Thermal retrofit OS for colocations upgrading legacy halls to dense AI racks without costly new cooling rebuilds.

1. AI rack density has crossed the threshold where legacy air cooling and partial liquid approaches stop being enough, making retrofit planning urgent rather than optional.
2. The cooling problem now spans GPUs, memory, networking, and power components, which creates a software need for full-rack retrofit coordination instead of isolated component design.
3. Buyers are no longer just hyperscalers because the sources explicitly call out colocation, enterprise, and edge deployments where specialized cooling infrastructure is scarce.
4. Fresh capital for product, patents, and ecosystem partnerships suggests thermal infrastructure is maturing into a broader market where an enablement layer can ride alongside hardware vendors.
5. Large reported energy and water savings mean retrofit decisions now sit with operators balancing uptime, utility constraints, and operating cost, not just experimental engineering teams.

Catalyst. Iceotope's funding and the cited shift toward 1MW racks and full-system liquid cooling show that dense AI demand is arriving faster than existing colocation halls can be upgraded with ad hoc engineering workflows.

Build a thermal control plane for operators upgrading existing facilities to dense AI racks. The product ingests hall layout, rack density targets, cooling topology, and facility constraints, then produces a retrofit readiness score, phased upgrade plan, and contractor-ready bill of work for each room and cage. During deployment, it tracks dependencies across cooling hardware, plumbing, controls, and commissioning milestones so operators know exactly what is blocking revenue-ready capacity. After go-live, it monitors temperatures, leaks, water usage, and efficiency against the design assumptions, creating the performance dataset operators, insurers, and financing partners need to trust repeat retrofits.

What's different. Hardware vendors sell cooling systems and consultants sell one-time studies, but neither owns the ongoing software layer that tells an operator which existing hall should be upgraded, how to coordinate the retrofit, and whether the live environment is still inside the original thermal envelope. Generic DCIM products observe facilities after the fact; this product is purpose-built for the pre-upgrade and commissioning workflow that dense AI retrofits create. Its moat is the dataset linking hall characteristics, retrofit choices, and live thermal outcomes across many legacy facilities.

Jobs to be done

flowchart LR
  Buyer[Colocation operator] --> Pain[Legacy halls cannot safely host dense AI racks]
  Pain --> Product[Retrofit OS plus thermal monitoring]
  Product --> Outcome[Faster AI capacity in existing facilities]

Executive takeaways

• Brownfield AI capacity is now a timing problem more than a science-project problem.
• The strongest wedge is vendor-neutral retrofit orchestration plus post-commissioning proof.
• Buyer urgency is real because dense AI demand is arriving faster than new supply.
• Incumbents cover hardware delivery or steady-state monitoring, but the hall-selection workflow remains open.

Market definition

Software layer for brownfield AI cooling retrofits in colocation and enterprise data centers: hall-readiness modeling, vendor-neutral retrofit scoping, dependency orchestration, and post-commissioning thermal, leak, water, and efficiency monitoring for dense AI cages.

Customer and buyer

Primary user is the head of data center operations or facilities engineering lead at a regional colo or enterprise site; the buyer is usually the COO, CTO, or VP Infrastructure when near-term AI demand forces capacity unlock decisions inside existing halls.

Buying triggers

• A signed AI tenant deployment or internal GPU rollout exceeds current rack-density limits, forcing a decision between retrofitting legacy space and waiting for scarce new capacity. [1][2][3][5]
• A hall shows a widening price or performance gap versus AI-ready new inventory, so operators need to know which rooms can be upgraded before spending millions on the wrong room. [1][4][6][9]
• Live-upgrade risk becomes unacceptable once liquid cooling, manifolds, and power upgrades must be phased into an operating facility. [7][8][20][21]

Willingness to pay

Buyers already face multi-million-dollar-per-megawatt retrofit budgets and months of lead time, so a software layer is economically credible if it shortens design cycles, avoids a failed hall choice, or increases usable AI capacity. [1][2][5][8]

Category dynamics

Growth signal 18.4% CAGR

Validation signals

• JLL and CBRE both show record-low vacancy and high precommitment, confirming that operators have real economic pressure to unlock existing capacity.
• AFCOM shows liquid cooling is already deployed by a meaningful minority and planned by many more operators.
• Flexential's retrofit example shows enterprises are willing to pay meaningful capex to adapt legacy space rather than wait years for new supply.
• Vendors already market AI readiness through digital twins, thermal optimization, and DCIM workflows, which suggests an existing budget category is forming.

