The Operating Layer

A town manager with six employees may now be expected to negotiate power, water, taxes, construction impacts, and community benefits with a data-center developer worth hundreds of billions of dollars.

Apparently “economic development” now means becoming a utility planner, infrastructure economist, environmental analyst, and contract negotiator between Tuesday’s pothole complaints and Thursday’s zoning meeting.

That is absurd.

It is also the best business opportunity I found this week.

The same pattern appears in federal subcontracting, international freight, connected products, and laboratory equipment. A new rule, dataset, or technical interface appears. The information technically exists. Then a real person still has to turn it into a sound decision, a clean handoff, or a machine that does not do something expensive and stupid.

Quick compare: five operating-layer businesses

Opportunity 1: The Data Center Community Deal Room

The problem. A developer arrives with consultants, demand forecasts, utility studies, tax projections, water estimates, construction schedules, and lawyers. The town gets a Dropbox folder. Documents conflict. Assumptions move. Promises made in a public meeting may never become measurable obligations in the final agreement.

Why now. Massachusetts issued an executive order on September 10 requiring more responsible data-center development, including attention to ratepayers, communities, and the environment. New Jersey’s Economic Development Authority has launched a technical- assistance program for communities evaluating data centers. On September 17, the Federation of American Scientists published a guide built from ten public agreements and warned that local governments often negotiate from a structural disadvantage.

What is known: states and municipalities are being pushed toward more formal review while projects are getting larger. My read: most communities do not need another dashboard; they need one defensible record of claims, assumptions, concessions, and deadlines. The bet: a neutral, software-assisted deal room can become the working system shared by municipal staff, outside experts, elected officials, and eventually the developer.

Existing solutions. Engineering firms, land-use lawyers, economic-development consultants, and utility advisers already do important pieces of this work. What is usually missing is the connective tissue: every developer claim linked to supporting evidence, every assumption compared with peer agreements, every unresolved question assigned to someone, and every negotiated commitment translated into a measurable deadline with a remedy.

The AI advantage. AI can extract claims and obligations from hundreds of pages, normalize terminology, compare provisions across agreements, maintain a cited question log, and draft meeting briefs. Power, water, traffic, tax, and financial models should remain deterministic. Engineers, lawyers, utility experts, and public officials make the judgments. This is decision support, not an algorithm wearing a hard hat and pretending to be town counsel.

Competition moat. The moat is not the document chat. It is a growing corpus of agreements and outcomes, reusable impact models, local-government relationships, and a reputation for being useful without becoming the developer’s salesperson or the opposition’s slogan machine.

Risk. Government sales move slowly. Documents can be confidential. Advice can create liability. The service needs clean boundaries, professional partners, and a visible audit trail. Still, the need is real and the buyer’s alternative is often a heroic spreadsheet assembled at 11 p.m.

Signals: Massachusetts Executive Order 658, NJEDA’s Data Center Technical Assistance Program, and FAS’s September 17 local-government guide.

Opportunity 2: The DOE M&O Opportunity Radar

The problem. Small companies searching SAM.gov for Department of Energy work are watching the front door while much of the freight is moving through the loading dock. DOE says roughly 80 percent of its contract dollars are spent through management-and-operating and facility-management contractors. Those contractors buy construction, fabrication, IT, maintenance, safety equipment, research support, and hundreds of other things through their own forecasts, supplier portals, events, and procurement teams.

Why now. DOE refreshed its acquisition forecast on September 11 and maintains a public directory that maps its sites and prime contractors to small-business contacts. What is known: the data exists, but it is distributed across agency pages, spreadsheets, contractor portals, and inconsistent descriptions. My read: a supplier rarely needs more opportunities; it needs fewer, better-matched ones and a credible next step. The bet: a focused research service can sell before the software is fancy.

Existing solutions. GovTribe and HigherGov do strong federal-market discovery, including forecasts and subcontracting data. The remaining gap is site-specific execution: what a particular laboratory or facility buys, which contractor controls the purchase, where the forecast lives, who handles small-business outreach, how the supplier’s past performance maps to the requirement, and what to do this week.

