Building a power strategy: a capstone project

The previous lessons examined each element of data centre power supply: market structures and prices, interconnection, flexibility, renewable procurement, storage and firm clean power, on-site generation, gas supply, contracts, nuclear and emerging sources, and economic comparison. A real power strategy combines them over time, because no single source meets a large campus's needs for speed, cost, reliability and carbon at every stage of its life. This final lesson sets out a framework for building such a strategy and applies it to a capstone project: a 500 MW AI campus that must be energized quickly, grow over five years, and reach a high level of hourly carbon-free energy within a decade.

A framework for power strategy

A power strategy answers five questions, in roughly this order:

• How much, when and how firm? The load forecast and ramp, the reliability required (which may differ between AI training and conventional workloads, DCE-340), and how much of the load could be flexible (Lesson 4).
• Where can power be obtained, and how fast? Grid capacity and the interconnection timeline at candidate sites (Lesson 3), on-site generation options and gas supply (Lessons 7 and 8), and co-location opportunities at existing plants.
• At what cost and risk? Delivered cost of each path, exposure to energy, capacity, fuel and carbon prices, and the hedges available (Lessons 2, 9 and 11).
• How clean, and how credible? Hourly matching targets, the portfolio of renewables, storage and firm clean power that achieves them, and the accounting rules that will apply (Lessons 5, 6 and 10).
• How will it evolve? A sequence of stages, with decision points, that moves from fast initial supply to long-term supply as grid upgrades and clean firm resources arrive.

The strategy is documented as a staged plan with the contracts, permits and investments needed at each stage, the dependencies among them, and the risks and fallbacks.

The capstone project

A developer has secured a 200 ha site in a Canadian province for an AI campus ramping to 500 MW of average load over five years. The provincial utility can provide 150 MW of firm transmission service in year 3, rising to 300 MW in year 4 after a substation expansion, and 500 MW in year 6 after a new transmission line. A gas transmission pipeline crosses the site. The province has a nuclear refurbishment and uprate program, and wind and solar resources are good. The customer, a large technology company, has committed to 90% hourly carbon-free energy by year 10.

Stage 1: Bridging power (years 2 to 5)

The first building is ready in year 2, a year before any grid service. On-site reciprocating engines (chosen for speed of delivery, modularity and part-load efficiency, Lesson 7), connected to the pipeline with firm transport (Lesson 8), supply the first 100 MW, sized N+2 and paired with a battery for transient support and smoothing of AI power swings. The bridging plant is designed to transition to backup duty when grid service is firm, avoiding a stranded asset; air permits are sought for both prime and later emergency operation.

Stage 2: Flexible grid connection (years 3 to 6)

The developer signs an electric service agreement with a ramp matching realistic leasing (Lesson 9), and negotiates a flexible connection: grid service beyond the firm 150 MW and 300 MW stages, interruptible during system emergencies, with the on-site engines and batteries covering curtailment (Lesson 4). Together with the on-site plant, this lets the campus grow to 400 MW in year 5 before the new transmission line is complete; the simplified model below conservatively assumes the extra 100 MW in year 5 comes from on-site generation.

Stage 3: Clean energy procurement (years 3 onward)

Long-term PPAs for wind and solar paired with storage, in the same grid region and contracted with hourly attribute data (Lessons 5 and 6), start in year 3 and grow with the load, combined with the grid's own carbon-free share.

Stage 4: Firm clean supply (year 8 onward)

A 20-year contract for 150 MW from the nuclear uprate program, starting in year 8 (Lesson 10), fills the night-time and low-wind gaps that renewables and batteries cannot economically cover, lifting the hourly CFE score above the 90% target.

