can ship before the bus leaves

Jul 20 • Subbu

22

1

2

Qualcomm High Bandwidth Compute (HBC): More Questions Than Answers

Who's making the memory stack? How good is the compute die? Will it really ship mid-2027?

Jul 10 • Subbu

12

1

Inside Attention-FFN disaggregation: What Groq LP40 will look like

How AFD works, from ByteDance's MegaScale-Infer paper to NVIDIA's Groq LPUs, and why the LP40 roadmap makes the acquisition hard to explain.

Jul 6 • Subbu

15

2

June 2026

The Psychology of Hype Cycles

A priest and the dot-com bubble, FOMO and the rational mind, Why we do this every time

Jun 14 • Subbu

49

4

8

May 2026

DRAM was the worst business in chips, now Micron is worth $1 trillion: 6 cycles of boom and bust and what could come next

How the cycle that built and buried memory companies broke, and the mistake Micron can't afford to repeat

May 31 • Subbu

14

5

1

The Uncomfortable Truth Behind Deploying the Latest NVIDIA GPUs: MFU, Silent Data Corruption, and the Real Moat for CSPs and Hyperscalers

Lessons from DeepSeek, ByteDance, Meta, Google and the long deployment gap between a Jensen keynote and production reality

May 10 • Subbu

19

1

1

March 2026

Why speculative decoding wants two kinds of silicon: NVIDIA, Groq, d-Matrix, Gimlet Labs, NVLink Fusion

Disaggregated speculative decoding, why the new class of accelerators are a better fit than GPUs, how 3D DRAM will put pressure on HBM based GPUs

Mar 14 • Subbu

37

8

Arista, AMD Pensando, and the Making of a Billionaire CEO

How Cisco and its culture of spin-ins led to the creation of Pensando and Silicon Valley's richest Indian-American CEO

Mar 1 • Subbu

31

9