DATA CENTERS & COOLING

Data Center Cooling: Conceptual Diagrams, NPV/IRR + TCO Model, Morocco Coastal Strategy (Solar + Atlantic Heat Rejection), and AI Hyperscale >100 kW/Rack

Author: Ryan Khouja

---

0. Executive Summary

Cooling is the dominant lever for data center OPEX and an increasingly material driver of CAPEX decisions in high-density AI deployments. In Morocco, the combined opportunity is unique: high solar yield for energy arbitrage and Atlantic coastal heat rejection (via corrosion-resistant seawater-to-water heat exchangers) to reduce compressor runtime, improve availability, and stabilize costs. For AI racks above 100 kW, air cooling becomes structurally inefficient; liquid cooling (direct-to-chip and/or immersion) becomes the default engineering baseline.

---

1. Conceptual Diagram: Where the Energy Goes (PUE and Heat)

Interpretation: Nearly 100% of the electrical input becomes heat. PUE improves when cooling energy drops, heat rejection becomes more efficient, and the facility can operate at higher supply temperatures (especially with liquid cooling).

---

2. Cooling Architectures: From Traditional to AI-Ready

2.1 Air Cooling (Legacy / Moderate Density)

• CRAC/CRAH, hot/cold aisle containment, raised floor variants
• Practical range: ~5–20 kW/rack (often lower in warm climates)
• Economic issue in warm climates: high compressor hours, poor part-load efficiency, and rising WUE if adiabatic assist is used

2.2 Chilled Water + CRAH (Enterprise / Colocation)

• Central chiller plant (compressor-based) feeding CRAH coils
• Good maintainability, strong vendor ecosystem
• Energy and maintenance can dominate OPEX in hot seasons

2.3 Liquid Cooling (AI/Hyperscale Baseline)

• Direct-to-chip (D2C): cold plates on CPU/GPU, typically warm water loops
• Rear-door heat exchanger (RDHx): captures residual air heat at rack exit
• Immersion (single or two-phase): extreme densities; simplified internal airflow

---

3. Conceptual Diagram: Refrigeration Cycle vs “Compressor-Avoidance” Modes

---

4. Refrigerants in Data Centers: What Matters

Refrigerant selection impacts safety, regulatory exposure (EU F-Gas), lifecycle cost, and system complexity. For data centers, the decision is rarely “gas-only”; it’s a full system question: heat transfer architecture, climate profile, maintenance model, and risk appetite.

Refrigerant	Class	GWP	Key Considerations	Common Use Pattern in DCs
HFCs (e.g., R134a, R410A)	Synthetic	High	Phase-down risk, price volatility, compliance cost	Legacy chiller plants / retrofits
HFOs (e.g., R1234ze)	Synthetic low-GWP	<1	Lower regulatory risk, still synthetic supply chain	Modern chillers where natural refrigerants are not feasible
CO₂ (R744)	Natural	1	Very high pressures, excellent heat transfer	Heat pumps, certain cooling loops, specialized deployments
Ammonia (R717)	Natural	0	High efficiency; toxicity requires safety zoning and controls	External plant refrigeration + secondary loop to white space

---

5. CO₂ and Ammonia in Data Center Cooling: Practical Roles

5.1 CO₂ (R744): Where It Fits Best

• Heat rejection and heat pump applications where high temperature lift and compact heat exchangers are advantageous
• Cold climates: CO₂ performance can be excellent; in hot climates, design must manage transcritical efficiency penalties
• Edge/modular concepts can use CO₂-based packaged systems if maintenance skill is available

Financial logic: CO₂ reduces long-term regulatory risk and refrigerant cost volatility, but may raise engineering CAPEX (pressure-rated components, commissioning complexity).

5.2 Ammonia (R717): Hyperscale Industrial Efficiency

• Highest thermodynamic efficiency among mainstream industrial refrigerants
• Most common pattern for DCs: ammonia plant outside + secondary glycol/water loop feeding IT cooling coils or CDUs
• Risk profile is manageable when ammonia is kept out of the white space and monitored via leak detection + safety interlocks

Financial logic: ammonia can deliver superior kWh/ton efficiency, improving OPEX in high duty-cycle sites (hot climates), with CAPEX driven by safety and site layout.