The AI advantage. Use AI to match messy capability statements against forecast descriptions, classify fit, identify missing proof, and draft a short cited introduction. Humans decide whether the company can actually perform. Semantic similarity is not a license to recommend a roofing contractor for a nuclear instrumentation package because both mention “high-performance systems.”

Competition moat. Build a history of which suppliers pursued which opportunities, where introductions led, what each site actually bought, and which capability claims produced meetings. Distribution can come through APEX Accelerators, trade groups, community colleges, and regional manufacturers rather than another heroic attempt to win Google search traffic.

Risk. Forecasts move, contractor pages break, and suppliers overstate readiness. That makes research discipline and honest fit scoring part of the product. The first version should be a paid matching sprint with a human in the loop, not a national procurement platform.

Signals: DOE’s September 11 acquisition forecast, DOE’s small-business program manager directory, GovTribe’s forecast coverage, and HigherGov’s subcontracting workflow.

Opportunity 3: The ICS2 MRN Exception Desk

The problem. European import filings are becoming less tolerant of vague descriptions, missing parties, late data, and the general freight-forwarding tradition of solving everything through an email chain called RE: RE: FINAL DOCS 7. A missing Movement Reference Number can now stop cargo before loading. The container does not care that someone “thought customs had it.”

Why now. On September 7, Maersk announced a “No MRN, No Load” policy for relevant EU imports beginning September 30, with MRN submission required 24 hours before the estimated arrival at the first load port. The European Commission continues to warn that low-quality or incomplete ICS2 data can be rejected or delayed. What is known: carriers and customs systems are enforcing earlier, cleaner handoffs. My read: the pain sits upstream, where exporters, buyers, brokers, and forwarders still disagree about who owns each field. The bet: exception management is more valuable than another filing screen.

Existing solutions. Customs platforms, carrier portals, freight-management systems, and brokers handle declarations and transmission. The gap is the coordination layer before filing: assign responsibility, validate the minimum dataset, monitor cutoffs, chase missing information, and preserve the evidence trail when something fails.

The AI advantage. AI can extract shipment facts from commercial invoices, packing lists, booking confirmations, and email; normalize descriptions; spot likely inconsistencies; and draft precise requests for missing fields. Deterministic rules should validate required fields, formats, EORI structure, route logic, and cutoff times. A qualified broker or filing party remains responsible for the declaration.

Competition moat. A clean record of recurring exceptions by shipper, lane, supplier, and document type becomes valuable quickly. Add broker relationships and embedded templates, and the product becomes harder to replace than the extraction model itself.

Risk. The initial market is concentrated in EU-bound freight, carrier policies vary, and errors can create real costs. Keep the system away from unsupervised filing, define roles clearly, and sell the reduction in exceptions and missed cutoffs rather than “AI customs compliance.”

Signals: Maersk’s September 7 policy notice, the European Commission’s ICS2 overview, and the Commission’s ICS2 FAQ.

Opportunity 4: The Connected Product Data Access Test Lab

The problem. The EU Data Act says users of connected products should be able to access and share the data their products generate. A manufacturer can add an export button, write a compliance memo, and declare victory. Whether the export is complete, usable, timely, machine-readable, secure, revocable, and available to an authorized third party is a separate question.

A slide deck can say the product is compliant. A slide deck cannot execute a GET request.

Why now. The Data Act’s access-by-design obligations began applying on September 12, 2026, to connected products and related services placed on the EU market after that date. What is known: product teams now have to turn legal language into observable software behavior. My read: many will document an intended workflow without repeatedly testing the workflow users actually receive. The bet: an independent, technically literate test service can sit between counsel, product, support, security, and engineering.

Existing solutions. Data-sharing platforms such as Steelbridge and Dativo help companies expose, govern, and contract around product data. They validate the underlying need. The opening is not another general portal. It is independent behavior and regression testing: can the customer find the feature, retrieve the expected data, authorize another party, revoke access, understand failures, and repeat the process after the next firmware release?