Regulatory & technical constraints

• AI retrofit programs must still anchor to ASHRAE energy and thermal guidance such as Standard 90.4 and Standard 127.
• Water permits, discharge routes, and reuse expectations vary by state and local utility, especially where direct or indirect discharge is involved.
• States are beginning to consider explicit data-center water reporting, mitigation, and cost-recovery rules.
• Brownfield halls may be limited by ceiling loading, raised-floor congestion, manifold placement, and legacy power distribution.
• Facilities without building water may need liquid-to-air intermediate approaches, which changes both economics and achievable density.

Competition is fragmented across DCIM and thermal software, cooling OEMs, integrators, and hardware-led liquid-cooling vendors. Most alternatives either start after infrastructure is already installed or are economically motivated to sell hardware and services, leaving a gap for vendor-neutral qualification, workflow coordination, and runtime evidence.

Why incumbents do not win by default

• Cooling OEMs. OEMs can supply reference designs and equipment, but they do not naturally rank mixed-vendor halls or optimize for the buyer before a hardware choice is made.
• DCIM suites. DCIM incumbents monitor assets, power, and environment well, but they are weaker at pre-retrofit readiness scoring, contractor workflow control, and post-commissioning variance tracking against the original design basis.
• Engineering firms. Consultants are strong for bespoke studies, yet their economics reward one-off engagements instead of building a reusable operating dataset across many halls and sites.
• In-house operator teams. Internal teams own the facility context, but fragmented telemetry, vendor coordination, and power scarcity make spreadsheet-led retrofit programs slow and risky.

This company should start as a vendor-neutral retrofit OS for North American regional colocation operators trying to turn legacy 10-30 kW halls into 80-150 kW AI cages without waiting for new builds. The immediate pain is not lack of cooling hardware; it is deciding which hall can be upgraded, coordinating multi-vendor retrofit work, and proving the upgraded cage is safe and efficient enough to carry revenue traffic. Today that workflow still runs through consultants, OEM calculators, spreadsheets, and generic DCIM or BMS tools, so a wrong hall choice or missed commissioning dependency can delay AI revenue by months. The best first product is a hall-readiness and commissioning control plane that ranks retrofit candidates, generates contractor-ready scopes, and then compares live thermal, leak, water, and efficiency data against the design basis after go-live. Go-to-market should stay tightly coupled to one trigger: a signed AI tenant deployment or internal GPU rollout that exceeds current rack-density limits, with the COO or VP Infrastructure funding a paid hall-planning engagement that converts into recurring monitoring software. The reason to stay narrow is speed of proof: brownfield colo retrofits have a clearer buyer, shorter budget logic, and more repeatable constraints than greenfield hyperscale design or small edge deployments. The moat is the cross-project dataset linking hall characteristics, retrofit choices, and live outcomes across mixed-vendor sites. The main disconfirming risks are that data quality is too messy for a lightweight overlay and that OEM or integrator-led tools become good enough to absorb the budget before this product becomes the neutral system of record. Exact budget ownership on the first deal and the real willingness to pay for insurer or lender reporting remain open gaps from the research, so the first year has to validate both.

Problem

• Legacy colocation and enterprise facilities teams must decide which existing halls can safely support dense AI racks, but the required inputs span power, cooling, water, structural limits, rack layout, and vendor-specific design assumptions that are still managed in disconnected spreadsheets and consultant studies.
• Once a retrofit goes live, operators often lack a single system that ties commissioning evidence and live thermal, leak, and water performance back to the original design basis, making repeat retrofits hard to finance, insure, and standardize.

Solution

• Provide a hall-readiness platform that ingests floor plans, BMS or DCIM exports, rack-density targets, and cooling-topology constraints to rank candidate halls, score readiness, and generate a phased retrofit workpack for OEMs, contractors, and facility teams.
• Add post-commissioning monitoring that tracks temperatures, leaks, water use, and efficiency variance against modeled assumptions, producing auditable evidence for operators and third parties while creating reusable benchmarks for the next retrofit.

