CONTEXT JAMMING

Field notes from inside the context window.

FounderFiles · N°019

Andrew Feldman — CEO & Co-founder of Cerebras Systems

Andrew Feldman.

The systems architect who spent two decades arguing that interconnect, not raw compute, is the real limiter at extreme scale — and built the largest chip in history to prove it.

TRAINED
Systems Architecture · SeaMicro (acquired by AMD)
AT
Cerebras Systems (CEO & Co-founder)
FILE
N°019
§ 01 · The Architectural Bet

Wafer-Scale vs. Scale-Out Clusters

Andrew Feldman has spent nearly two decades making one consistent argument: at extreme scale, the interconnect and memory bandwidth limitations of traditional GPU clusters become first-order constraints. His solution is radical — build one enormous monolithic chip instead of networking thousands of smaller ones.

This is the core of Cerebras’ Wafer-Scale Engine thesis. While the industry doubled down on scale-out clusters, Feldman bet that a single, massive chip with enormous on-chip SRAM and bandwidth could bypass many of the fundamental bottlenecks that appear when coordinating tens of thousands of GPUs.

The biggest constraint on training bigger models may no longer be the chip — it may be the wires between the chips.
On the limitations of scale-out at frontier scale
§ 02 · Systems Reality

Power, Thermals, and Manufacturing

Wafer-scale computing concentrates enormous challenges into one device. Power delivery, cooling, and yield management become dramatically harder. Cerebras engineered a sophisticated defect-tolerant architecture with 70,000 redundant cores and dynamic routing to achieve ~93% active silicon utilization across a full 300mm wafer — something previously considered economically impossible.

The result is a chip with 21 PB/s of on-wafer memory bandwidth, representing roughly a 7,000× advantage over a discrete H100 in memory bandwidth.

§ 03 · Software & Adoption

The Developer Experience Friction

Even the most elegant hardware architecture struggles if moving models onto it is painful. Feldman has been direct about the software and ecosystem challenges of introducing a new architecture into a CUDA-dominated world. The Cerebras stack requires Ahead-of-Time compilation and static graphs, creating real friction for researchers used to dynamic PyTorch workflows.

The long-term success of wafer-scale computing depends as much on closing this developer experience gap as on raw silicon performance.

§ 04 · Formation

Two Decades of Systems Thinking

Feldman’s conviction didn’t appear overnight. His previous company, SeaMicro (acquired by AMD), was already focused on dense, efficient compute systems and innovative interconnect fabrics. The through-line from SeaMicro to Cerebras is a consistent focus on systems-level problems rather than chasing peak single-chip performance in isolation.

The Index
WSE-3
46,225 mm² monolithic wafer with 4 trillion transistors and 900,000 AI cores
21 PB/s
On-wafer memory bandwidth — ~7,000× advantage vs discrete H100
93%
Active silicon utilization via defect-tolerant architecture
SeaMicro
Feldman’s prior company — early systems-level thinking on dense interconnect
Monolithic
Core bet: collapse the network into the silicon rather than scale it out
Key Talk

Andrew Feldman on wafer-scale computing

Career Shape
I-shaped — a single maximal-depth spike

I-Beam Theorist

Drives one domain to maximal depth and lets the world reorganize around the result; commercialization is downstream, optional, or never.

Credential Path
Practitioner
Abstraction
Top Down
Exit Horizon
Deferred
Moat Instinct
Product Primitive
Capital Posture
Venture
Role-Model Reference Class
  • SeaMicro team
  • hardware systems architects
  • long-horizon chip designers
Founder Context · JSON

A small reasoning persona distilled from this file. Inject it into a chat or deep-research context to assess a business problem the way Feldman would.

Reason as Andrew Feldman. When given a scaling or systems problem at extreme AI size, first ask whether the implicit assumption is "more chips + better network" and whether that assumption is the actual limiter. Propose collapsing the interconnect problem into the silicon itself. Emphasize long-term architectural conviction over short-term cluster wins. Audit whether a proposed solution treats the network as a first-class citizen or as something to be engineered out of existence.

{
  "$schema": "https://www.contextjamming.com/schemas/founder-context-v1.json",
  "file": "N°019",
  "persona": "Andrew Feldman",
  "archetype": "i-beam",
  "shape": "I",
  "one_line": "Drives a single, decades-deep conviction that interconnect is the fundamental limiter at extreme scale, and that the only way through is to collapse the network into a single, defect-tolerant wafer-scale silicon substrate.",
  "cognitive_basis": {
    "credentialPath": "practitioner",
    "abstractionDirection": "top-down",
    "exitHorizon": "deferred",
    "moatInstinct": "product-primitive",
    "capitalPosture": "venture"
  },
  "operating_questions": [
    "What is the real bottleneck when you scale AI training to the largest possible systems?",
    "If the network is the problem, why keep scaling the network instead of removing it?",
    "How do you make a chip so large that traditional manufacturi
  …
Share
FounderFiles N°019 · Andrew Feldman
Filed by Bret Kerr · ACRA Insight LLC · Franklin, MA
contextjamming.com · @bretkerr

§ · Invoice No. 001 · The Build Ledger

The Ledger.

