Infrastructure Without Reciprocity
On what Project BOSC may mean for Allen County
The first piece planted a question. This one asks what the answer costs, and what a fair exchange would look like.
Let me say directly what I believe: data centers are infrastructure, and infrastructure is good. A facility that puts real compute capacity into a community, employs local workers, pays taxes, and becomes woven into the local economic fabric is a net asset. The question has never been whether Allen County should host one for me. The question is whether Allen County was told what it was actually hosting, and whether the terms of the exchange reflect what the community put in.
When I published “Decoding Project BOSC,” the response was largely what I expected. The question I anticipated most, and the one that arrived on schedule, came from a specific kind of reader: practical, development-minded, not unsympathetic. Okay, but isn’t a Google data center good for the county? It is a fair question. It deserves a real answer.
That answer requires understanding what a government-grade cloud data center actually is, why it is built differently than what the press releases imply, and what the economics of that difference do to a community that was never told it was bidding on something other than a commercial cloud deployment. The issue is not the facility. It is the terms under which the community received it.
Two clouds, one building
The data center industry has a taxonomy problem. “Data center” conjures a neutral image: a refrigerated building full of servers. But within that category sits a spectrum of facilities so different in purpose, architecture, and capital structure that they share little more than the word on the sign outside.
At one end: the commercial hyperscaler cloud. Google, Amazon, Microsoft. Multi-tenant. Globally load-balanced. Optimized for elasticity, cost efficiency, and geographic redundancy. When Netflix serves your Tuesday night television, it moves through AWS facilities designed for exactly this workload: ephemeral, distributed, repriced quarterly, and indifferent to which physical server actually runs your stream. I know AWS in Columbus as us-east-2 — actually many clusters of data centers called availability zones in AWS terminology.
At the other end: the government-grade sovereign cloud, and its classified sibling, the govcloud. These are not commodities. Boston Consulting Group has documented that hyperscalers charge a premium of up to 30 percent for sovereign cloud offerings, and that figure understates the structural difference.¹ Sovereign cloud requirements exist specifically because of legal and jurisdictional constraints that standard public cloud cannot satisfy, including the 2018 U.S. CLOUD Act, which enables American law enforcement to compel data access from U.S.-based cloud providers even when that data resides outside U.S. borders. For DoD workloads classified at Impact Level 6 — the Secret level — the architecture is not a variant on commercial cloud. It is a different product category.
Google Distributed Cloud makes this explicit. GDC achieved DoD IL6 authorization in May 2025, enabling the company to provide “a secure, compliant, and cutting-edge cloud environment at IL6” for “the most sensitive Secret classified data and applications,” building on existing IL5 and Top Secret accreditations.² This is not AWS selling extra encryption. GDC is designed to operate in full, indefinite disconnection from the public internet. The architecture is built to meet “NIST SP 800-53-FedRAMP High security controls, ICD 503/703, FedRAMP+ for IL6 authorization, and FIPS140-2L3”: accreditations with no relevance to commercial cloud workloads.³
The question BOSC raises — and which nobody in Allen County has officially answered — is which product line was built in American Township.
Why location matters more for govcloud than for commercial cloud
A commercial hyperscaler data center is, in theory, locationally agnostic. You pick for power cost, land cost, tax incentives, fiber infrastructure, and, lastly, local partnership opportunities. Latency matters at the margin. You do not need to be near your customers because your customers are everywhere and your load balancer handles it.
A govcloud deployment serving tactical edge applications has a fundamentally different location logic.
The GDC air-gapped appliance is designed for “tactical edge environments outside a data center,” creating an “isolated sovereign cloud in a box” that is physically disconnected from the internet. The appliance weighs approximately 100 lbs and is designed to be transported by two people, moved on and off vehicles, and operated in “uncontrolled environments” including tents and repurposed buildings.⁴ It meets MIL-STD-810H ruggedization standards. GDIT and Google Public Sector demonstrated this architecture operationally during Exercise Mobility Guardian 2025, enabling Air Force teams to run language translation, optical character recognition, and retrieval-augmented generation at the tactical edge — and, critically, to “quickly and securely update mission software” while forward deployed.⁵
That last capability defines the architecture. An appliance that can receive mission software updates in the field requires somewhere to receive those updates from. In a classified environment, that somewhere is not the public internet. It is a primary facility: one that holds the authoritative model weights, the validated software packages, and the integration environment where updates are tested against vehicle and sensor configurations before they are cleared for field deployment. That facility is not a rack in a basement. At IL6 classification, it is a purpose-built, generator-redundant, physically secured data center — the kind whose specifications you redact from an EPA permit.