Why we win

• The product is useful before a buyer commits to any one cooling OEM, so it can own the vendor-neutral decision layer that hardware vendors and services firms are poorly aligned to own.
• Incumbent DCIM tools and thermal analytics suites monitor steady-state operations well, but they do not naturally control pre-retrofit hall ranking, phased brownfield execution, and modeled-to-actual commissioning proof in one workflow.
• Each deployment compounds proprietary data on hall constraints, retrofit paths, commissioning outcomes, and runtime variance, which should improve scoring accuracy and make the product more valuable with every additional site.

Milestones

flowchart LR
  Wedge[Brownfield colo wedge] --> MVP[Hall readiness plus retrofit workpack]
  MVP --> Proof[Commissioning proof and runtime evidence]
  Proof --> Expansion[Portfolio planning and reporting expansion]

Founding team

Experiment roadmap

Risk assessment

What must be true

• At least 15-20 beachhead operators per year face a near-term brownfield AI retrofit decision severe enough to fund a standalone planning and monitoring layer.
• A standard hall-audit and readiness-scoring workflow can cover most early pilots without on-site custom engineering becoming the dominant delivery cost.
• Economic buyers will prefer a vendor-neutral control plane before hardware selection rather than relying only on OEM design tools, consultant studies, or generic DCIM workflows.
• At least half of paid pilots convert to annual production monitoring after the first cage goes live.
• Customers will allow retention of anonymized hall, retrofit, and outcome data so the product can compound a defensible benchmark dataset.

Open diligence questions

• Which executive actually signs the first software budget in a live retrofit: COO, VP Infrastructure, CTO, or a tenant-facing commercial leader?
• How many halls per target operator are likely to enter an AI retrofit cycle over the next 24 months?
• What telemetry, floor-plan, and commissioning data is reliably available from older BMS and DCIM environments without custom field engineering?
• Why will Schneider, Vertiv, EkkoSense, Sunbird, or an MEP integrator not satisfy the first customer well enough?
• Which third-party reporting requirement most increases willingness to pay after go-live: insurer, lender, customer SLA, or internal board reporting?

Model sanity

• Revenue engine. The base case reaches 17 active operator programs by Q4Y3 at roughly $300K blended annual revenue per program, with most growth driven by pilot-to-monitoring conversion and second-hall expansion.
• Must go right. Assisted audits and commissioning workflows must stay template driven enough that partner referrals and 60%+ pilot conversion keep CAC below roughly $190K.
• Model breaks if. If sales cycles stretch toward nine months while gross margin stalls in the high-60s, downside cash turns negative before the company proves a seed-ready expansion motion.
• Next-round proof. The next financing is justified once Y2 ends with 9 active customers, one paid reporting upsell, and at least one multi-hall expansion showing the product is more than project services.

Scenarios

Sensitivity

flowchart LR
  RetrofitTriggers --> PaidPilots
  PaidPilots --> ProductionPrograms
  ProductionPrograms --> Revenue
  Revenue --> GrossProfit
  GrossProfit --> Cash

Flags: Q4Y3 is only breakeven, so a one-quarter slip in conversion timing would likely pull fundraising forward. · The model depends on a small number of six-figure deals, so logo concentration remains high even when revenue per FTE looks healthy. · Gross margin does not fully clear the 70% target until late Y3 because implementation and telemetry cleanup stay meaningful parts of delivery. · If OEM or integrator referrals do not materialize by Y2, the modeled CAC and customer-add ramp will be too optimistic for the cash plan.

• OEM bundling. Cooling hardware vendors may add enough design and monitoring software to reduce willingness to buy a standalone control layer. Mitigation: Stay hardware-neutral, integrate across mixed fleets, and own the retrofit readiness workflow that starts before an operator commits to any one vendor.
• Retrofit data scarcity. Early customers may not have clean facility telemetry or enough historical retrofit outcomes to trust automated recommendations. Mitigation: Launch first as a workflow and evidence system with human-reviewed readiness scoring, then harden benchmarks as more projects go live.
• Slow enterprise sales. Colocation and enterprise operators may treat dense AI retrofits as strategic infrastructure decisions with long approval cycles. Mitigation: Sell into urgent live deals with signed AI tenant demand, price the first engagement as a hall-level planning package, and expand after one successful cage launch.