Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/PEOPLE/REINER POPE
// PERSON

Reiner Pope

ROLE CEOAT MADDOXMENTIONS 14LAST SEEN MAY 22, 2026
// BIO

CEO of Maddox chip startup and former Google TPU architect.

// RECENT MENTIONS
// SIGNALS
14 SIGNALS
01
mention·Dwarkesh·MAY 22, 2026

We've talked publicly about something which we call a splittable systolic array, which is in some sense you can think of as big systolic arrays that can be small systolic arrays as well.

Source
02
product·Dwarkesh·MAY 22, 2026

We've talked publicly about something which we call a splittable systolic array, which is in some sense you can think of as big systolic arrays that can be small systolic arrays as well.

Source
03
mention·Dwarkesh·MAY 22, 2026

That ratio is actually slightly wrong. It should be like an even bigger, you should get an even bigger speedup than you might otherwise think. Nvidia's product specs have sort of started acknowledging that in B300 and beyond, where the FP4 is three times faster than the FP8. Right. Though it should be 4X.

Source
04
mention·Dwarkesh·MAY 22, 2026

Seven-eighths of the cost is in the reading and writing the register file. And only a tiny fraction of the cost is in the logic unit itself. So this is the problem to solve.

Source
05
mention·Dwarkesh·MAY 22, 2026

The processes that are inside a lot of AI chips actually also have deterministic latency too. Groq has advertised this, TPUs have that in the core as well.

Source
06
hire·Dwarkesh·MAY 22, 2026

CEO of Maddox, former chip architect (likely Google TPU background given depth of TPU knowledge).

Source
07
hire·Dwarkesh·APRIL 29, 2026

Today I'm interviewing Reiner Pope, who is CEO of Maddox, which is a new chip startup. Previously, he was doing TPU architecture and many other things at Google.

Source
08
mention·Dwarkesh·APRIL 29, 2026

Today I'm interviewing Reiner Pope, who is CEO of Maddox, which is a new chip startup. Previously, he was doing TPU architecture and many other things at Google.

Source
09
mention·Dwarkesh·APRIL 29, 2026

From Hopper to Blackwell is mostly just the decision to switch from trays as the form factor... switching to racks as the form factor. That's a product decision.

Source
10
product·Dwarkesh·APRIL 29, 2026

DeepSeek V3 has about 37 billion active parameters and then 700 billion total parameters.

Source
11
mention·Dwarkesh·APRIL 29, 2026

DeepSeek mixture of experts has said actually activate more experts but finer grained experts — was a big innovation.

Source
12
mention·Dwarkesh·APRIL 29, 2026

Character AI has a blog post talking about that — alternating long and short context — and like in the global context which is really what we're talking about here, global context was shared across all the layers.

Source
13
mention·Dwarkesh·APRIL 29, 2026

Jensen math there, but there is at least a genuine 4x increase

Source
14
mention·Dwarkesh·APRIL 29, 2026

Reiner Pope, who is CEO of Maddox, which is a new chip startup. Previously, he was doing TPU architecture and many other things at Google.

Source

AI-extracted from podcast / newsletter / paper summaries. May contain errors.