The use case this serves is not abstract. The M1A2 SEPv3, the current production variant rolling off JSMC’s assembly lines, incorporates a Vehicle Health Management System and is described by the Army as “the foundation for future incremental system upgrades” that can “host any mature technology the Army deems operationally relevant.”⁶ The Army’s AI2C program is actively developing predictive maintenance AI models for rotary and ground platforms — modular, iteratively updated, tested against operational feedback.⁷ Each vehicle leaving JSMC’s 1.9-mile test track and entering service is a potential node in that update chain.
The Joint Systems Manufacturing Center in Shawnee Township is the sole manufacturer of the M1 Abrams family in the western hemisphere.⁸ GDIT — General Dynamics Information Technology, a sister business unit to JSMC’s operator under the same General Dynamics corporate structure — announced an expanded collaboration with Google Public Sector in November 2025 specifically targeting deployment of Secret-level AI to defense vehicles at the tactical edge.⁹ The integration and validation work that bridges a primary data center to those field-deployed appliances has to happen somewhere close to the vehicle program it serves. Proximity to the manufacturing and test facility is not incidental to that function. It is the function.
The tax structure in Ohio
Here is where “good for the county” requires a balance sheet.
Ohio has built one of the most aggressive data center incentive regimes in the United States. The structure stacks three layers of subsidy: a state sales tax exemption on data center equipment, state job creation tax credits, and local property tax abatements that typically run 75 percent for 15 to 30 years.¹⁰ These are not incidental concessions. They are, in the words of the Ohio Chamber of Commerce, “central” to Ohio’s data center attraction strategy.¹¹
The numbers are stark. According to Policy Matters Ohio, 13 data center agreements approved through September 2024 covered nearly $5.1 billion in investment but generated only 356 jobs and $31.6 million in annual payroll; a state revenue loss exceeding $281.9 million, or approaching $1 million per job, before accounting for local sales tax losses that push the total above $343 million.¹² These figures derive from the Ohio Department of Development’s own data.
Ohio’s state tax expenditure report projects the data center exemption will cost the state more than $140 million in the current fiscal year alone, though analysts regard that figure as a significant underestimate given recent mega-project announcements.¹³
Compare this to Virginia. The Tax Foundation has documented that Ohio imposes no tangible sales tax. That means the most capital-intensive assets a data center possesses, including servers, networking gear, uninterruptible power supplies, and generators, generate zero local tax liability.¹⁴ In Virginia, that equipment is taxable. Loudoun County alone estimates approximately $663 million in data center real and personal property tax revenue in fiscal year 2022 (a figure that has since grown substantially).¹⁵ The county’s own FAQ states that data centers now yield 38 percent of general fund revenue, allowing the county to lower the residential property tax rate every year for a decade.¹⁶
Ohio is hosting the capital. It is not capturing the returns if the American Township site is actually a govcloud facility.
The construction-phase employment argument is real and should be acknowledged plainly. Chuck Brooky, president of the North Central Ohio Building and Construction Trades Council, has said what a lot of people in the trades feel: “In the rural areas, we’re struggling. We’re just kind of scraping by. So when I hear of a data center coming into town, it’s a good thing for us.”¹⁷ The argument here is not against construction employment. It is about what comes after the ribbon cutting: a facility that draws on the community’s grid, water infrastructure, and land capacity while contributing, by design, almost nothing to the local tax base for decades.
If the American Township facility carries defense contracts — if it is a govcloud deployment serving GDIT’s Secret-level AI programs — the public subsidy is compounded by opacity. The county negotiated with a Delaware shell company (Bistrozzi LLC) under NDA. The generator specifications were redacted from the Ohio EPA permit as proprietary. The facility’s codename was never explained. A community that absorbed the land use, the power grid demand, and the permanent tax abatement was told: Google data center. As if that ended the inquiry.
What a commercial deployment could actually do
This is where the “good for the county” argument deserves its most serious engagement — not dismissal, but precision.
