MGHPCC AI Compute Resource, Holyoke
FJI source-qualified: four-document gate met, including project-specific government evidence. Qualification evidence and decisions →
MGHPCC said on September 29, 2026 that researchers and startups were using its AI Compute Resource. October technical documentation lists 400 GPUs. This is a shared compute installation at an existing campus, not a verified new building.
What the record establishes
- Evidence base
- 7 reviewed documents
- Current stage
- Operating shared cluster, according to operator
- Installed configuration
- 400 GPUs; 50 GPU nodes
- Funding boundary
- $31 million state grant plus initial university match
Investment takeaway
AICR has a stronger delivery basis than an announced campus: its operator describes users already accessing it and publishes an installed-system specification. The next test is sustained access, useful research output and operating cost. GPU counts establish configuration, not utilization, economic productivity or supplier earnings.
Project at a glance
- Location
- 100 Bigelow Street, Holyoke, Massachusetts, United States
- Project scope
- Nonprofit shared research AI computing
- Named participants
- MGHPCC nonprofit university consortium; Cambridge Computer; Dell; NVIDIA; VAST Data
- Capacity boundary
- 400 GPUs across 50 GPU nodes: 248 NVIDIA B200 and 152 RTX PRO 6000. Dedicated AICR electrical load undisclosed.
- Map precision
- Approximate Census street-address reference for the documented address; not a surveyed building or parcel boundary.
A shared research installation in Holyoke
The Massachusetts Green High Performance Computing Center is operated by a nonprofit consortium of six universities. Its AI Compute Resource extends shared research computing within the existing Holyoke campus. In September, MGHPCC said researchers from universities, startups and research organizations were using AICR. Institutional access arrangements can differ; that statement does not establish unrestricted commercial availability. [2, 5, 7]
What the hardware record shows
The October 8 technical overview lists 31 nodes with eight B200 GPUs each and 19 nodes with eight RTX PRO 6000 GPUs each: 248 plus 152, or 400 GPUs. It separately lists service nodes and effective all-flash VAST storage. Service nodes are not additional GPU nodes. Other systems at MGHPCC, including Harvard’s Kempner cluster, are separate installations and are not added to AICR’s count. [3]
Funding is broader than incurred capex
The state’s December 2025 announcement identifies a $31 million grant, an initial university match and expected total funding exceeding $120 million by 2030. It names Cambridge Computer as integrator with Dell, VAST and NVIDIA technology. These are funding and procurement descriptions; they do not establish spending already incurred or a supplier’s recognized revenue. [1]
The campus power figure is not AICR’s load
WBUR reported in September that the existing campus was permitted to draw up to 12 MW and was using less. That is an entire-campus boundary. AICR’s dedicated electrical load is not disclosed in the reviewed documents. Campus energy and water descriptions cannot be treated as metered AICR performance. The city’s restrictions on new data centers also distinguish the existing MGHPCC campus; this profile does not represent an additional approved building. [4]
Public-company disclosures
These reports describe the suppliers’ businesses. They do not allocate revenue, orders, capital or profit to this installation. Equipment presence does not disclose contract value or supplier revenue.
- NVIDIA (NVDA): Investor relations and shareholder disclosures · Q2 FY2027 results, published 2026-08-26
- Dell Technologies (DELL): Investor relations and shareholder disclosures · Q2 FY2027 results, published 2026-09-01
What remains unverified
Dedicated AICR electrical consumption; metered utilization; allocation among members; recurring access costs; incurred project expenditure and vendor contract values; independent acceptance or operating performance.
Next evidence checkpoint
Track institution access, sustained utilization, measured resource consumption and outcomes; separate further equipment additions from the existing 400-GPU configuration.
Delivery timing
Operator reported use September 29, 2026; technical configuration dated October 8. No utilization percentage or independent performance audit verified.
Related Northeast projects
- New Jersey Supercomputer Initiative · New Jersey
- Aligned IAD-06, Frederick · Maryland
- Concord Locke Road Micro Data Center · New Hampshire
- Servistar Westfield Data Center Campus · Massachusetts
Sources and claim boundaries
- Massachusetts AI Hub / MassTech · December 12, 2025
Announces the state grant, initial university match, expected program funding and selected delivery suppliers; it predates the operator’s use announcement.
Reviewed November 15, 2026.
- MGHPCC · September 29, 2026
Says researchers and startups are using AICR. It does not disclose utilization, service revenue or independent operating measurements.
Reviewed November 15, 2026.
- AICR technical documentation · October 8, 2026
Lists 248 B200 and 152 RTX PRO 6000 GPUs across 50 GPU nodes; service nodes are listed separately.
Reviewed November 15, 2026.
- WBUR · September 11, 2026
Reports existing-campus energy, water and local-policy context. Its campus-wide 12 MW permit figure is not AICR’s dedicated load.
Reviewed November 15, 2026.
- MGHPCC · Governance
Identifies nonprofit operation and university membership; it does not disclose installation-level operating economics.
Reviewed November 15, 2026.
- MGHPCC · Directions and campus address
Identifies 100 Bigelow Street in Holyoke. The address supports an approximate campus reference, not a surveyed equipment location.
Reviewed November 15, 2026.
- Yale Research Computing · AICR access guidance
Describes institution-specific application and access arrangements, which should be distinguished from the operator’s system-use announcement.
Reviewed November 15, 2026.
Map reference: U.S. Census address geocoder. Approximate Census street-address reference for the documented address; not a surveyed building or parcel boundary.
AI-assisted public-source research by Fjord Journal. No site visit, private contract inspection or independent metering. Unknown operating IT load and paid capital remain undisclosed; no whole-project certification or investment recommendation is claimed.
Company investor information & quarterly statements
Financial-disclosure review remains incomplete for one or more participants. The company-level limitations below remain unresolved; this section does not establish full corporate-disclosure verification.
Corporate disclosures provide company-wide context. They do not establish this project’s revenue, spending, profit, ownership share or operating capacity. Disclosure verification records →
