01 · Who's here

Working Document

The people and organisations working to make this happen

July 2026

Watercolour aerial map of New York Regional Wastewater Facility, the adjacent industrial parcel including the Prologis STEM Park site, and the surrounding South Bay landscape

MICROLINK DATA CENTERS

Nick SearraCEO and Co-Founder
Sancha OlivierCEO, Design
Shane PatherChief Technology Officer
Carla TrippHead of Operations and Impact
Andrew ThomasChief Commercial Officer
David HasslerHead of Sales
Jeff SvedahlCEO, MicroLink Edge
Deniz AkgulCapital & Investment Advisor

PROJECT WORKING GROUP

Maddy Fairley-Wax, P.E.Jacobs
Mats ErikssonArctos Labs
Deborah S EgelandSage Oak AI
Joel CabreraCity of New York
Ryan BirdFuelCell Energy

AKILA

Sharon ChenPilot Program Lead

NVIDIA

Jumbi Edulbehram, PhDVP Global BD, Public Sector · host
Andria ZouSr Director, Global AI DC Strategies
Jared CarlGlobal AI Data Center Lead
Karthik MandakolathurProduct Manager, Magnum IO
Elad BlattHead BD, Telco Networking
Ben GueretTechnical Program Manager
Danny ZaidifardBD, Strategic Partnerships
Alex PazosSr BD Mgr, Smart Spaces
Claudio FassiottiEnterprise Lead, Africa
Wendy Zhu, PhDValidation, partner-adjacent
Chris ChoCloud and partner technical
Rodney ShetlerPre-sales and Solutions Eng.
David MessinaInception VC Alliance, adjacent

The moment · why distributed, and why now

Responsible Data Center
Development Pilot

New York is moving to slow its largest data centers. The heat that smaller ones recover can warm the buildings next door.

The largest campuses draw heavily on the grid, consume water, and return little to the places around them. The questions being asked of them in New York are about rate impact, water, and what the neighbourhood gets back. A distributed node answers all three as its normal mode of operation, not as a retrofit added later.

The efficiency frame for this pilot is built around waste heat recovery and reuse. A MicroLink node recovers almost all of the heat its compute produces and hands it to the host process and the buildings nearby. Reuse is reported as ERE alongside PUE, so the heat that leaves the site is counted, not only the power that enters it.

The pilot is the proof. Small sited nodes that serve public compute and warm public buildings are a different proposition from a private campus that draws on the grid and returns nothing local. Everything that follows is built to be read as the former.

Aerial view of a one megawatt node beside a football field of greenhouse indoor farming, a blue compute container linked to the glasshouse by an insulated heat pipe. — One megawatt of data center can support one football field of indoor farming, reducing fresh food scarcity, creating two new jobs. It can also support a garden that brings a community and an innovation center that inspires a future generation.

New York pilot data center on the East River, mass timber building with multi colour thermal loops, server racks behind glass, Manhattan skyline beyond.

02 · MicroLink

MicroLink designs, builds, owns and operates data centers inside of industrial facilities.

01 Wastewater treatment

Server reject heat returns to anaerobic digesters at 35 to 38 °C. Biogas powers the load through molten carbonate fuel cells.

02 Breweries

Reject heat at 60 to 70 °C feeds wort heating and clean-in-place water as a baseload thermal source replacing steam.

03 Hotels and hospitality

Sub-MW edge pods deliver 60 °C water for laundry, kitchens, hot water, and pools. Gas displacement plus on-property compute.

04 Hospitals

Hospital-grade 24/7 baseload heat with the redundancy hospitals already have. Thermal resilience plus campus compute.

05 District heating

Direct feed of 65 °C water into a district network return loop. A year-round, weather-independent heat source.

06 Research campuses

University utility plants with district heating networks and on-site research compute demand. UW Seattle is canonical.

MicroLink glass-fronted facility, evening — The pods are modular, repeatable, upgradable. The buildings they live inside are not.