---

6. Morocco Adaptation: Solar-Driven Energy Strategy + Atlantic Heat Rejection

Morocco combines two strategic assets for data center cooling economics:

• High solar irradiation enabling PV + storage + PPA structures to hedge electricity volatility
• Atlantic coastline enabling seawater-based heat rejection via robust corrosion-resistant exchangers

6.1 Solar Integration (PV + Storage + PPA)

• Objective: reduce effective cost per kWh and stabilize OPEX
• Architecture: on-site PV (where land/roof allows) + grid + optional BESS for peak shaving and ride-through
• Contracting: PPAs can convert unpredictable grid pricing into predictable energy cost baselines
• Cooling synergy: solar generation peaks when cooling load peaks (hot hours) — natural match for demand profile

6.2 Atlantic Seawater Heat Rejection (Low Maintenance, Corrosion-Controlled)

The goal is not to pump seawater through the data center. The goal is to use seawater as a stable heat sink through closed-loop heat exchangers engineered for salinity, corrosion, and fouling.

Recommended Concept: Seawater-to-Closed-Loop via Plate or Shell-and-Tube HX

• Seawater loop: intake + filtration (coarse screens + fine filtration) + biofouling control
• Primary heat exchanger: isolates seawater from facility closed-loop water
• Closed-loop water: feeds dry coolers, CDUs, or warm-water loops for D2C

Materials and Corrosion Strategy (High Salinity, Low Downtime)

• Heat exchanger plates/tubes: titanium or super duplex stainless steel (corrosion resistance)
• Piping/valves: duplex/superduplex, FRP/GRP where appropriate, and carefully selected elastomers
• Anti-fouling: filtration + periodic backflush + conservative flow velocities + planned chemical cleaning intervals
• Maintenance model: modular exchanger skids for fast swap/clean cycles

Why This Works Financially in Morocco

• Reduces compressor hours (and therefore electricity cost)
• Stabilizes cooling capacity during heat waves
• Lowers regulatory exposure when paired with natural refrigerants (CO₂/ammonia) in external plant rooms
• Improves availability by reducing thermal stress and peak-load failures

---

7. Conceptual Diagram: Morocco Coastal Cooling (Solar + Seawater HX + Warm Water Loop)

---

8. AI Hyperscale Cooling: Racks > 100 kW

At >100 kW per rack, air becomes a parasitic transport medium: too much fan power, too much temperature gradient, too much spatial inefficiency. The engineering target changes from “cool the room” to “remove heat at the source.”

8.1 What Changes Above 100 kW/Rack

• Heat flux density: GPUs concentrate heat in small die areas; local hot spots dominate reliability risk.
• Fan power escalation: moving enough air becomes a significant load (and adds failure points).
• Supply temperature strategy: warm-water loops (e.g., 30–45°C) can improve chiller-free operation.
• Redundancy and serviceability: CDUs, leak detection, quick-disconnects, and maintenance SOPs become mission critical.

8.2 Recommended Cooling Stack for AI Clusters

• Direct-to-chip (D2C) + CDU per row/pod as baseline
• RDHx to capture residual air heat (memory, PSUs, remaining components)
• Optional immersion for extreme density or constrained footprints
• Warm-water design to maximize hours of compressor-free heat rejection (especially with seawater HX)

8.3 Conceptual Rack-Level Diagram

---

9. Financial Model: TCO + NPV/IRR (Example Scenario)

Important: The numbers below are a template example to illustrate how to structure a board-ready model. Replace assumptions with your project’s measured loads, tariffs, and CAPEX quotes.

