Modelled 10MW data center could save $120,000 per megawatt a year with HVDC architecture

The person signing the purchase order for an 800V DC backbone can check half of this pitch on their own electricity bill. A typical modelled 10MW deployment saves well over $1.2 million a year in operating cost, the column says ^[6]. That is $120,000 per MW per year ^[13]. The 100MW case is given as over $130 million across a decade ^[7], which works out to about $130,000 per MW per year ^[14], roughly 8% above the smaller model ^[15]. The saving scales close to linearly with load here, and its source is conversion stages: four or more in a typical AC architecture, two with an HVDC path ^[5]. The column also credits a modest hyperscale facility with efficiency improvements of several percentage points ^[19].

The claim about the utility's calendar works differently. The column argues that operators should shift from "a mindset that asks for permission to a mindset that offers a solution" ^[16], and rests that argument on EPRI's Powering Intelligence and Flex MOSAIC work ^[4]. No operator is named as having reached energization sooner this way, and no figure is given for how much sooner. The column says a growing number of operators are beginning to explore battery storage and microgrid principles in site design ^[11].

Operators wait in the utility queue, or ask an already strained network for an immediate connection as a favor ^[2]. The fast alternative, a standard diesel generation block, gets refused, because utilities see volatility and emissions with little systemic value to the local network ^[3]. So the realistic comparison is a grid-asset build against waiting.

Does the loss saving pay back the DC and storage gear at your load and your power price? And does your utility have a live way to pay for flexibility today, a tariff or a program with a number in it? Yes and yes, and you buy it now. Yes and no, and you buy it for the losses and treat any earlier connection as a free option. No and yes, and you are buying a grid product, so price it against what the utility will actually pay. No and no, and the case rests entirely on the column's argument that projects are stalling because operators present themselves as passive consumers ^[17].

Where the flexibility has value is where transmission is already bottlenecked. The column points to ERCOT in Texas and the US Midwest, where renewable output is frequently curtailed ^[9]. A DC-native site can absorb that power into storage and feed surplus onsite generation back ^[8][18]. What a utility will pay for it is the open item: Flex MOSAIC was developed with utilities, system operators and hyperscalers, and the column describes such frameworks as pointing toward "a more structured way of defining and valuing load flexibility" ^[10].