HD Hyundai Power Costs About 8.4% More per MW • The Weighted Average

HD Hyundai Heavy Industries signed a $673.8 million contract to supply 1,000 MW of engine generation for U.S. data centers, equal to $673,800 per ordered MW. Using HD Hyundai’s rounded $425 million value and Yonhap’s 684-MW capacity for the April order gives about $621,345 per MW, making the new disclosed value approximately 8.4% higher—a warning that behind-the-meter power needs scope-adjusted pricing, not a generic dollars-per-megawatt shortcut.

One gigawatt gets a visible equipment price

The company’s August 9 announcement calls this its largest power-generation engine order. Corban Energy Group will receive systems based on 9.6-MW HiMSEN engines for a major U.S. technology company’s data centers. HD Hyundai’s own engine release confirms HiMSEN is designed for land-based power generation, although that page describes a different ammonia-capable model rather than the Corban configuration. The release describes continuous 24/7 operation but does not name the end customer, sites, delivery dates, fuel arrangement, emissions profile, financing, maintenance price, or commissioning guarantees.

The arithmetic creates a diligence anchor. Divide $673.8 million by 1,000 MW and the disclosed order value is $673,800 per MW. Dividing capacity by the stated 9.6-MW engine rating implies roughly 104 engine-equivalents—1,000 ÷ 9.6—but actual unit count could differ because the release does not provide configuration details. The calculation is a scale translation, not a bill of materials.

A prior order supplies the comparison. HD Hyundai’s August release describes its April agreement as $425 million, while Yonhap’s contemporaneous report gives 684 MW and converts 627 billion won to $424 million. Combining the company’s rounded dollar baseline with Yonhap’s capacity gives about $621,345 per MW and an 8.4% premium: ($673,800 ÷ $621,345) − 1. Using Yonhap’s $424 million conversion instead produces about 8.7%; the 0.3-point spread reflects currency and disclosure rounding, so the headline remains approximate.

That premium is not evidence of inflation or richer margins. Contract scope may differ in controls, auxiliaries, logistics, installation support, warranty, service, delivery urgency, engine mix, currency timing, and site conditions. The announcements do not provide enough detail to normalize those variables. The useful conclusion is contractual: a buyer comparing onsite-power proposals should demand scope per deliverable MW before interpreting the headline value.

The market is moving because the grid cannot always move first. The Electric Power Research Institute estimates data centers at 4–5% of U.S. electricity demand today and 9–17% by 2030. It places 2024 data-center load near 184 TWh and projects 14–27% annual growth through 2030 across scenarios. Large campuses are therefore shopping for generation, switchgear, transformers, storage, and grid service simultaneously.

HD Hyundai is selling into both sides. A separate July framework agreement covers up to 1.1212 trillion won of North American distribution and high-voltage equipment through 2028. The Korea Times reported the package at about $720 million, while noting that purchase orders will follow construction schedules. That package approach reduces compatibility and schedule interfaces but increases concentration in one supplier family. The new engine order extends the same logic from power delivery into generation.

For developers, this changes the buy-versus-wait calculation. A multi-year utility interconnection delay can make onsite generation economically rational even when equipment cost, fuel, maintenance, permits, and carbon exposure exceed ordinary grid service. But one equipment contract does not make power “secured.” The archive’s Texas 474-GW queue audit showed why queue position now needs a proof package, while Amazon’s GW Ranch analysis showed how private generation simply relocates risk to permits, fuel, emissions, and commissioning.

Today’s Claude Code governance lead makes the same structural point in software: removing an obvious bottleneck does not remove the control problem. It changes where the operator must enforce it.

Contract for delivered power, not engine nameplate

Data-center developers considering onsite generation should convert vendor quotes into four comparable layers. First is equipment: engines, generators, controls, switchgear, auxiliaries, cooling, and emissions equipment. Second is delivery: logistics, civil work, installation, interconnection, testing, and commissioning. Third is operation: fuel, variable maintenance, overhaul intervals, staffing, spares, and availability guarantees. Fourth is risk: permits, noise, emissions, water, community opposition, and schedule remedies.

The disclosed $673,800 per MW sits somewhere inside those layers, but the release does not say where. Procurement should request a responsibility matrix and price each omission. A cheaper engine-only quote can become more expensive once the owner absorbs integration. A higher turnkey quote can still be attractive if delay damages are real and performance guarantees are enforceable.

Capacity should also be phased. One gigawatt arriving as a single commercial milestone creates correlated schedule risk. Tie payment and take-or-pay obligations to commissioned blocks, demonstrated output, emissions compliance, fuel availability, and black-start or ride-through tests where required. Keep grid, storage, renewable, and demand-response paths alive rather than treating reciprocating generation as a permanent binary choice.

Reliability claims need denominators. “24/7” describes intended duty, not achieved availability. Ask for equivalent forced-outage rate, maintenance intervals, derating at site conditions, ramp behavior, minimum stable load, and spares lead time. Model N+1 or N+2 configurations and the fuel infrastructure needed to survive upstream disruption. The implied 104 engine-equivalents can create redundancy, but also a maintenance fleet.

Carbon and local permitting are not footnotes. Combustion generation can shorten an electrical queue while lengthening air, land-use, and community review. Buyers with emissions covenants should require hourly fuel and output data, control guarantees, ownership of environmental attributes, and remedies if actual performance exceeds agreed intensity. An undisclosed end customer cannot rely on a sustainability narrative written for another site.

The strongest counterpoint is that equipment-value-per-MW is too crude to compare. That is correct. The April and August contracts name different developers, capacity, timing, and possibly scope. The 8.4% gap is a diligence signal, not a market index. If detailed schedules show the later order includes more integration or longer service, the premium may be entirely rational. If scope is identical, it may reveal tighter supply or faster delivery.

Evidence that would change the verdict includes delivery schedules, unit counts, fuel and emissions design, service coverage, guaranteed net output, availability, and customer identity. A fully turnkey, performance-backed contract would make the higher value more compelling. An engine-only purchase with long owner-supplied integration would weaken it.

The operator move is clear:

Developers should normalize bids by net commissioned MW and a complete responsibility matrix, not nameplate alone.
Procurement should price schedule remedies by phased blocks and withhold payment until performance and permit conditions clear.
Finance should model the whole stack: equipment, integration, fuel, maintenance, emissions, grid backup, and stranded-capacity risk.
Sustainability teams should require hourly evidence and contractually assign emissions, reporting, and environmental attributes.
Change the verdict with scope disclosure: the 8.4% premium is informative only after like-for-like deliverables are known.

The new order is a real marker: a gigawatt of data-center generation now carries a publicly visible nine-figure equipment contract. It is not yet a power price. Builders should use the disclosure to ask sharper questions, then pay for commissioned electrons rather than a round number on a press release.