# Ten-year power plan for a campus ramping to 500 MW (annual averages; MW of average load)
load = [0, 100, 200, 300, 400, 500, 500, 500, 500, 500]
grid_cap = [0, 0, 150, 300, 300, 500, 500, 500, 500, 500]        # firm grid service available (MW)
nuclear = [0, 0, 0, 0, 0, 0, 0, 150, 150, 150]                    # MW from a nuclear PPA (restart or uprate)
grid_clean_share = 0.35                                           # carbon-free share of grid energy
ppa_clean_mwh = [0, 0, 0.4e6, 0.9e6, 1.4e6, 1.8e6, 1.8e6, 1.8e6, 1.8e6, 1.8e6]   # hourly-matched clean energy from PPAs + storage
cost = {"grid": 78, "onsite": 74, "nuclear": 95, "ppa_premium": 6}   # $/MWh
ef = {"grid": 0.40, "onsite": 0.44}                                # t CO2/MWh
print("year  load  grid  on-site  nuclear   cost $M  CO2 kt  approx CFE   (CO2 is location-based: grid energy at the average grid rate)")
for y in range(10):
    total = load[y] * 8760
    if total == 0:
        print(f"{y + 1:4d}  {0:4d}    construction only"); continue
    nuc = nuclear[y] * 8760
    grid = min(grid_cap[y], load[y] - nuclear[y]) * 8760
    onsite = total - nuc - grid
    c = (grid * cost["grid"] + onsite * cost["onsite"] + nuc * cost["nuclear"] + ppa_clean_mwh[y] * cost["ppa_premium"]) / 1e6
    co2 = (grid * ef["grid"] + onsite * ef["onsite"]) / 1e3
    clean = nuc + min(grid, ppa_clean_mwh[y] + grid * grid_clean_share)
    print(f"{y + 1:4d}  {load[y]:4d}  {grid / 8760:4.0f}  {onsite / 8760:7.0f}  {nuc / 8760:7.0f}  {c:8.0f}  {co2:6.0f}  {clean / total:9.0%}")

year  load  grid  on-site  nuclear   cost $M  CO2 kt  approx CFE   (CO2 is location-based: grid energy at the average grid rate)
   1     0    construction only
   2   100     0      100        0        65     385         0%
   3   200   150       50        0       137     718        49%
   4   300   300        0        0       210    1051        69%
   5   400   300      100        0       278    1437        66%
   6   500   500        0        0       352    1752        76%
   7   500   500        0        0       352    1752        76%
   8   500   350        0      150       375    1226        96%
   9   500   350        0      150       375    1226        96%
  10   500   350        0      150       375    1226        96%

The plan energizes the first building two years earlier than grid service alone would allow, at an energy cost broadly comparable to grid supply. Emissions are highest in the early years, when on-site gas and a partly fossil grid supply most of the load, and fall sharply when nuclear supply begins. The approximate CFE score passes 90% in year 8. Each figure in the model is a simplification (hourly matching, in particular, requires the hourly analysis of Lesson 6), but the staged structure is typical of how large campuses are now planned.

Stage 5: Risks and fallbacks

Presenting the strategy

A power strategy is ultimately a decision document for executives, investors and often regulators and communities. It should present:

• The load forecast and reliability requirements, with their uncertainties.
• The staged supply plan, with a timeline of energization, contracts and investments.
• Costs and risks for each stage, with scenario and sensitivity results rather than single numbers (Lesson 11).
• The carbon trajectory, with hourly CFE estimates and the accounting basis used, and how claims will be verified.
• Commitments to the community and grid: flexibility, payment of costs caused, water and noise impacts of on-site generation, and local benefits (DCE-410).
• Decision points and triggers: what will be decided when, and what would cause the plan to change.

Data centre power has changed from a utility commodity bought on a tariff into one of the most strategic and complex decisions a data centre company makes. The skills of this course, understanding markets, grids, technologies, contracts and economics together, are what make that decision well.

Lessons from recent projects

Recent large campus projects suggest some practical lessons. Equipment availability, especially for large transformers and gas turbines, has repeatedly set schedules more than permits or construction. Communities and regulators have become more demanding about who pays for new generation and transmission, so projects that bring their own resources, accept flexibility and commit to paying costs they cause gain approvals faster. Co-location at existing power plants has attracted regulatory scrutiny over its effect on other customers, making grid-connected arrangements with clear cost allocation more durable. And hourly carbon-free energy claims are being examined more closely, rewarding procurement of new, deliverable, round-the-clock clean resources over annual certificate purchases.