OUTSIDE THE BUBBLE

Micron reported record fiscal Q3 results and disclosed a new class of strategic customer agreements that could change how memory gets financed.

The company said it has completed 16 strategic customer agreements across data center, consumer, and auto markets. Its Q3 filing says those agreements include binding multi-year volume commitments and are expected to bring $22 billion of cash deposits and related financial commitments, with roughly $18 billion in cash deposits. Micron also said remaining performance obligations tied to minimum committed volumes and minimum pricing were approximately $100 billion.

Reuters framed the point directly: AI-driven shortages are forcing large data center customers to fund capacity, reshaping the memory market.

The headline story is not just Micron beating numbers.

It is customers treating memory like strategic infrastructure.

That is different.

Memory has historically been a brutal commodity cycle. Too much supply, pricing collapses. Too little supply, pricing spikes. Customers wait it out. Suppliers eat volatility.

AI is changing the behavior.

HBM is not generic DRAM sitting on a shelf. It is tied to accelerator roadmaps, advanced packaging, qualification cycles, and supply windows that cannot be replaced at the last minute.

The binding constraint is no longer just who can design the fastest chip.

It is who can secure the memory stack around it.

The market may be missing the contractual shift underneath the earnings surge.

This is not only price inflation.

It is reservation economics.

When customers put real deposits behind multi-year supply, the supplier’s business model starts to look less like spot commodity exposure and more like capacity allocation.

That does not eliminate cyclicality.

Memory will still be memory.

But customer prepayments, minimum commitments, and pricing floors can reduce the left-tail risk that historically crushed margins when supply caught up.

The under-modeled variable is not “AI demand is strong.”

Everyone knows that.

The under-modeled variable is customer behavior when AI infrastructure buyers decide supply certainty is worth paying for upfront.

Winners: HBM suppliers, advanced packaging, memory controllers, test equipment, substrate capacity, high-end storage, and anyone with validated capacity tied to AI roadmaps.

Squeezed: AI infrastructure buyers without balance sheet leverage, smaller model companies, downstream hardware vendors exposed to rising bill-of-material costs, and any cloud customer assuming inference costs keep falling smoothly.

The budget line expanding is not “chips.”

It is supply assurance.

The moat becomes committed capacity, not just design leadership.

That favors suppliers with process execution, customer qualification, and enough credibility to sign multi-year contracts.

This earns Level 3 because customer prepayments and take-or-pay style agreements can structurally alter memory cyclicality over the next 1–5 years. The market is pricing the near-term earnings surge, but may still be underpricing the contractual shift from spot component supply to reserved AI infrastructure.

This week’s power signals were louder than the chip headlines.

Bloom Energy and Brookfield expanded their AI infrastructure power partnership to $25 billion, up from a prior $5 billion framework, to finance fuel-cell power projects for AI data centers. Reuters also reported that data center investors are buying power developers to secure electricity and accelerate buildouts, highlighting DigitalBridge’s $1.1 billion deal for ArcLight Capital as part of a broader convergence between digital infrastructure and power infrastructure.

The numbers explain the urgency.

Reuters cited Goldman Sachs estimating U.S. data center power demand rising from 31 GW in 2025 to 66 GW in 2027.

The market still talks about AI infrastructure like the bottleneck is only GPU supply.

That is stale.

The edge is shifting from compute procurement to energy control.

A data center without firm power is not capacity.

It is a rendering.

Power is becoming part of the AI product stack: generation, interconnection, grid position, dispatchability, backup, cooling, permitting, and local political tolerance.

This is not just real estate.

It is industrial project finance.

The operator that can secure power faster can monetize compute faster.

The operator waiting in a queue owns a press release.

The market may be missing that “megawatts” are not all equal.

The real variable is usable megawatts.

Fast megawatts.

Permitted megawatts.

Dispatchable megawatts.

Megawatts with interconnection rights and credible local execution.

That is why data center capital is moving upstream into power developers. It is not because cloud companies suddenly want to be utilities for fun.

It is because the grid is becoming the deployment bottleneck.

The real tell is not one deal.

It is the merger of two capital stacks that used to be modeled separately: digital infrastructure and energy infrastructure.

Winners: power developers, fuel cells, grid equipment, switchgear, transformers, power electronics, cooling systems, battery storage, gas generation, nuclear developers, and landowners with credible energy access.

Squeezed: data center developers selling shell capacity, cloud buyers without power leverage, local utilities forced to absorb load growth, and AI platforms whose economics assume cheap, frictionless compute expansion.

The budget line expanding is energy assurance.

Power contracting becomes a moat.

Balance sheet scale matters more.

Permitting capability matters more.

Depreciation and utilization matter more because idle compute tied to constrained power destroys the model.

This earns Level 3 because power is now a durable gating factor for AI infrastructure over the next 1–5 years. The market is pricing the AI capex boom, but still underpricing how much leverage shifts to energy access, grid position, and infrastructure execution.

Microsoft made Work IQ APIs generally available on June 16 and priced them through Copilot Credits. There is no separate Work IQ API subscription, SKU, or per-user license. Microsoft says consumption includes variable charges for grounding, retrieval, and reasoning, plus a static usage component for tool actions.

Microsoft also made Agent 365 broadly available as a control plane to observe, govern, and secure agents across the enterprise.

The same pattern is showing up elsewhere. ServiceNow and Accenture launched services to move companies from legacy risk platforms into agentic AI workflows, and Salesforce agreed to acquire Fin for roughly $3.6 billion to deepen Agentforce in customer service automation.

The market is still asking whether agents replace SaaS.

Wrong question.

Agents do not kill enterprise software overnight.

They change what enterprise software sells.

Seats were the old unit.

Work becomes the new unit.

Control becomes the monetization layer.

The real tell is Microsoft’s Work IQ pricing. When an agent retrieves context, reasons across enterprise data, invokes tools, and takes action, the billing unit starts to look less like “employee with access” and more like “machine work performed inside a governed system.”

That is a different pricing model.

It is also a different margin model.

The market may be missing that governance is not a feature.

It is the toll booth.

Enterprise agents need identity, permissions, data boundaries, logging, audit, rollback, observability, cost controls, and policy enforcement.

Without that, they stay trapped in demos and shadow workflows.

This is where the control plane matters.

Not because every agent is ready to replace labor.

Because every enterprise now needs to answer basic operating questions.

Who authorized the agent?

What did it access?

What did it do?

What did it cost?

How do we stop it?

That becomes a budget line.

Winners: Microsoft, ServiceNow, Salesforce, identity platforms, workflow systems, security vendors, observability tools, GRC vendors, integration platforms, and systems integrators that can turn messy workflows into controlled agent deployments.

Squeezed: point SaaS vendors whose value was interface ownership, not workflow authority.

If agents become the interface, the application UI loses leverage.

Pricing power shifts toward platforms with context, permissions, execution rails, and auditability.

Labor changes too.

The first wave is not total replacement.

It is task erosion.

Support, IT ops, compliance triage, sales ops, finance workflows, research, and ticket resolution get measured at the activity level.

The seat model hides underutilization.

Usage pricing exposes it.

This earns Level 2 because the pricing and governance rails are now real, but enterprise adoption will be uneven over the next 1–3 years. The market is correctly pricing that agents matter, but underpricing the control-plane layer and overpricing generic agent wrappers with no workflow authority.

The headline story is not smarter models — it is reserved capacity, controlled power, and governed execution turning AI from capability into infrastructure.

— Connor
Alpha Before It Prints

🔗 Click to Connect with me

© 2026 Alpha Before It Prints
Unsubscribe