The secondchapter is heat

In 2012, MGHPCC was certified LEED Platinum, the first university research data center in the United States to achieve that mark. The "green" in the consortium's name was earned on two structural facts: the carbon-free hydroelectric supply from Holyoke Gas and Electric's Hadley Falls station, and a leading-edge PUE in the 1.2 to 1.3 range. Both achievements remain, and both are durable. They have also been the entire sustainability story for fourteen years.

The next chapter of that story is heat. ERE, Energy Reuse Effectiveness, is the published metric for waste-heat reuse, and it sits alongside PUE in modern data center sustainability frameworks. The existing MGHPCC architecture has no published ERE figure because there is no waste-heat offtake; all of the compute waste heat is rejected to atmosphere via rooftop and yard-side cooling. MicroLink's three-loop architecture, with the Loop 3 host termination at a thermal offtake, is engineered for ERE below 0.5 with tri-gen. This is a step change in the consortium's sustainability metrics and a credible LEED narrative for any new build on the undeveloped acreage.

Why heat is the right frame for the state-side audience

The carbon-free electricity argument is durable, but it is also fixed. HG&E's portfolio has been roughly 87 percent owned hydro for many years; the consortium has been on that supply since 2012; the 100 percent carbon-free framing in the May 2025 MassTech announcement is true but not new. The Healey administration's affordability narrative, the Mass AI Hub's "responsible AI" framing, and the state's economic-development push under Secretary Paley all need a forward-looking sustainability story for AI compute beyond what is already on the press release shelf.

Waste heat is that story. Massachusetts has the highest-cost retail electricity rates in the continental United States, average residential rates near 28 cents per kWh in 2025, and a structurally constrained grid that is now under additional pressure from AI compute load. A state-anchored ERE-below-0.5 demonstration at MGHPCC tells a different story: AI compute can be a thermal asset to its host community, not just a load on the grid. This is the lede the state-side communications team has been waiting for.

The metric pairing that matters

ERE is calculated as (Total Facility Energy minus Reuse Energy) divided by IT Equipment Energy. A facility with no heat reuse has ERE equal to PUE. A facility that exports half of its rejected waste heat as a usable thermal product reports ERE substantially below PUE. The MicroLink target of ERE below 0.5 with tri-gen represents net energy efficiency that no other research data center in the United States reports publicly at scale.

For Mass AI Hub communications, this is the headline metric of the next press release. For MGHPCC's LEED narrative, it is a credible path to a fresh certification on the new build. For EPRI DCFlex, it is the thermal-side complement to Emerald Conductor's load-side flexibility, and the basis for the first stacked thermal-and-load demonstration at academic-research scale in the United States.

MGHPCC's site at 100 Bigelow Street sits in the heart of Holyoke's historic industrial Flats, three blocks from the Connecticut River, one block from the Holyoke Canal System, and within walking distance of the city's mill-redevelopment district and Innovation District identity. The infrastructure that defined Holyoke since 1849, paper mills and canal-driven hydropower and dense industrial heat use, is exactly the infrastructure that a heat-recovering AI compute facility can plug back into.

Mayor Garcia, OPED, and the Holyoke reform window are aligned

Mayor Joshua A. Garcia's second-term agenda explicitly includes zoning and special-permitting reforms to make Holyoke competitive in the commonwealth. Eric Nakajima, appointed Director of the Office of Planning and Economic Development in early 2026, brings a Berkeley city-planning background, prior MBI directorship experience, and a clear early-tenure mandate for process efficiency on industrial and innovation-district projects. The political and procedural conditions in Holyoke are unusually well-aligned with a clean, well-organised containerised deployment proposal at MGHPCC.

The Holyoke Redevelopment Authority holds urban renewal authority under M.G.L. Chapter 121B over the Center City Renewal Plan area where MGHPCC sits, and HRA endorsement is procedurally meaningful for site-plan or special-permit work on the undeveloped acres. The Holyoke 2030 plan, the city's climate action plan, and the Innovation District designation all align with the MicroLink narrative; this is the kind of development the city's own planning documents anticipate, not a project the city's planning architecture would receive as a surprise.