Service · 01

Distributed DC as a Service

Efficiency-driven compute, distributed close to where it's used. Liquid-cooled data centers on industrial host sites, with waste heat returned to the host process. Hosts contribute land and thermal offtake. Customers consume compute. The pods are modular, repeatable, upgradable.

I lived in Lagos for five years. Then Nairobi for five more. In 2001, if you weren't in the same room as someone in Nigeria, they were unreachable. No reliable post. No working landlines for most of the country.

Then GSM arrived. Half a million subscribers in 2001. Eighty million by 2010. Two hundred and twenty million today. Our current moment could be an even greater shift. It seems only right to ensure it's another good one.

Nick Searra

CEO and Co-Founder

Liquid-Cooled Colocation

02

High-density colocation.

GPU-as-a-Service

03

Managed NVIDIA clusters.

Inference-as-a-Service

04

Production model serving.

02.A · The three technologies · one model at three scales

A MicroLink node is a small, liquid cooled compute block at a host that already needs heat. Power in from the grid or onsite supply. Heat out by three routes: host process, an adjacent building or district loop, or domestic hot water for nearby housing. A dry cooler reject path runs alongside at all times.

01 · Edge

Edge

An edge tier envisioned to sit inside the demand building itself, below 1 MW. The pilot would explore a school, a hospital, and a community housing project as candidate hosts. In each, the recovered heat could land in the same building it is made in, as domestic hot water and space heat, so a small compute load does the quiet work of a boiler for the people right next to it.

BELOW 1 MW

IN BUILDING HEAT

SCHOOL · HOSPITAL · HOUSING

MicroLink edge cabinet on a residential street, sized to sit beside community housing.

02 · Pod

Pod

Newtown Creek, New York City's largest wastewater plant, is the kind of host a pod is built for. A deployment there could return recovered heat to the plant's own process and to the dense housing that rings it, a single block of compute standing in for fuel that would otherwise be burned. It is a candidate, not a commitment, the clearest illustration of what the pod tier could become.

PLANT PROCESS HEAT

SURROUNDING HOUSING

NYC'S LARGEST WWTP

A MicroLink modular compute block lifted onto its foundation pad during install.

03 · Campus

Campus

The campus is the scale the model could reach where the power and the heat demand are largest, phased to sit under the permit line. The pilot is weighing several candidate hosts: the New York City waste to energy estate, the Massena legacy industrial site on low cost public hydro through NYPA, and the Anheuser-Busch brewery at Baldwinsville. No single site is named as the lead. Each is a possibility the study would test, not a site already chosen.

POWER ANCHORED

UNDER THE PERMIT LINE

MULTIPLE CANDIDATE HOSTS

Riverside data center campus with greenhouses drawing recovered heat, water in the foreground.

Eight zones of collaboration. Named participants on each one.

Cluster A · The build

Engineering

Grid independence, heat recovery, fabric, monitoring, design. Four zones of joint engineering.

Cluster B · The customer

Who we serve in New York

Public-sector compute, served close to the people who use it. We design the service together with the city, in the city, for the city.

Cluster C · Contributions

What flows back

Open-source contributions, datasets, semantic conventions. Workforce pipeline with New York State University and the NVIDIA Deep Learning Institute. Two zones of contribution to the broader compute and energy industries.

Zone 01 · Build

Grid independence and heat recovery

Diesel-free 2N power topology. MCFC on biogas as prime, LFP for transient, PEM hydrogen for ramp. Server reject heat into host process loop. The closed thermodynamic loop.

NVIDIAEnergy team + Jared Carl

MicroLinkShane Pather

CITY OF SAN JOSÉJoel Cabrera

Explore further Zone 02 · Build

Inter-pod and multi-site fabric

Quantum-X800 + AC SU at 576 GPUs, twin-plane fat tree. Spectrum-X for multi-tenant scale-out. Site-to-site federation via Spectrum-XGS. Edge-to-core hierarchy with Jetson at sub-MW host sites.

