Layered-Labs/nyc-clinic-ai-infrastructure

🏥 NYC Clinic AI Infrastructure Dataset

Can clinics run local AI? This dataset answers that question for every ZIP code in New York City.

Published by a nonprofit researching open-source, on-premise AI deployment for community health clinics.

Coverage	311 ZIP codes · all 5 NYC boroughs
Dimensions	Broadband access · Electricity cost · Grid reliability
Data vintage	FCC NBM Jun 2025 · Census ACS 2022 · EIA Form 861 2024
License	CC BY 4.0
Explore	🗺️ Interactive Map

---

Why This Dataset Exists

Running AI models locally on a clinic's own hardware can reduce costs, eliminate protected health information (PHI) transmission to third parties, and provide AI functionality even in low-connectivity environments. Whether a clinic can viably do this depends on three infrastructure questions:

1. Internet connectivity: reliable broadband is needed for model updates and telemetry, even when inference runs fully offline
2. Electricity cost: inference hardware running 24/7 adds to operational costs
3. Grid reliability: power outages can corrupt model state and interrupt patient-facing tools

This dataset was built to help clinics, researchers, and policymakers evaluate where local AI deployment is most feasible and where infrastructure gaps create barriers.

---

Key Findings

🔌 Electricity: A Non-Issue Citywide

Con Edison serves all 5 boroughs with exceptional reliability:

Metric	Value
SAIDI (excl. major events)	14.9 min/year
SAIDI (incl. major events)	27.8 min/year
SAIFI	0.13 outages/year
Reliability tier	Excellent (EPRI benchmark)

Running a local AI model is also inexpensive at NYC commercial rates (17.82 ¢/kWh):

Hardware	Power	Monthly cost
Mac Mini M4 (Llama 3.2 3B)	35W	$4.56/mo
GPU workstation (Mistral 7B)	350W	$45.57/mo
Dual-GPU server (Llama 3 70B)	900W	$117.17/mo

📡 Broadband: The Real Barrier

Fixed broadband (cable/fiber/DSL) access varies dramatically across boroughs:

Borough	Avg Fixed Broadband %	Avg No-Internet %
Manhattan	~83%	~5%
Queens	~76%	~10%
Brooklyn	~70%	~13%
Staten Island	~72%	~8%
Bronx	~66%	~14%

South Bronx ZIPs (10452, 10454, 10455) have the city's lowest fixed broadband rates at 55-63%, while being home to some of NYC's most underserved health populations. This is where local, offline-capable AI matters most.

🌐 Broadband Infrastructure

Despite household subscription gaps, FCC infrastructure coverage is near-universal:

• Fiber-to-premises available: 197/197 ZCTAs with FCC data
• Cable available: 196/197 ZCTAs
• Average ISPs per ZIP: 6.0

The gap is between infrastructure availability and household subscription: a cost and digital equity problem, not an infrastructure one.