HG&E as the utility partner and the institutional memory

Holyoke Gas and Electric has been the municipal utility since 1902 and the operator of the Hadley Falls hydroelectric station since 1859. James Lavelle has been General Manager since 2004, sits on the boards of MMWEC and NEPPA, served as the 2025 Grand Marshal of the Holyoke St. Patrick's Parade, and is the institutional memory of every major HG&E and MGHPCC engagement. His relationship to the consortium is durable; the substation provisioning of 30 MW at the North Canal Substation in 2011 to 2013 happened under his direct sponsorship.

HG&E's owned generation portfolio is approximately 87 percent hydro by capacity: Hadley Falls at 33 MW nameplate plus the smaller canal-side hydro plants, the 5.76 MW Mount Tom Solar array (commissioned 2018 on the site of the former Mount Tom coal plant, with 3 MW / 6 MWh of co-located battery storage), and an interest in regional carbon-free supply through the Massachusetts Municipal Wholesale Electric Company. The 2024 supply mix reports 37.4 percent RPS-qualifying renewables HG&E owns or retires RECs for, plus 49.7 percent additional carbon-free generation without REC ownership, plus 12.9 percent Null Energy. HG&E rates are roughly 30 to 50 percent below MA investor-owned utility rates and are the most-cited reason businesses and adaptive-reuse developers are choosing Holyoke.

The Holyoke district heat aggregator that does not yet exist

Holyoke does not have a city-scale district heating system today. There is an HG&E "district steam" service mentioned in utility About materials, with scope and customer list not in the public record. There is no Stockholm-scale thermal loop. What Holyoke does have is a dense set of plausible thermal offtake candidates within a 1 km (0.6 mi) radius of the MGHPCC parcel: the Hazen Paper Company (active paper converter on hydro and solar power, real process-heat demand), the mill adaptive-reuse cohort (Open Square, Residences on Appleton, Skinner Silk Mill, Gateway City Arts, all heat-pump-augmented under municipal rates), and Sublime Systems' early-stage green cement operations.

1.5 to 2 km south sits the Holyoke Water Pollution Control Facility, operated by Veolia North America under a multi-decade contract running to 2035. The plant treats an average 30,300 m³/day (8 MGD) with capacity for 66,200 m³/day (17.5 MGD), and as of 2025 runs Veolia's Hubgrade AI-enabled predictive operations platform. Year-round thermal demand for sludge handling and building heat at a wastewater plant of this scale is a high-fit offtake at the temperature MicroLink's Loop 3 host loop delivers.

The positioning is not "we have a signed offtake today." It is: "MicroLink's three-loop architecture creates an exportable warm-water resource at 65 °C return temperature, and a half-dozen credible offtake candidates sit within a kilometre of the parcel. HG&E and the City become the natural anchor for a Holyoke district heat aggregator that grows around the MicroLink resource over a five- to ten-year horizon, supported by state and federal heat-reuse incentives." For the Healey administration, this is exactly the kind of place-based, decarbonisation-anchored economic development the Mass Leads Act and Mass Wins Act are written to enable.

The Gateway City and Green Community context matters

Holyoke is a designated Gateway City under Massachusetts Department of Housing and Economic Development definitions, qualifying for targeted economic-development incentives, Brownfields and Underutilized Sites funding, and Gateway City Manufacturing Mentorship Network resources. Holyoke is also a designated Green Community under the state's Green Communities programme, which carries energy efficiency planning support and small grants for municipal energy projects.

Holyoke Community College, a Mass Reconnect free-tuition community college with HCC President George Timmons on Governor Healey's AI Strategic Task Force, anchors the workforce pipeline. The MassHire Holyoke Career Center and the Holyoke Codes program extend that pipeline into adult workforce and K-12 community programming. The local-buy-in question that often slows large-scale state-anchored projects in other locations is, at Holyoke and MGHPCC, structurally favourable to the project from day one.

