P

Perplexity

$20B · AI Search · Founded 2022

AnalyticsSupply ChainCompetitorsFunding

Supply Chain Control Room

A more visual map of how Perplexity depends on infrastructure, partners, and operational chokepoints.

The stack below turns tier data into an exposure view: where dependencies cluster, which layer carries the most vendors, and how many live risk signals are already mapped.

Mapped tiers

6

Critical Elements

Node clusters

14

20 listed dependencies

Risk signals

13

0 formal assessments

Highest severity

critical

13 flagged nodes

T01Models & Products
T02Infrastructure
T03Processors
T04Memory & Network
T05Foundries & Machines
T06Critical Elements

Exposure Mix

Operational risk by signal density

Live map
Signals13across 6 tiers
critical4
high5
medium4
low0

Dominant concentration

Critical Elements carries the widest cluster with 4 mapped nodes.

Dependency Lanes

Each tier is rendered as a lane, with grouped nodes and dependency payloads surfaced directly inside the mapped layer.

6 tiers · 14 nodes · 20 dependencies
01

Tier 01

THE MIND (RENTED)

Models & Products

Perplexity does NOT train frontier models — it orchestrates them.

Node clusters

2

Dependencies

7

NodeHIGH

Model Layer (External)

Items

3

  • GPT-4o / GPT-5.2 via OpenAI APIRENTED
  • Claude 4.6 via Anthropic APIRENTED
  • Sonar (own search model, built on Llama)OWN
Node

Revenue Hubs

Items

4

  • Perplexity Pro/Max subscriptionsPRIMARY
  • Sonar API (enterprise search)GROWING
  • Comet browserNEW
  • Computer (agentic agent)NEW 2026
02

Tier 02

THE LANDLORD

Infrastructure

Single-cloud Azure dependency via $750M 3-year commitment.

Node clusters

2

Dependencies

3

NodePRIMARYHIGH

Microsoft Azure

Items

2

  • $750M compute commitment (Jan 2026)LOCKED
  • GPU capacity for Deep Research + AgentsCRITICAL
NodeMEDIUM

No Multi-Cloud

Items

1

  • No GCP or AWS fallback reported
03

Tier 03

THE ORGANS

Processors

NVIDIA-only via Azure. Zero custom silicon.

Node clusters

2

Dependencies

2

NodeHIGH

NVIDIA GPUs (via Azure)

Items

1

  • H100 / Blackwell for inferenceVIA AZURE
NodeMEDIUM

No Custom Silicon

Items

1

  • Inference-heavy, not training-heavy
04

Tier 04

NERVOUS SYSTEM

Memory & Network

Standard cloud infrastructure. Bottleneck is API rate limits, not raw HBM.

Node clusters

2

Dependencies

2

NodeMEDIUM

HBM4 (Indirect)

Items

1

  • Indirect exposure via Azure cluster availability
NodeHIGH

API Rate Limits

Items

1

  • Multi-model orchestration requires stable API accessCRITICAL
05

Tier 05

THE FACTORY

Foundries & Machines

Every AI chip comes from these two companies.

Node clusters

2

Dependencies

2

NodeCRITICAL

TSMC

Items

1

  • Only 3nm/2nm foundryMONOPOLY
NodeCRITICAL

ASML

Items

1

  • High-NA EUV. No alternative.NO ALT
06

Tier 06

THE INGREDIENTS

Critical Elements

The Silicon Curtain of 2026.

Node clusters

4

Dependencies

4

NodeCRITICAL

Gallium & Germanium

Items

1

  • 25% revenue tax on chip salesGEOPOLITICAL
NodeCRITICAL

Neon Gas

Items

1

  • ~50% of semiconductor-grade neonWAR ZONE
NodeHIGH

Copper

Items

1

  • Record prices hit DC costsCOST
NodeMEDIUM

Silicon Wafers

Items

1

  • Shin-Etsu & SUMCODUOPOLY