Data Centers, Demystified: A Field Guide for Lodi

LodiEye · Explainer

Data Centers, Demystified: A Field Guide for Lodi

The buildings behind the apps you already use every day — and how Google, Amazon, Microsoft, Apple, Meta, and the new wave of AI operators differ from one another.

The invisible buildings behind your phone

When you check Instagram with your morning coffee, ask Siri for tomorrow’s forecast, stream a movie on Netflix tonight, or tap your card at the gas station on Cherokee Lane, something invisible happens. Your phone or the payment terminal sends a message to a building you have never seen, often hundreds or thousands of miles away. A few milliseconds later, that building sends an answer back. The building is a data center, and there are now several thousand of them scattered across the United States.

For most of the internet’s history, those buildings were not something the average user thought about. They worked. They stayed in the background. Recently they have started showing up in local news more often, partly because they consume enormous amounts of electricity and water, and partly because a new category of facility, specifically engineered to train artificial intelligence models, did not really exist five years ago and now does. That category is growing faster than any other type of industrial construction in the country.

This is a field guide to what those buildings actually do, who builds them, and how the major operators differ from one another. It is written for people who use the products coming out of them every day but have not had a reason to think about the buildings themselves.

So what is a data center, exactly?

Strip away the jargon and a data center is a single, purpose-built warehouse for computers. Inside are long rows of metal cabinets, each filled with rack-mounted servers, storage arrays, and networking equipment. Those machines run software that, depending on the operator, might be holding your Netflix watch history, processing the credit card transaction you just ran at Lodi Hardware, training an artificial intelligence model, or simply storing the photos in your iCloud library. The building exists because that hardware needs three things consistently and at industrial scale: electricity to run, cooling to keep it from overheating, and fiber-optic cables to move data in and out.

Three universal constraints follow from those needs. First, power. A medium-sized data center draws tens of megawatts, comparable to a small city; a large hyperscale campus can draw hundreds of megawatts, comparable to an aluminum smelter or a small steel mill. That is why utility capacity and transmission lines drive site selection more than almost anything else. Second, cooling. Most traditional facilities use chilled water and large air-handling units; newer AI-focused buildings increasingly pump liquid coolant directly to the chips, because modern processors run hot enough that air alone cannot keep up. Third, fiber. A data center is only as useful as its connection to the outside world. The best sites sit on multiple long-haul fiber routes with diverse paths to major peering points in the Bay Area and Sacramento.

Everything else — tax incentives, labor pools, seismic risk, climate — matters at the margin. Power, cooling, and fiber are the three constraints that decide whether a site is even feasible.

The operators most people have heard of

Most residents who follow technology news have heard the names Google, Amazon, Microsoft, Apple, Meta, and a growing list of AI companies. Those names refer to very different kinds of buildings, built for very different reasons. The simplest way to understand the differences is to look at which apps and services from your daily life come out of each one.

Google Hyperscale Cloud + Own Services

Google owns more of its own data centers than any other major cloud provider, with 23 campuses in 15 US states. Its facilities run two things in parallel: the company’s own services and Google Cloud, which sells computing capacity to outside businesses. Google has designed its own custom AI chips, called TPUs, and runs most of its training and inference workloads on them rather than on general-purpose hardware. It is also the cloud provider that has invested most aggressively in long-term renewable power agreements, often pairing each new campus with a dedicated solar or wind project.

What runs here: Google Search, YouTube, Gmail, Google Maps, Google Photos, the Play Store, Android phone syncing, Google Drive and Docs, Waze, Nest thermostats, and the back-end services for any business that runs on Google Cloud.

Nearest Google region to Lodi: us-west2 in Los Angeles. Google has no owned data center in the Sacramento or Bay Area region.

Amazon Web Services (AWS) Hyperscale Cloud

AWS is the largest cloud provider in the world, holding roughly a third of the global market. Unlike Google, AWS is almost entirely a business-to-business operation: it sells raw computing power, storage, and networking services to other companies large and small. That means an enormous share of the consumer internet runs on AWS without consumers ever seeing the name. AWS is notoriously secretive about facility locations, but its US West (Northern California) region, called us-west-1, is spread across multiple buildings in Santa Clara, San Jose, and surrounding Bay Area cities. The company has invested heavily in custom chips of its own, branded Graviton for general-purpose computing and Trainium for AI training.