The AI advantage. AI can turn product documentation and policies into candidate test cases, map failures to requirements, cluster repeated defects, and draft a cited report for engineers and counsel. Deterministic test code must perform the actual requests, compare schemas, measure latency, verify permissions, and retain reproducible evidence.

Competition moat. Reusable device-and-cloud test harnesses, failure taxonomies, a library of product-specific edge cases, and trusted relationships with technical counsel become more valuable with every release tested. Portability matters: customers should own the evidence and be able to hand it to another lab, regulator, partner, or auditor.

Risk. Legal interpretation, trade secrets, cybersecurity, and product safety create hard boundaries. The lab tests observable behavior and packages evidence; it does not certify legal compliance. Start with one willing manufacturer and one clearly scoped access journey.

Signals: the European Commission’s Data Act explainer, the official regulation text, Steelbridge, and Dativo.

Opportunity 5: The MHS Driver & Safety Harness Studio

The problem. Laboratories and small manufacturers own valuable equipment that technically speaks software and practically speaks a vendor SDK from 2014, an industrial protocol, three PDF manuals, and whatever Dave meant when he labeled a button “DO NOT PRESS DURING CAL.” Connecting AI agents to that world requires more than a clever tool description.

Why now. On August 27, Anthropic released a research preview of the Model Hardware Standard, an open approach for describing and controlling scientific hardware with AI systems. Giving language models access to robot arms sounds hilarious right up until the robot arm discovers the expensive microscope. What is known: the interface layer is becoming more open. My read: the commercial value will sit in tested drivers, simulation, permissions, fault handling, and evidence that a device behaves safely. The bet: a small studio can build those packages for one neglected equipment category at a time.

Existing solutions. NI LabVIEW, Tulip, OPC UA tooling, integrators, and vendor software already connect machines reliably. Do not replace working control systems to make an AI demo look modern. The gap is a bounded agent-facing layer that exposes approved capabilities while preserving the PLC, instrument driver, or vendor application underneath.

The AI advantage. AI can extract command semantics from manuals, scaffold adapter code, generate simulation scenarios, translate natural-language procedures into candidate plans, and summarize test logs. Safety remains deterministic: ranges, sequences, interlocks, permissions, rate limits, and emergency stops live outside the model. The model can request an action. The safety layer decides whether reality is allowed to care.

Competition moat. Tested drivers, failure-mode libraries, simulated devices, installation knowledge, safety evidence, and trust with equipment owners are difficult to copy. The valuable artifact is not the prompt. It is the boring proof that a command was allowed, bounded, executed, observed, and stopped correctly.

Risk. The market is early, integration is bespoke, and physical mistakes carry real liability. Start with low-energy equipment, experienced operators, visible approvals, and a scope narrow enough to test exhaustively. This is the highest-upside technical bet in the report, but it earns a prototype before it earns a company.

Signals: Anthropic’s Model Hardware Standard research preview, NI’s LabVIEW hardware integration, and Tulip’s connectors and integrations.

Where I would place the bets

• Single best opportunity: The Data Center Community Deal Room.
• Highest-probability business: The DOE M&O Opportunity Radar. The public data, reachable channels, narrow deliverable, and service-first launch make the path unusually clean.
• Highest-upside business: The MHS Driver & Safety Harness Studio. A trusted catalog of safe hardware adapters could become important infrastructure if agent-controlled equipment moves beyond research previews.
• Fastest path to first revenue: Sell a fixed-price DOE matching sprint before building a subscription product.
• Most overlooked: The fact that DOE says roughly 80 percent of its contract dollars flow through management-and-operating and facility-management contractors.
• Best future That SI Guy article: The MHS Studio—essentially “MCP for machines,” except the error state may have momentum.
• Best standalone-company candidate: The Data Center Community Deal Room.

The next move

Take the ten public data-center agreements identified by FAS and turn them into one sample municipal scorecard. Show what each agreement promises, what it measures, what it conveniently leaves vague, and what happens when someone fails to deliver.

Then put that scorecard in front of five municipal managers or public-utility leaders.

Do not ask whether they want an AI platform.

Ask them to bring the ugliest data-center proposal currently sitting on their desk.

That document will tell you what to build.