A genuinely commercial hyperscaler deployment, structured for mutual benefit, is not nothing. It is potentially quite a lot. The economic literature on knowledge spillovers and agglomeration is extensive and consistent: regions with dense technical infrastructure attract technical workers, and technical workers generate the knowledge exchange that seeds new ventures. A landmark study from MIT Sloan and Harvard Business School found that physical proximity among startups raises the probability of technology adoption spillovers by 2.5 percentage points — and that this effect decays sharply beyond very short distances, underscoring how much the physical clustering of technical activity matters.²² Research published in Small Business Economics confirms that being located in a cluster increases patenting rates among niche market leaders by a factor of 1.49, with effects more pronounced in industrially diverse regions.²³
The mechanism relevant to Allen County is not just the data center itself. It is what a genuine commercial cloud presence signals and enables: lower-latency access to GPU and AI infrastructure for local manufacturers running predictive maintenance or quality-control models; a fiber and power environment that reduces the infrastructure barrier for R&D-intensive startups; the possibility of a cloud region designation that puts enterprise-grade compute within economically viable reach of firms in the surrounding area. The advanced manufacturing base already present in northwest Ohio — not only JSMC but the broader automotive, agricultural, and industrial supplier ecosystem — is precisely the kind of environment where local access to serious compute infrastructure creates compounding returns. Defense primes and their Tier 1 suppliers are increasingly running AI-assisted maintenance, logistics, and design workloads. Proximity to the infrastructure those workloads require is a genuine competitive input.
None of this requires magical thinking about trickle-down effects. It requires that the facility be genuinely accessible — that local firms can purchase compute capacity from it, that its workforce participates in the regional economy in visible ways, that the knowledge it generates is not entirely classified. A commercial Google Cloud region in northwest Ohio would be a real economic asset. The question this piece has been building toward is whether that is what was built and, thus, whether our municipal trust — the very resources and opportunity natural to this land — was corrupted.
A classified, air-gapped govcloud facility operating under an NDA does not generate these spillovers. Its workforce holds clearances that constrain what they can discuss with neighbors. Its infrastructure is not available for local firms to provision against. Its knowledge outputs are, by design, not accessible to the surrounding innovation ecosystem. The facility occupies land, draws power, and sits within the tax abatement regime — but it cannot function as the anchor of a technical cluster in the way a commercial deployment could, because the conditions that make cluster effects work require openness, and openness is architecturally incompatible with IL6 classification.
This is not an argument that defense infrastructure is bad or that classified facilities should not exist. It is an argument about what Allen County was told it was getting, and what it may have actually received. If the former, the case for the tax structure becomes easier to make — genuine reciprocity requires genuine access. If the latter, the community subsidized a federal asset while being told it was a commercial one, and the terms of that exchange were never available for public negotiation.
The defense dependency structure
Allen County has a specific relationship with federal defense spending that predates Google by 70 years. The Joint Systems Manufacturing Center has been the economic anchor of this region through cycles of drawdown, supplemental appropriation, and political rescue that reveal a structural vulnerability most residents understand in their bones even if they don’t name it.
The economic literature on this dynamic is instructive. Research from the Richmond Federal Reserve on local fiscal multipliers finds that defense spending generates local income multipliers in the 1.0–1.6 range, but the employment response is modest, roughly four jobs per million dollars spent.¹⁸ A separate peer-reviewed study in the AEA Papers and Proceedings found that a dollar of local defense spending increases city-level GDP by roughly a dollar, with labor earnings increasing by $0.35.¹⁹ Defense dollars create real economic activity. But they create activity that is structurally dependent on decisions made in Washington, reviewed at the Pentagon, and allocated through contract vehicles that a county commissioner in Lima has no leverage over.
The question is whether a govcloud deployment at the scale suggested by the BOSC site characteristics changes this dynamic — or deepens it.
A facility whose contracts are classified, whose operators are unknown, whose permanent employment may be measured in the dozens, and whose tax exemptions run for decades represents a continuation of the federal dependency structure, not a diversification away from it. The community would have traded agricultural land and infrastructure capacity for a facility that is just as tethered to federal appropriations decisions as the tank plant, but with less transparency, fewer local jobs, and an explicit NDA preventing the people who negotiated the deal from describing what they agreed to.
What extraction looks like at the local level
The standard model of data center development in Ohio — NDA negotiations through shell companies, redacted permit specifications, 30-year tax abatements, exemption from tangible property taxation, extraction of infrastructure capacity from the public grid while generating minimal permanent local employment — is a well-documented pattern, not a conspiracy theory.
Research from Policy Matters Ohio researcher Molly Bryden has found that community benefits agreements, legally binding arrangements requiring developers to commit on resource use, job creation, workforce training, and local investment, have not been successfully negotiated for any Ohio data center to date.²⁰ That is not because no one thought to ask. It is because the information asymmetry created by NDA-governed negotiations means the community is presented with the terms of a deal that was already finalized.