Prof. Ayse Coskun is Director of Boston University's Center for Information and Systems Engineering, Associate Dean for Research at the College of Engineering, full professor in Electrical and Computer Engineering, and Chief Scientist at Emerald AI. She is the academic voice of grid-responsive AI compute, and the introduction that brought MicroLink into the MGHPCC conversation.

Emerald AI launched publicly in June 2025 with a $24.5 million seed round led by Radical Ventures with strategic investment from NVIDIA's NVentures fund. The company was named to TIME100 Most Influential Companies 2026, to TechCrunch's AI Disruptors list, and to The Information's "50 Most Promising Startups of 2025." Emerald's flagship demonstration, published in Nature Energy and co-authored with NVIDIA, Oracle, Salt River Project, and EPRI, showed a 256-GPU cluster in Phoenix reducing power by 25 percent for three hours during a grid peak event, with full preservation of workload performance SLAs.

The non-overlap thesis matters operationally

Emerald's Emerald Conductor is software. It sits above the workload scheduler, adjusts power consumption in response to grid signals, and modulates slowdown across "flexible" jobs that tolerate small performance adjustments. MicroLink is hardware. It sits below the rack, runs three thermal loops independently of the building, and decouples the compute envelope from the host's thermal infrastructure. Neither displaces the other. They stack.

A MicroLink container block running an Emerald Conductor control plane offers four independent degrees of grid flexibility: (1) thermal flexibility at the host loop, where heat is exported when an offtake is ready and rejected to dry cooler when it is not; (2) thermal flexibility at Loop 2, where the CDU temperature setpoint can be modulated; (3) load flexibility at the workload layer via Conductor, where AI jobs respond to ISO-NE or HG&E grid signals; and (4) tri-gen flexibility where on-site generation supplements the grid feed during constrained windows. The four levers are independent; a grid event that exhausts one is absorbed by the others.

EPRI DCFlex and the missing academic hub

EPRI's DCFlex initiative is a three-year programme balancing power-system demands with the energy needs of large data center loads. As of public reporting, DCFlex operates nine demonstration hubs with roughly 45 collaborators. Demonstrations are scattered across hyperscale and colocation operators, with limited academic-research representation. MGHPCC is not yet on the DCFlex map.

A stacked MicroLink and Emerald Conductor deployment at MGHPCC, with the thermal independence of the container block and the load flexibility of the Conductor software, would be the first DCFlex demonstration that combines hardware and software grid-flexibility in an academic-research setting. EPRI funding pathways for the demonstration are well-established. The proposed sequencing is straightforward: a co-authored joint proposal in Q3 2026, EPRI committee evaluation in Q4 2026, programme award and partnership instrument in H1 2027, demonstration commissioning aligned with ST2 operational date.

For the state-side audience, the demonstration value is direct. ISO-New England has the most expensive wholesale electricity market in the continental United States, the most constrained generation capacity outlook, and the strongest case for demand-side flexibility from large new loads. A documented MicroLink-plus-Emerald demonstration at MGHPCC, anchored in EPRI's national programme, is national-press-quality content for the Mass AI Hub.

The MicroLink block is a data center, not a server. Each megawatt is delivered as a pair of 40 ft (12.2 m) ISO containers, one carrying the Compute Module and one the TPM and electrical equipment. They arrive on a flatbed, sit on a slab, and connect to a dry cooler array and a host thermal loop. They are operated by MicroLink. The compute equipment inside the Compute Module is the consortium's to specify, procure, capitalise, and own.