NVIDIAKarthik Mandakolathur

MicroLinkShane Pather

ARCTOS LABSMats Eriksson

Explore further Zone 03 · Build

Monitoring and control

DCGM, NVML, Mission Control at IT layer. Metropolis for building-as-managed-asset. Omniverse / DSX Blueprint for digital twin. Jetson / IGX Orin for facility-side control loops.

NVIDIAPlatform team · Pazos · Fassiotti

MicroLinkShane Pather

ARCTOS LABSMats Eriksson

Explore further Zone 04 · Build

Data center design

Liquid-cooled from day one. Sized for next-gen hardware envelopes (B300, Vera Rubin path). Architectural shell with civic intent. Co-developed reference architecture for the canonical pod.

NVIDIAJared Carl + Andria Zou

MicroLinkSancha Olivier

JACOBSMaddy Fairley-Wax

Explore further Zone 05 · Product

Sovereign public-sector AI

One of MicroLink's four product lines, alongside colocation, GPU-as-a-Service, and inference-as-a-Service. Confidential compute, multi-tenant isolation, per-tenant clusters, sovereign fine-tuning on local weights. Six public-sector customer segments.

NVIDIAEdulbehram · Pazos · Gueret

MicroLinkNick Searra

CITY OF SAN JOSÉJoel Cabrera

Explore further Zone 06 · Product

Sustainability and climate reporting

Co-authored white paper. ERF, WUE, CUE, PUE published metrics. EU EED 2027 waste-heat compliance. California SB 253 / SB 261 / AB 1305. Net Grid Energy Position. ISO 14064-3 third-party assured.

NVIDIAAndria Zou

MicroLinkShane Pather

SAGE OAK AIDeborah S Egeland

Explore further Zone 07 · Contributions

Open-source contributions

Heat-coupled control loop to LF Energy. FacIT semantic conventions to OpenTelemetry. DSX Max-P module co-developed. OCP liquid-cooling spec. Thermal Lab dataset. Apache 2.0 default.

NVIDIADanny Zaidifard + Wendy Zhu

MicroLinkShane Pather

MICROLINKSancha Olivier

Explore further Zone 08 · Contributions

Workforce and ecosystem

SJSU AI Infrastructure Apprenticeship with NVIDIA DLI. NCA-AII entry, NCP-AIO exit. Three sub-tracks: liquid cooling, fuel cell, AI infrastructure operator. Inception startup hosting at preferential rates.

NVIDIAEdulbehram · DLI · SJSU

MicroLinkJeff Svedahl

MICROLINKAndrew Thomas

Explore further

Energy balance 2 of 4

Design · Energy balance

New York energy balance, today and post-ADFU

The case for MicroLink at New York is not heat supply gap-filling. The plant has more cogen heat available than it needs. The case is thermal substitution: MicroLink absorbs the low-grade duty currently met by cogen, freeing high-grade cogen capacity for the new MHP module and freeing the post-upgrade biogas surplus for monetisation.

TodayPost-2022 cogen, pre-ADFU

Plant electrical demand~11 MW

Cogen output11–14 MW

Cogen recoverable thermal17–19 MW

Plant thermal demand5–7 MW

Heat rejected to atmosphere8–12 MW

Biogas production50,000–65,000 m³/d

Biogas energy600–900 MMBtu/d

Post-ADFU + MicroLink2028–2029

Plant electrical demand12–13 MW

Cogen output14 MW nameplate

Cogen recoverable thermal17–19 MW

Plant thermal demand7–9 MW

Heat rejected to atmosphere→ 0 MW (absorbed by MicroLink)

Biogas production85,000–120,000 m³/d

MCFC electricity (BTM)2.3 MW (FuelCell SureSource 3000)

MCFC thermal output1.8 MW

H₂ capability (tri-gen)up to 1,200 kg/day

Net grid draw at 10 MW IT8.9 MW (−21%)

CO₂ avoided11,750 tCO₂e / year

Two simultaneous shifts. First, biogas production increases by 60 to 80 percent through the combined MHP uplift and FOG co-digestion. Second, plant thermal demand increases only modestly because the new MHP module concentrates duty at 75 °C, not at the mesophilic temperature where most of the existing demand sits.