Policy Matters Ohio points to Virginia not as a model to be uncritically replicated, but as evidence that the same capital can generate categorically different public returns depending on how the tax structure is designed.²¹ Loudoun County has used data center revenue to fund schools, reduce residential property taxes, and invest in public safety infrastructure. Allen County, under Ohio’s current regime, is in line to receive construction-phase employment and decades of tax-exempt occupancy.
The difference is not a matter of which companies decide to be generous. It is a matter of what the legal framework requires them to give back.
Genuine expansion of the tax base, the kind that produces durable fiscal capacity for schools, roads, and local services, requires either high-employment facilities that generate income tax revenue regardless of property tax abatements, or property-intensive facilities whose real property taxes are not abated, or a community benefits agreement with binding commitments on both. Data centers as currently structured in Ohio are designed to avoid all three. Ohio taxes no tangible personal property on data center assets. Its abatement regime exempts real property improvements. Its NDA-governed negotiations prevent communities from knowing what commitments to demand.
The result is not an accident. It is a design — one that a French economist identified in a different context nearly two centuries ago: “Property is not a fact; it is a legal fiction.”²⁴ The fiction here is that the exchange is mutual. It is not.
What the April 9 meeting should be asking
The April 9, 2026 Shawnee Township Zoning Commission hearing on amendment #6172-2026 is an opportunity — however belated — to ask some of what should have been asked before the first shovel broke ground in American Township.
If a second facility is under consideration near the JSMC campus, the Shawnee Township record should establish: what entity is the actual end user of the facility’s compute capacity; what the expected permanent employment will be in year five of operation and year fifteen; what, if any, federal contracts will be served by the facility; and what community benefits the township is prepared to negotiate before any abatement agreements are executed.
None of this requires opposing data center development. It requires treating data center development as a negotiation, not a concession.
I have been clear about what I can establish and what remains inferential. The convergence of the GDIT-Google Public Sector announcement, the IL6 authorization, the Air Force RSO confirmation, JSMC’s position in the GDC air-gapped supply chain, and the anomalous secrecy around Project BOSC’s specifications does not, by itself, prove a defense contract relationship. It raises a question the public record does not yet answer.
But a community that does not know what it has agreed to host cannot negotiate the reciprocal obligations that would make the agreement fair. The information asymmetry is not incidental to the deal structure. For Allen County, and now potentially for Shawnee Township, it has been the mechanism.
The author is an Allen County resident and technology professional specializing in cloud infrastructure governance for regulated industries including healthcare, finance, and commercial real estate. A potential conflict of interest as a recently incorporated local business with partial overlap in edge and backup technologies was publicly disclosed in the original piece:
Notes
¹ Boston Consulting Group, “Cloud Cover: Price Swings, Sovereignty Demands, and Wasted Resources,” August 2025. https://www.bcg.com/publications/2025/cloud-cover-price-sovereignty-demands-waste
² Google Cloud Blog, “Google Distributed Cloud (GDC) & GDC Air-Gapped Appliance Achieve Department of Defense (DoD) Impact Level 6 (IL6) Authorization,” May 28, 2025. https://cloud.google.com/blog/topics/public-sector/google-distributed-cloud-gdc-gdc-air-gapped-appliance-achieve-dod-impact-level-6-il6-authorization
³ Google Cloud Blog, “Google Distributed Cloud Hosted is GA,” March 2023. https://cloud.google.com/blog/products/infrastructure-modernization/google-distributed-cloud-hosted-is-ga
⁴ Google Cloud Documentation, “About Google Distributed Cloud air-gapped appliance.” https://docs.cloud.google.com/distributed-cloud/hosted/docs/latest/appliance/overview; Google Cloud Blog, “Google Distributed Cloud air-gapped appliance is GA,” July 2024. https://cloud.google.com/blog/topics/hybrid-cloud/google-distributed-cloud-air-gapped-appliance-is-ga
⁵ GDIT, “GDIT Successfully Demonstrates Google Cloud Air-Gapped Appliance at the Tactical Edge During the Mobility Guardian 2025 Exercise,” September 2025. https://www.gdit.com/about-gdit/press-releases/gdit-successfully-demonstrates-google-cloud-air-gapped-appliance-at-the-tactical-edge/ The RSO statement quoting Michael Roquemore, Director, Air Force RSO Rapid Agile Integrated Capabilities Team, is reported in Data Center Dynamics, “Google launches air-gapped Distributed Cloud Edge hardware for AI workloads.” https://www.datacenterdynamics.com/en/news/google-launches-air-gapped-distributed-cloud-edge-hardware-for-ai-workloads/
⁶ U.S. Army, “Army rolls out latest version of iconic Abrams Main Battle Tank,” October 2017. https://www.army.mil/article/194952
⁷ U.S. Army, “Commoditizing AI/ML Models,” October 2024. https://www.army.mil/article/280160/commoditizing_aiml_models
⁸ U.S. Army TACOM, “Joint Systems Manufacturing Center – Lima.” https://www.tacom.army.mil/jsmc-lima
⁹ General Dynamics, “General Dynamics Information Technology and Google Public Sector to Deploy Secret AI to Defense Vehicles at Tactical Edge,” November 13, 2025. https://www.gd.com/Articles/2025/11/13/general-dynamics-information-technology-google-public-sector-deploy-secret-ai-defense
¹⁰ Signal Ohio, “Data centers get huge tax breaks while straining utilities,” October 2025. https://signalohio.org/data-centers-have-claimed-2-5-billion-in-tax-breaks-since-2017-report-says/
¹¹ Ibid.