---

Dataset Schema

Column	Type	Description	Source
zipcode	string	5-digit ZIP code	GeoNames/TIGER
borough	string	NYC borough	GeoNames/TIGER
neighborhood	string	Neighborhood / place name	GeoNames/TIGER
lat, lon	float	ZIP centroid coordinates	GeoNames/TIGER
census_hh_total	int	Total households	Census ACS 5yr 2022
census_hh_any_subscription	int	HH with any internet subscription	Census ACS B28002_002E
census_hh_fixed_broadband	int	HH with cable/fiber/DSL	Census ACS B28002_007E
census_hh_no_internet	int	HH with no internet access	Census ACS B28002_013E
census_fixed_broadband_pct	float	% HH with fixed broadband	Derived
census_no_internet_pct	float	% HH with no internet	Derived
fcc_isp_count	int	Unique ISPs offering service in ZIP	FCC NBM Jun-2025
fcc_max_dl_mbps	int	Max advertised download speed (Mbps)	FCC NBM Jun-2025
fcc_max_dl_mbps_practical	int	Max DL speed capped at 10 Gbps	FCC NBM (derived)
fcc_max_ul_mbps	int	Max advertised upload speed (Mbps)	FCC NBM Jun-2025
fcc_max_ul_mbps_practical	int	Max UL speed capped at 10 Gbps	FCC NBM (derived)
fcc_location_count	int	Unique addressable locations with service	FCC NBM Jun-2025
fcc_cable_available	bool	Cable (HFC/DOCSIS) ISP present	FCC NBM Jun-2025
fcc_fiber_available	bool	Fiber-to-premises ISP present	FCC NBM Jun-2025
fcc_copper_available	bool	Copper/DSL ISP present	FCC NBM Jun-2025
utility_name	string	Primary electric utility	EIA Form 861 2024
commercial_rate_cents_kwh	float	Commercial electricity rate (¢/kWh)	EIA API 2023
residential_rate_cents_kwh	float	Residential electricity rate (¢/kWh)	EIA API 2023
monthly_cost_mac_mini_m4_usd	float	Monthly cost: 35W device, 24/7 ($)	Derived
monthly_cost_workstation_gpu_usd	float	Monthly cost: 350W GPU, 24/7 ($)	Derived
monthly_cost_dual_gpu_usd	float	Monthly cost: 900W dual-GPU, 24/7 ($)	Derived
saidi_with_major_events_min_yr	float	SAIDI incl. major events (min/yr)	EIA Form 861 2024
saidi_excl_major_events_min_yr	float	SAIDI excl. major events (min/yr)	EIA Form 861 2024
saifi_with_major_events_per_yr	float	SAIFI incl. major events (events/yr)	EIA Form 861 2024
saifi_excl_major_events_per_yr	float	SAIFI excl. major events (events/yr)	EIA Form 861 2024
caidi_with_major_events_min	float	CAIDI incl. major events (min/outage)	EIA Form 861 2024
reliability_tier	string	excellent / good / fair / poor	EPRI benchmarks
data_sources	string	Full provenance string	n/a

Notes:

• fcc_* columns cover 197 of 311 ZIPs (ZIPs without Census ZCTAs have NaN broadband columns; ZIPs without residential population have NaN census columns)
• fcc_max_dl_mbps includes carrier-grade dark fiber (e.g. Zayo at 1.2 Tbps, business-only); use fcc_max_dl_mbps_practical (capped at 10 Gbps) for clinic-relevant speeds
• Electricity metrics are uniform citywide (all ZIPs served by Consolidated Edison Co-NY Inc)

---

Reproducing the Dataset

All data sources are free and public. The pipeline is fully automated.

Prerequisites:

pip install -r requirements.txt
export EIA_API_KEY=your_key   # free at https://www.eia.gov/opendata/register.php

Step 1: Base dataset (Census ACS + EIA rates + EIA-861 reliability):

python build_dataset.py
# Output: outputs/nyc_clinic_infrastructure.csv

Step 2: Add FCC broadband data:

1. Create a free account at broadbandmap.fcc.gov
2. Download for New York state: Cable, Fiber to the Premises, Copper (Fixed Broadband)
3. Place CSVs in requested/

python join_fcc.py
# Updates: outputs/nyc_clinic_infrastructure.csv

Data sources accessed programmatically (no account needed):

• US Census ACS 5-year API: api.census.gov
• EIA Open Data API: api.eia.gov
• EIA Form 861 bulk download: eia.gov/electricity/data/eia861
• Census ZCTA-Tract crosswalk: www2.census.gov/geo/docs/maps-data/data/rel2020/zcta520

---

Citation

If you use this dataset in research, please cite:

@dataset{nyc_clinic_ai_infrastructure_2025,
  title     = {NYC Clinic AI Infrastructure Dataset},
  author    = {{Layered-Labs}},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Layered-Labs/nyc-clinic-ai-infrastructure},
  note      = {FCC NBM Jun-2025, Census ACS 2022, EIA Form 861 2024}
}

---

License

CC BY 4.0. You are free to share and adapt this dataset for any purpose, provided you give appropriate credit.

Underlying data is from US government sources (FCC, Census Bureau, EIA) and is in the public domain.