The three thermal loops. Loop 1 is the server loop, direct-to-chip cold plates at 45 °C (113 °F) supply, Rubin-compatible. Loop 2 is the facility loop inside the TPM/electrical container, with a coolant distribution unit lifting the return to 65 °C (149 °F). Loop 3 is the host loop, terminating at either a heat-recovery offtake or a dry cooler rejection path. The two terminations are not alternatives operationally; the dry cooler is always available, and the host offtake is added on top when a thermal partner is ready. The dry cooler is the architectural commitment that MicroLink does not depend on the host's thermal sink to operate.

What stays untouched in the existing building. The MicroLink deployment does not touch the existing 90,000 sq ft (8,360 m²) MGHPCC building. The hot-aisle pods, the in-row chilled-water cooling, the existing rejection topology, the annual 24-hour facility shutdown, and the LEED Platinum certification envelope are all preserved exactly as they are. The existing white space continues to absorb ST1 and any workload that fits its 14 kW mean, 25 to 90 kW peak envelope, for as long as that envelope is useful to the consortium. This is the operational fact that protects the state's existing AICR investment while creating the new build the trajectory requires.

The AICR RFP, Section 5E, states that the procurement is by MGHPCC and that "more than 80% of the funds supporting this RFP are obligated from sources that require their use for capitalizable expenditures under accepted accounting principles." Both clauses are determinative for any procurement structure MicroLink proposes. The clean structure is straightforward.

MGHPCC owns the compute equipment. The B200, RTX, Rubin, and Rubin Ultra hardware that sits inside the Compute Module is procured, contracted, capitalised, and depreciated by MGHPCC under the same accounting treatment as any other AICR component. The compute is the asset; the consortium's capital outlay on that asset preserves the AICR programme's capitalisable-spend requirement verbatim. MicroLink owns and operates the containerised shells. The Compute Module container, the TPM and electrical container, the three thermal loops, the dry cooler array, and the slab on the undeveloped acreage are MicroLink assets, depreciated on MicroLink's books, operated under a host-partner agreement consistent with the AICR governance model.

The consortium pays MicroLink a hosting fee, project-dependent and not globally locked, for the containerised infrastructure service. The fee structure aligns with how MGHPCC contracts for other facility services today; it does not require new governance or new procurement instruments. The Mass AI Hub at MassTech continues as programme sponsor through the existing $31 million state grant and any follow-on Mass Leads or Mass Wins capital authorisations.

For the Healey administration's communications team, the procurement story is simple. "The state continues to fund the AI Hub. MGHPCC continues to own and operate the AI compute. MicroLink builds and operates the infrastructure on the undeveloped acreage that MGHPCC's master plan reserved for exactly this purpose. The existing building stays untouched. The new build delivers ERE below 0.5 and DCFlex demonstration value." That is the press release.

The container deployment runs three traditional sequential steps in parallel. Site preparation, foundations, electrical service drops, and water and glycol loop installation can run concurrently with factory build of the containers themselves, and equipment installation inside the containers is a factory process not a field process. The compression from this parallelism is typically 9 to 12 months for a deployment in the 5 to 20 MW range.

For an ST2 operational target of January 2027, that means a partnership structure agreed in H1 2026, site preparation in Q3 2026, term sheet by Q4 2026, partnership instrument signed by year-end 2026, and operational lands in Q1 2027 with two to three months of slack for commissioning. ST3 in August 2028 is comfortable from the same baseline, with a phased addition rather than a fresh build. The ask, restated as a timing question, is straightforward: a 60- to 90-day technical evaluation in May to July 2026, a site visit in Q3 2026, a term sheet by Q4 2026, and a partnership instrument signed by year-end 2026 lands ST2 in January 2027.

The honest risk register and what it does not include

The risks of not moving on ST2 in time are different and larger. The alternative to a containerised deployment on the undeveloped acreage is to retrofit Rubin-class density into the existing building, with substantial disruption to existing AICR ST1 workloads, the LEED Platinum certification under substantive review, and multi-year multi-million-dollar engineering project risk. The honest comparative reading is that the containerised path is materially lower risk for the state's existing AICR investment, not higher.