¹² Policy Matters Ohio, “Indefensible Tax Breaks for Data Centers Will Cost Ohio,” January 2025. https://policymattersohio.org/research/indefensible-tax-breaks-for-data-centers-will-cost-ohio/
¹³ Signal Ohio, “Ohio data centers exploit tax break,” July 2025. https://signalohio.org/ohio-data-centers-tax-breaks/
¹⁴ Tax Foundation, “State Taxation of Data Centers,” February 2026. https://taxfoundation.org/research/all/state/data-centers-taxation/
¹⁵ Olney Enterprise, “How Much Revenue Do Data Centers Generate For Local Governments?,” November 2025, citing Loudoun County fiscal impact statements. https://www.olneyenterprise.com/news/how-much-revenue-do-data-centers-generate-local-governments; see also WUSA9 Verify, “Yes, at least 29% of Loudoun County’s projected tax revenue is expected to come from data centers,” 2022. https://www.wusa9.com/article/news/verify/loudoun-county-fiscal-year-2023-budget-data-center-projected-tax-revenue
¹⁶ Loudoun County, Virginia, “FAQs: Data Centers — Tax Revenues and the County Budget.” https://www.loudoun.gov/faq.aspx?TID=241
¹⁷ News5 Cleveland, “Ohio’s spending billions on tax breaks for data centers. Now an incentive battle is brewing,” April 2026. https://www.news5cleveland.com/news/local-news/ohios-spending-billions-on-tax-breaks-for-data-centers-now-an-incentive-battle-is-brewing
¹⁸ Richmond Federal Reserve, “Impacts of Government Spending Changes on Local Economies,” Economic Brief, October 2025. https://www.richmondfed.org/publications/research/economic_brief/2025/eb_25-28
¹⁹ Shoag, D., et al., “Local Fiscal Multipliers and Fiscal Spillovers in the United States,” AEA Conference Paper, 2019. https://www.aeaweb.org/conference/2019/preliminary/paper/2TkbN259
²⁰ WOUB Public Media, “In Ohio’s rural areas, zoning is sometimes a four-letter word. Data centers could change that,” March 2026. https://woub.org/2026/03/30/ohio-rural-areas-zoning-data-centers/
²¹ Policy Matters Ohio, op. cit.
²² Catalini, C., Roche, M.P., and Oettl, A., “Proximate (Co-)Working: Knowledge Spillovers and Social Interactions,” NBER Working Paper No. 30120 / HBS Working Paper 21-024 (2022). Summary: MIT Sloan, “Close proximity is the key to knowledge spillovers among startups.” https://mitsloan.mit.edu/ideas-made-to-matter/close-proximity-key-to-knowledge-spillovers-among-startups
²³ Meissner, D., et al., “Hidden champions and knowledge spillovers: innovation-enhancing agglomeration effects and niche technology specificity,” Small Business Economics (2025). https://link.springer.com/article/10.1007/s11187-025-00999-3
²⁴ Pierre-Joseph Proudhon, What Is Property? Or, an Inquiry into the Principle of Right and Government (1840), Ch. 4. The full formulation: “Property is not a fact; it is a legal fiction… it only exists in virtue of the law.” Proudhon’s distinction between possession earned through active use and property sustained by legal title that permits extraction without obligation maps precisely onto the abatement structure described here.