SHARK MONEYYACHT CLUB

Search Shark Money

Search public and private companies, people, funds, markets, tools, and research.

Join free
All market currents
AI InfrastructureNVDAAVGOAMDANETVRTSMYC / MARKET MAP
Technology · Compute buildout

AI Infrastructure

The systems that turn model demand into usable compute: accelerators, networking, power delivery, cooling, and data-center infrastructure.

6 constituentsEqual-weighted viewLive delayed data
1 day—
1 week—
1 month—
1 year—
3 years—
What is included

Constituent companies

NVDA
NVIDIAAccelerated compute

Designs GPUs, networking products, and software used across accelerated computing workloads.

Price—
1D—
1W—
1M—
1Y—
AVGO
BroadcomNetworking & custom silicon

Supplies connectivity products and custom accelerators for hyperscale data centers.

Price—
1D—
1W—
1M—
1Y—
AMD
Advanced Micro DevicesCompute platforms

Builds CPUs and accelerators serving cloud, enterprise, and embedded markets.

Price—
1D—
1W—
1M—
1Y—
ANET
Arista NetworksData-center networking

Provides high-speed switching and routing platforms for large-scale data centers.

Price—
1D—
1W—
1M—
1Y—
VRT
VertivPower & cooling

Supplies critical digital infrastructure for power management and thermal control.

Price—
1D—
1W—
1M—
1Y—
ETN
EatonElectrical systems

Produces power-management equipment used across data centers, grids, and industrial systems.

Price—
1D—
1W—
1M—
1Y—

Returns are calculated from adjusted prices and may be delayed. Themes are research groupings, not investable indices.

Full-stack map

Where value can accrue

Move from the headline theme to the operating chain, measurable evidence, indirect beneficiaries, and questions that can disprove the thesis.

Value chain

01

Semiconductor equipment and foundries

02

Accelerators and custom silicon

03

Memory, optics, and networking

04

Power, cooling, and data-center construction

05

Cloud capacity and model deployment

Key metrics

01

Hyperscaler capex and AI mix

02

Accelerator lead times and utilization

03

High-bandwidth memory supply

04

Network attach per compute cluster

05

Power availability and energization dates

06

Inference cost per useful workload

Second-order exposure

01

Electrical equipment and grid construction

02

Liquid cooling and thermal management

03

Optical interconnects

04

Data-center real estate and natural-gas generation

Diligence questions

01

Which bottleneck captures economics after accelerator supply normalizes?

02

How much demand is recurring inference versus one-time training buildout?

03

Where is customer concentration highest?

04

Which capacity additions create oversupply risk?

Deep dive

What the theme
must prove.

AI spending is becoming a full-stack infrastructure cycle. The durable opportunity may migrate from the most visible chip to the recurring bottlenecks around networking, power, cooling, and deployment.

Structural drivers

Accelerator deploymentNetwork bandwidthPower availabilityCooling density

Evidence to track

Hyperscaler capital spendingNetworking backlogPower-system lead timesInference utilization

Thesis breakers

Customer concentrationCapacity overshootArchitecture shiftsCapital-cycle compression
People shaping the current

Operators, founders & capital allocators

People intelligence
Satya NadellaExecutive

Technology executive

Satya Nadella

Microsoft

An operating profile centered on cloud infrastructure, enterprise software, artificial intelligence distribution, partnerships, and large-scale capital allocation.

Marc AndreessenBillionaire

Venture investor

Marc Andreessen

Andreessen Horowitz

A broad venture-capital network spanning artificial intelligence, enterprise software, defense technology, crypto, consumer platforms, and American dynamism.

Related intelligence

Read across the current

All articles
Deep Water11 min read

Mapping the Physical AI Stack

The robot is the visible product. The durable economics may sit several layers beneath it.

Read intelligence
Infrastructure10 min read

Power Is the New Compute Constraint

The next wave of compute may be shaped as much by interconnection, generation, and cooling as by chips.

Read intelligence
Playbook7 min read

How to Kill Your Own Thesis

The goal of research is not to defend the original idea. It is to discover what is true before the market makes the lesson expensive.

Read intelligence
Research grouping—not an index.

Theme performance is an equal-weighted average of available constituent returns for orientation only. It excludes dividends, fees, taxes, rebalancing, and investability constraints. Nothing here is investment advice.