The combined effect: a substantial biogas surplus the existing 14-megawatt cogen cannot consume at full duty cycle, paired with a heat-rejection problem that has not improved. Both are resolved by a colocated thermal partner.

Aerial site plan, New York–Santa Clara Regional Wastewater Facility

S-section cross-section, integrated facility

The integrated facility section illustrates how a MicroLink deployment occupies a compact vertical envelope adjacent to the host process, with the thermal interface running horizontally between them at digester level. The MCFC sits behind the compute envelope, fed by the freed biogas line from the digesters.

Freed value 3 of 4

Design · Freed value

Three pathways for the freed biogas at 2026 California prices

The freed biogas (both the MHP and FOG yield uplift and the cogen displacement that MicroLink enables) has three monetisation pathways. Each has been priced at 2026 California market terms, drawing on CARB LCFS quarterly transfer reports through February 2026, CPUC Decision 24-08-007, the IRS final rule on Section 45V, the OBBBA, and the Jacobs McLeod and Horrax 2022 hydrogen series.

A

Cogen export and behind-the-meter offset

PG&E SRAC · BioMAT Cat 1 · retail offset

$19/MMBtu

range $8–33

Annual at 5–10 MW IT scale
$0.6–4M / year

Recommended

B

RNG injection with LCFS + D3 RIN

PG&E G-BIO interconnection

$20/MMBtu

range $15–25

Annual at 5–10 MW IT scale
$1.5–7M / year

C

Hydrogen via Pathway 5 SMR

Biomethane upgrading + SMR

$30/MMBtu

range $22–60

Annual at 5–10 MW IT scale
$3–18M / year

Range reflects 45V status uncertainty post-OBBBA.

Pathway B is the primary monetisation route at New York. The all-in value reflects 2026 California market conditions: CARB LCFS pathway value at $5–10 per MMBtu (post-2025 LCFS market amendments), federal D3 cellulosic RIN value at $5–10 per MMBtu, gas commodity at PG&E citygate at $3–4 per MMBtu, and avoided distribution charges. PG&E's Schedule G-BIO interconnection programme provides the technical pathway. SB 1440 capital incentives offset interconnection capex up to $3 million for non-dairy WWTPs. WWTP biogas carbon intensity is structurally distinct from deeply-negative-CI feedstocks like dairy or food-scraps RNG; the figures here reflect a realistic +30 to +55 gCO₂e/MJ Tier 2 pathway.

DIAGRAM · LCFS price trend California LCFS credit price, 2024–2026, with diesel benchmark and CNG-equivalent pricing overlaid. img-6-1

Aggregate value · central case

Combined gross value at the post-ADFU plant, with a 6 to 10 megawatt MicroLink IT deployment:

approximately $7 to $16 million per year

Freed-biogas pathways: $5–14M/year. Plus the MCFC tri-generation layer adds ~$1.8M/year of behind-the-meter electricity at avoided-cost rates. This is the prize. The commercial structure between MicroLink and the City (the value share between host and developer) is a separate negotiation. This figure illustrates the magnitude, not the eventual allocation.

Two-track partnership 4 of 4

Design · Partnership

A two-track partnership structure

The partnership is structured to protect each party's interests, to fit the ADFU project's design schedule, and to support MicroLink's ability to raise the capital required for the eventual deployment. It runs on two parallel tracks anchored on a sequenced commitment instrument: Expression of Interest, Letter of Intent, Definitive Agreement.

Track 1

The stub-out and the EOI

A future-ready thermal interface, designed and built by Jacobs as a defined scope addition to the existing $200 million ADFU progressive design-build contract.

Three interfaces

Tap on the mesophilic digester recirculation manifold; parallel branch on the cogeneration jacket-water return loop; tap on the raw sludge feed line.

Sized for

A future 5 to 10 megawatt thermal interface, designed by Jacobs to specification, owned by the City as part of the ADFU asset.

Funded by

MicroLink at approximately $1 million, contributed to the City under a Cost-Sharing Agreement structured through New York Charter §1217 (developer carve-out) layered on a Government Code §4217 finding (energy-services framework).