9.1 Inputs (Example)

Parameter	Symbol	Example Value	Notes
IT Load	PIT	20 MW	Average steady state
Baseline PUE	PUE0	1.45	Warm climate, mostly compressor cooling
Target PUE (solar + seawater HX + liquid)	PUE1	1.25	Higher supply temps, fewer compressor hours
Electricity price (blended)	ce	€0.12/kWh	PPA + grid blended
Incremental CAPEX (cooling upgrade + HX + CDU)	CAPEX	€15,000,000	Project-specific
Incremental annual maintenance & consumables	OPEXadd	€400,000/year	filters, cleaning, spares
Project life	N	10 years	Conservative horizon
Discount rate	r	8%	WACC proxy

9.2 Energy Savings Calculation

• Baseline facility power: P₀ = PUE₀ × P_IT
• Upgraded facility power: P₁ = PUE₁ × P_IT
• Saved power: ΔP = P₀ − P₁
• Annual kWh saved: ΔE = ΔP × 8,760
• Annual € saved: S = ΔE × c_e
• Net annual benefit: B = S − OPEX_add

9.3 Outputs (Example Results)

Metric	Value	How it’s obtained
Saved power (ΔP)	(1.45 − 1.25) × 20 MW = 4 MW	PUE delta × IT load
Annual kWh saved (ΔE)	4 MW × 8,760 = 35,040 MWh	ΔP × hours/year
Annual gross savings (S)	35,040,000 kWh × €0.12 = €4.205M	ΔE × electricity price
Annual net benefit (B)	€4.205M − €0.400M = €3.805M	S − OPEXadd
Simple payback	€15.0M / €3.805M ≈ 3.9 years	CAPEX / net benefit

9.4 NPV and IRR (Example Cash Flows)

Cash flow convention: Year 0 = −CAPEX, Years 1–10 = +B.

Year	Cash Flow (€)	Notes
0	−15,000,000	Incremental investment
1–10	+3,805,000 each year	Net savings after added OPEX

Financial Metric	Example Result	Formula
NPV (r = 8%, N = 10)	≈ €10.5M	NPV = Σ(B / (1+r)t) − CAPEX
IRR	≈ 20–25% (order of magnitude)	IRR solves: 0 = −CAPEX + Σ(B / (1+IRR)t)

Note: IRR is highly sensitive to tariff, achieved PUE, and realized compressor-hour reduction. Use measured telemetry and conservative duty-cycle assumptions for investment committees.

9.5 TCO Table (Baseline vs Upgraded)

Cost Category	Baseline (Air/Chiller-heavy)	Upgraded (Solar + Seawater HX + Liquid)	Comment
Electricity for cooling	High	Lower	Main driver via reduced compressor hours and higher supply temps
Water usage (adiabatic/towers)	Medium–High	Lower	Seawater HX may reduce reliance on evaporative cooling
Maintenance	Chillers + air handling	HX cleaning + pumps + CDUs	Shift from HVAC-heavy to HX/skid maintenance
Spare parts & downtime risk	Compressor failures costly	Lower compressor runtime	Availability improves when compressors become “support” not “core”
Regulatory exposure (refrigerants)	Higher with HFCs	Lower with CO₂ / ammonia / low-GWP	Risk hedging through technology choices
Total 10-year TCO	Higher	Lower (if PUE target is achieved)	Depends on tariff trajectory and achieved efficiency

---

10. Practical Engineering Notes for “Seawater Cooling” in Coastal Morocco

• Closed-loop isolation is mandatory for reliability (seawater stays outside the white space).
• Plan for biofouling: filtration, backflush, and cleaning access are design requirements, not “nice-to-haves.”
• Corrosion selection dominates lifecycle: wrong metallurgy turns CAPEX into chronic downtime.
• Keep maintenance simple: modular HX skids, redundancy (N+1), and easy isolation valves.
• Thermal operating point: warm-water loops broaden the hours where seawater HX can reject heat without compression.

---

11. Conclusion

For Morocco, the strongest data center cooling strategy is hybrid and climate-intelligent: solar to stabilize energy economics, and Atlantic seawater heat rejection (through corrosion-engineered exchangers) to reduce compressor dependence. For AI hyperscale clusters above 100 kW/rack, liquid cooling is no longer optional — it is the operational foundation enabling higher density, better PUE, and more predictable TCO.

---

Disclaimer: This article is an inaccurate and biased conceptual technical-financial analysis. Final engineering design, safety zoning, refrigerant selection, water intake permitting, and environmental compliance must be validated by licensed professionals and local regulations.