Blog

I write about semiconductors, AI infrastructure, memory architecture, and public markets. Full articles are also published on my Substack, where paid subscribers get the full analysis.


Four Layers, One Moat: What Four AI Companies Taught Me This Week

April 17, 2026

Ben Pouladian analyzes: Ayar Labs, Tenstorrent, Databricks, and Wonder. Four companies at four altitudes. One playbook.


Oracle Just Signed the MSA Bloom Was Waiting For

April 13, 2026

Ben Pouladian analyzes: The market read 2.8 gigawatts. The structural read is that the warrant vested two business days early.


Credo Just Confirmed the Framework

April 13, 2026

Ben Pouladian analyzes: Same-day flash update to last night’s follow-up · Free post · Share widely


Credo Was the Wrong Question

April 12, 2026

Ben Pouladian analyzes: A reader correction, an on-camera primary source from NVIDIA’s networking SVP, and a Jefferies BUY initiation that all landed in the same Sunday. The Credo bear case has bee…


Sold Out Through 2028: Hurlston’s Tokyo Interview and the Optical Cold War

April 10, 2026

Ben Pouladian analyzes: Hurlston says Lumentum sells out through 2028 in two quarters. The AI infrastructure market is organizing around two power centers — and live news is validating the thesis i…


AI for the Rest of the World: Why Muse Spark Changes the Meta Thesis

April 9, 2026

Ben Pouladian analyzes: Meta is not chasing the frontier race. It is building a surface-specific consumer toolkit. Muse Spark is the first visible confirmation.


The Compute Confession

April 7, 2026

Ben Pouladian analyzes: Anthropic Just Built a Frontier Model They Cannot Afford to Run


The Token Dollar

April 4, 2026

Ben Pouladian analyzes: Oil is scarce because of war. Tokens are scarce because of physics.


2.7x on the Same Iron: MLPerf v6.0 Just Proved Nvidia’s Extreme Co-Design Thesis

April 1, 2026

Ben Pouladian analyzes: NVIDIA’s software stack just delivered 2.7x more tokens from the same GB300 NVL72 hardware in six months. That’s not a product launch. That’s a moat.


The Orchestration Layer Is the New Platform War — NVIDIA’s AI Agent Strategy | GTC 2026 (Podcast)

March 29, 2026

Ben Pouladian presents “The Orchestration Layer Is the New Platform War — NVIDIA’s AI Agent Strategy | GTC 2026 (Podcast)” — deep-dive analysis from BEP Research covering AI infrastructure, semicon…


TurboQuant and Dynamo: Why the Market Has the Memory Thesis Backwards

March 27, 2026

Ben Pouladian analyzes: Google resurfaced a year-old compression paper. NVIDIA just shipped the software that explains why it doesn’t matter.


The Agentic CPU: Arm’s First Silicon Shot at x86

March 25, 2026

Ben Pouladian analyzes: Arm’s first-ever production silicon targets a fast-forming CPU gap in agentic AI infrastructure


The Token Explosion: Why GTC 2026 Was Really About Building the World’s Largest Token Factory

March 21, 2026

Ben Pouladian analyzes: The infrastructure map for a world that needs orders of magnitude more tokens — with exclusive NVIDIA interviews from GTC 2026.


Micron Just Proved the Memory Thesis

March 18, 2026

Ben Pouladian analyzes: $23.9 billion. 75% gross margins. Revenue nearly tripled. And the supply gap is widening


The Fourth Piece Ships: NVIDIA’s Groq 3 LPX Changes the Inference Business Model

March 16, 2026

Ben Pouladian analyzes: Live from GTC 2026: Seven chips, five racks, and the slide that rewrites how Wall Street should model AI infrastructure revenue


The Foundry Behind the Photonic Boom: Why Tower Semiconductor Is the Most Important Chip Company You’re Not Watching

March 14, 2026

Ben Pouladian analyzes: Tower’s Q4 earnings confirm what the optical supply chain already knew: the 1.6T transition may be bottlenecked by one foundry.


The NeoCloud Hypothesis

March 10, 2026

Ben Pouladian analyzes: Why CoreWeave and Oracle Are NVIDIA’s Real Distribution Partners—and What That Means for Your Portfolio


The Quiet Architect

March 7, 2026

Ben Pouladian presents “The Quiet Architect” — deep-dive analysis from BEP Research covering AI infrastructure, semiconductors, and datacenter investment.


The Reasoning Tax: Why GPT-5.4 Just Validated the Memory Wars

March 5, 2026

Ben Pouladian analyzes: OpenAI’s latest benchmarks prove smarter models don’t save infrastructure — they devour it. And they explain why NVIDIA paid $20 billion for Groq


The Other Memory Wall, Part 2

March 4, 2026

Ben Pouladian analyzes: Apple’s Two-Layer AI Play Is Happening in Real Time


View all 55 posts →   |   Subscribe on Substack →