FabSpeak
Ram Agarwal · with a robot, naturally

About FabSpeak

FabSpeak is a weekly breakdown of what is actually happening in AI and semiconductor manufacturing, aimed at the layer most coverage skips.

Hey, I’m Ram. Twenty years around manufacturing software and the factory floor. Every week, one story from the news gets taken apart at the layer that decides whether it ships.

44 pieces · 38 tips · 30 fourteen-second explainers · 12 videos on the channel · since July 2026

Every week, one story from the news gets taken apart: what the mechanism underneath it actually is, one number that makes it real, the strongest argument against my own reading, and a question for the people who actually run the line.

The recurring finding is simple enough to state up front. AI’s constraint in manufacturing was never the model. It is the data plumbing, the yield and the governance, and those are owned by manufacturing people, not AI people. Most coverage films the machine. The story is almost always the invisible layer that made the machine manufacturable.

What this covers

Manufacturing execution systems and integration. Process control, yield and metrology. Traceability and factory data. Industrial AI and how it gets governed on a floor where being wrong is expensive. Chip and packaging news, read as a manufacturing story rather than a market one.

How it is made

  • Every number comes from the piece it belongs to. Where a source is not on the page, the page says so.
  • Every piece names the strongest objection to itself.
  • No vendor has ever paid for a placement here, and none ever will.
  • Nothing on this site comes from any employer’s internal information. All of it is public.

Who writes it

Ram Agarwal in the studio, smiling

Ram Agarwal. Twenty years around manufacturing software and the factory floor. FabSpeak exists because the most interesting layer in this industry is the one nobody points a camera at. If something here is wrong, tell me, and the correction gets published.

YouTube · LinkedIn · RSS

Before and after

Drag the handle
Before · what gets filmed

A $400M machine prints the world’s most advanced chips.

The optics, the launch event, the render of the cleanroom. Every outlet runs the same shot.

After · what decides whether it ships

Holding 85% yield on a focus window one-third the size.

The process-control engineer, the overlay budget, the recipe that never gets a headline. That is the layer FabSpeak covers.

The thesis
AI’s constraint in manufacturing was never the model. It’s the data plumbing, the yield and the governance. Manufacturing people own all three.
SubjectMES, process control, yield, metrology, traceability, and how AI gets governed on a floor where being wrong is expensive.
FormatOne story a week. The mechanism underneath it, one number that makes it real, the strongest argument against my own reading, and a question for the people who run the line.
RegisterPlain language, real companies named, every acronym expanded on first use.
RuleOne number per claim, traced to the piece it came from. A missing source is stated on the page, never hidden.

How a piece is made

01

Pick one story

From the week’s AI and chip news: the one with a manufacturing mechanism under it, not the loudest one.

02

Find the layer beneath

What actually made it work, or fail: MES, process control, yield, data plumbing, governance.

03

Get one number, sourced

One figure that makes it real, traced to a public paper, filing or analyst report. Never invented.

04

Name the caveat, ask the question

The strongest argument against my own reading, then a question for the people who run the line.