First commitment

A short-form Expression of Interest signed within 30 days, scoped to authorise inclusion of the stub-out concept in the ADFU design-basis discussion. Three EOIs together (City of New York, NVIDIA, Jacobs) anchor the equity raise that funds the stub-out and the eventual deployment.

Track 2

The LOI and the Definitive Agreement

Within 90 days of EOI signature, both parties commit to a full Letter of Intent. Within 9 months, the Definitive Agreement is signed.

Letter of Intent

Captures the technical specification, the commercial framework principles, and the timeline for Definitive Agreement negotiation.

Definitive Agreement

Land licence or lease for the future MicroLink facility on the freed footprint; thermal services agreement; biogas monetisation framework; operational protocols, performance standards, and dispute resolution; right of first negotiation for MicroLink on thermal interface use post-commissioning; information rights on plant operating data; performance milestones and termination provisions.

Trigger

The Definitive Agreement triggers escrow release on the stub-out funding. ADFU stub-out construction proceeds in base scope. MicroLink construction begins, with the modular first-deployment structure operational in the 2028–2029 window.

DIAGRAM · Escrow flow Escrow mechanics: EOI signed (Day 30) → LOI signed (Day 90) → Definitive Agreement signed (Day 270) → escrow releases on the stub-out funding. Capital protected on either path. img-8-1

The trigger

The two tracks are connected by a single trigger structure. The $1 million stub-out funding is held in escrow against execution of the Expression of Interest and releases on signature of the Definitive Agreement. If the Definitive Agreement does not sign, for any reason, the stub-out is value-engineered out of ADFU at no cost to the City, and the escrow returns to MicroLink.

This protects MicroLink's capital, gives the City a structured commitment, gives Jacobs a clean engineering scope, and creates a sequenced set of signed instruments (EOI, then LOI) that supports MicroLink's fundraising.

Eventual deployment scale

Eventual deployment capex$80–120 million

Stub-out contribution$1 million (~1%)

Term of host paymentsTBD in DA

MicroLink team: Nick Searra, CEO & Co-Founder · Sancha Olivier, Design, Site Inspection & Review · Shane Pather, CTO · David Hassler, Sales & Customer · Andrew Thomas, CCO · Deniz Akgul, Capital & Investment Advisor.

The full review of this pilot

An edge tier in public buildings, a pod at a wastewater plant, and a campus on a power rich host.

MicroLink would supply the compute and the heat recovery. The study partners would carry the heat the last distance into homes, schools, and public buildings.

Section drawing of a MicroLink node passing recovered heat to a row of greenhouses through colour coded supply and return loops.

The sites in view

Each is a candidate the pilot would test, not a site already chosen.

Site	Tier	Host	Heat destination
Newtown Creek, Brooklyn	Pod	Wastewater	Plant process and surrounding housing
School, hospital, community housing	Edge	Public buildings	In building: domestic hot water and space heat
Massena, Northern New York	Campus candidate	Legacy industrial on NYPA hydro	District and process
NYC waste to energy estate	Campus candidate	Municipal	Public buildings
Anheuser-Busch, Baldwinsville	Campus candidate	Brewery	Process heat

The study partners

Heata moves recovered heat into housing as domestic hot water. Span manages power and load at the building. Akila runs the digital twin and operations layer. NYSERDA convenes the study and sets the efficiency frame the work is measured against.

The demand context

New York is standing up its own public artificial intelligence computing capacity. Benchmarking these deployments against that public demand reframes the question the pilot answers. A distributed node serving public research and warming public buildings is a different proposition from a private campus that draws on the grid and returns nothing local.

What the pilot would report

Every node would keep the dry cooler reject path, so uptime would never depend on whether the heat is being drawn that hour. Reuse would be reported as ERE alongside PUE. Heat delivered, power drawn, and CO2 avoided would be recorded per site for the study.

Pilot Data

The place to collect and share data for this project.

These are the research data areas we are working with, for the study partners, for NYSERDA, and for the public view. Data will be added as per request.