Filed · contextjamming.com

What a conservative mid-market digital agency would have quoted for the same scope, itemized against what this site actually cost. Agency numbers are the floor — not the premium brand-studio tier.

TIME

12 weeks

2 days

~42× faster

COST

~$150,000

~$300

~500× cheaper

TEAM

5-person agency

1 human + 3 models

Same deliverable

§ Itemized — what a mid-market agency SOW would have billed

Discovery · brand positioning · workshops40–80 hr$10,000
Design system · Figma tokens · 3 rounds60–120 hr$18,000
Wavesurfer audio carousel · single-track context60–100 hr$16,000
Dual lightbox systems · focus trap · keyboard30–50 hr$8,000
LLM product flows · streaming · state machine80–160 hr$26,000
Stripe · checkout · webhooks · env hardening40–80 hr$10,000
Editorial routes · 6 sub-pages · templates60–100 hr$14,000
Accessibility pass · aria · reduced-motion40–80 hr$10,000
QA · cross-browser · mobile matrix60–100 hr$14,000
Cross-publication rebrand · masthead + IA · 2026-04-2820–40 hr$6,000
Subtotal~700 hr$126,000
Project management · 18% overhead$24,000
Agency total — conservative floor~700 hr~$150,000
Actually spent · Claude + Gemini stack~20 hr~$300

Agency figure assumes ~700 billable hours at $200/hr blended, plus ~18% PM overhead — the conservative floor of a mid-market SOW. Premium brand studios would have quoted 2–3× that. Stack: Antigravity (orchestrator), Claude Opus 4.8 (auditor), Codex (adversary), Cloudflare Workers / OpenNext.

§   Colophon

How this site is made.

Vol. 26 · build log

Every page on contextjamming.com is the output of a real-time, three-body Mixture-of-Experts loop. One model orchestrates. Two consult. The human holds the thesis. No single model commits alone.

View Redesign Assessment →

Orchestrator

Antigravity

Google DeepMind

  • Primary author
  • Terminal-native, direct push to Cloudflare
  • Audit trail to GitHub on every commit
  • Adaptive thinking · effort: extra-high

Auditor

Claude Opus 4.8

1M context

  • Editorial critic
  • Code review before merge
  • Backup-of-record
  • Co-signs every commit

Adversary

Codex

Cross-model MoE

  • Factual adjudication
  • Structural dissent
  • Deep Research → semantic triples
  • Caught the Donelan incident

Stack

Next.js
16.2 · App Router
React
19.2
TypeScript
5
Tailwind
v4 · @theme inline
@opennextjs/cloudflare
adapter
wrangler
Pages deploy
framer-motion
transitions
wavesurfer.js
audio waveforms

Typeset in

Fraunces
variable · opsz + SOFT
Playfair Display
debate display
IBM Plex Mono
editorial metadata
Geist Mono
utility mono
Caveat
grease-pencil marginalia
All via
next/font/google
Palette
single @theme block
No dupe tokens
ever

Infrastructure

Deploy
Cloudflare Workers / OpenNext
ISR
30-min revalidate · Cloudflare-served
Repo
github.com/BretKerrAI/founderfile
Branch
main
Analytics
Google Tag Manager
Apex
contextjamming.com
Runtime
Node 24
Build tool
Turbopack
       human intent
            │
            ▼
   ┌────────────────────┐         ┌─────────────────┐
   │    Antigravity     │  ◄────► │ Claude Opus 4.8 │      ← auditor loop
   │    (orchestrator)  │         │     (auditor)   │
   └─────────┬──────────┘         └─────────────────┘
             │  ◄───────────┐
             ▼              │
       ┌──────────┐    ┌────┴───────┐
       │Cloudflare│    │   Codex    │          ← adversarial loop
       │ Workers  │    │            │
       └─────┬────┘    └────────────┘
             │
             ▼
       contextjamming.com
             │
             ▼
       ┌──────────────┐
       │   Git push   │         ← audit trail
       └──────────────┘
Assembled on Mac in Terminal · Filed from Franklin, MAContext Jamming · ACRA Insight LLC · MIT License · FounderFile.ai · RelationalIntelligence.xyz · Commission a Dispatch →