SALIENCE PATHVERTICAL INTELLIGENCE SYSTEMS

AI Infrastructure Research

What must stay true for the buildout to work?

Most AI-infrastructure research starts with sectors, suppliers, or stocks. Salience Path starts one layer higher: with the assumptions that must remain true for the buildout to keep working.

The assumptions behind the buildout

Evidence cues → assumption → reconsideration

Select an assumption to see the evidence behind why we hold—or question—the view.

Dated research examples · November 2025–September 2026. Public observations with Salience Path interpretation—not a current-week model read.

Assumption 01

Capex & cash-flow credibility

Public observation

Cash generation is expanding

Amazon reported $161.4 billion in trailing-twelve-month operating cash flow, up 33%.

SupportsA larger operating cash base can help fund investment. It is not free cash flow or evidence of AI-project returns.

Public observation

Investment sets a demanding hurdle

Meta reported $31.08 billion in quarterly capital expenditures, including finance-lease principal payments.

PressuresLarge commitments raise the cash-generation and payback hurdle. Spending itself does not demonstrate a funding shortfall.

Connections show Salience Path interpretation of dated public observations—not proof of causation or a current-week model decision.

Assumption 02

Compute demand

Public observation

Cloud growth becomes operating income

Amazon reported AWS operating income of $16.6 billion, versus $10.2 billion a year earlier.

SupportsRealised cloud earnings are firmer evidence than a capacity announcement. AWS also includes non-AI workloads; GPU utilisation is not disclosed here.

Public observation

Named demand behind the buildout

Anthropic announced a $30 billion Azure compute purchase commitment and additional capacity contracting up to one gigawatt.

SupportsNamed commitments are stronger than unallocated capacity plans. Announced purchases are not delivered or utilised compute; supplier-linked investment also warrants scrutiny.

Connections show Salience Path interpretation of dated public observations—not proof of causation or a current-week model decision.

Assumption 03

Enterprise adoption & security

Public observation

Adoption alongside governance controls

Amazon reported hundreds of thousands of Bedrock customers and described governance and security controls alongside new agent-deployment capabilities.

SupportsReported adoption can coexist with control requirements. Vendor claims do not independently establish ROI or security effectiveness.

Public observation

Safeguards face a harder test

OpenAI disclosed an internal research incident involving its infrastructure and Hugging Face, and described tighter workload and network isolation.

PressuresThe incident increases the governance burden. It involved internal evaluation with reduced safeguards—not ordinary enterprise deployment.

Connections show Salience Path interpretation of dated public observations—not proof of causation or a current-week model decision.

Assumption 04

Macro & funding

Public observation

Broad conditions were not restrictive

At Jackson Hole, Federal Reserve Chair Warsh said he would be hard pressed to describe broad financial conditions as restrictive.

SupportsA supportive broad financing environment can reduce funding friction. An official's assessment is not evidence of financing access for each project.

Public observation

Further tightening remains possible

Governor Barr said policymakers should act decisively to raise rates if inflation does not moderate sufficiently.

PressuresConditional tightening is a risk to discount rates and debt service, not a policy decision or a measured increase in borrowing costs.

Connections show Salience Path interpretation of dated public observations—not proof of causation or a current-week model decision.

Assumption 05

Competition & pricing

Public observation

Lower-priced capability challenges premiums

Axios reported Kimi K3 launch pricing below premium rivals and competitive early leaderboard results. A weight release was still planned at publication.

PressuresCheaper alternatives can challenge pricing power. Launch claims do not establish realised economics, reproducible quality or completed weight release.

Public observation

An allegation is not a pricing measure

Anthropic alleged large-scale distillation campaigns by DeepSeek, Moonshot and MiniMax using fraudulent accounts.

Leaves unchangedA competitor's allegation does not quantify legitimate substitution, realised prices or market expansion. It is not independently adjudicated here.

Connections show Salience Path interpretation of dated public observations—not proof of causation or a current-week model decision.

How we decide what not to promote

Illustrative tests—not recorded suppression decisions. The question is whether a story adds evidence that changes an assumption, not how much attention it attracts.

Another product announcement
A new feature does not by itself establish paid production usage, safer deployment or improved returns. Those outcomes—not the launch headline—would change the read.
A familiar capacity commitment
Repeating an existing commitment adds no evidence of delivered capacity, utilisation or cash collection. A changed commitment or verified conversion would matter.
A leaderboard surge
A benchmark alone does not establish buyer switching or matched-workload economics. Reproducible performance and commercial displacement would cross the line.

Reports & Assessments

Sector reports follow the assumption map downstream: into industries, suppliers and companies, and the constraints connecting them. They explain where an assumption matters—not just which stock moved.