What runs here: Netflix, Disney+, Hulu, Airbnb, Pinterest, ESPN streaming, the NFL app, Peloton, Robinhood, Lyft, Slack, parts of Zoom, the Ring video doorbell, and the back-end systems for many banks, hospitals, insurance companies, and government agencies. If you have streamed a movie in the last month, AWS almost certainly delivered some of it.

Nearest AWS region to Lodi: us-west-1, distributed across the Bay Area roughly 80 miles southwest.

Microsoft Azure Hyperscale Cloud + AI

Microsoft’s data center footprint serves both its enterprise cloud business and, increasingly, its partnership with OpenAI, which runs ChatGPT and the GPT family of models on Microsoft infrastructure. The company’s primary western US region is anchored in Quincy, Washington, but Microsoft is actively building a new campus in the Alviso neighborhood of North San Jose and has filed plans for an Elk Grove site in the Sacramento area. Microsoft has also begun rolling out a class of AI-specific facilities branded “Fairwater,” designed around very high power density and direct liquid cooling.

What runs here: ChatGPT and the GPT API, Microsoft 365 (Outlook, Word, Excel, Teams), LinkedIn, Xbox Live and Game Pass, Bing search, Skype, GitHub, Minecraft online services, and the business systems many Lodi employers use behind the scenes.

Nearest Microsoft region to Lodi: West US (Bay Area) for Azure customers, with the San Jose Alviso campus under construction.

Apple Enterprise / Single-Tenant

Apple is the outlier among the consumer tech giants. It does not sell cloud capacity to anyone else; its data centers exist almost entirely to run Apple’s consumer services for Apple’s customers. Apple operates five owned facilities in the US, in Newark, California; Reno, Nevada; Maiden, North Carolina; Prineville, Oregon; and Mesa, Arizona, with another under construction near Des Moines. To supplement that capacity, Apple is one of the largest paying customers of both AWS and Google Cloud. The Apple footprint is therefore both visible (its own buildings) and invisible (capacity it rents from competitors).

What runs here: iCloud (your iPhone photos, contacts, calendar, and messages backup), iMessage and FaceTime, Siri, the App Store, Apple Music, Apple TV+, Apple Pay, Find My iPhone, and the device-to-device handoff features that move a phone call from your iPhone to your iPad.

Nearest Apple facility to Lodi: Newark, California, about 65 miles southwest. Apple sold the building to T5 Data Centers in 2020 but remains the tenant.

Meta Hyperscale Single-Tenant + AI

Meta sits in an unusual place on this map. Like Apple, it does not sell cloud capacity to outside customers; all of its data centers serve Meta’s own products. Unlike Apple, Meta operates at hyperscale, with 32 data centers in operation or under construction as of April 2026. Its Prineville, Oregon campus, opened in 2011, was where Meta developed and open-sourced the Open Compute Project, a hardware standard now used widely across the industry, including by some of the other operators on this list. Meta’s newer announcements are almost all described as AI-optimized, including multibillion-dollar campuses in Louisiana, Ohio, and Tennessee branded under names like Prometheus and Hyperion.

What runs here: Facebook, Instagram, WhatsApp, Threads, Facebook Messenger, Facebook Marketplace, the Meta Quest virtual reality platform, the Ray-Ban smart glasses cloud services, and the training of Meta’s Llama family of AI models.

Nearest Meta facility to Lodi: Prineville, Oregon, about 520 miles north. Meta operates no data center in California.

AI-Focused Operators New Category

The AI category is the newest and the most disruptive part of the industry. It includes specialist cloud providers such as CoreWeave, Crusoe, Lambda, and Nebius, alongside AI-dedicated wings of the hyperscalers and tenants that lease entire buildings from operators like Prime Data Centers. What distinguishes AI facilities is density. A traditional cloud server rack draws roughly 5 to 15 kilowatts of power; an AI rack built around modern Nvidia hardware draws 50 to 120 kilowatts — meaning the same building footprint consumes five to ten times the electricity. That density forces a shift from air cooling to liquid cooling, often piping coolant directly to each chip. A nearby example: Prime Data Centers opened a 33-megawatt facility in Vernon, near Los Angeles, that is one hundred percent leased to two unnamed AI companies.

What runs here: ChatGPT (via Microsoft Azure’s AI halls), Anthropic’s Claude, Google Gemini, Midjourney and other image generators, voice transcription services like Otter, the AI features built into Photoshop and Microsoft Word, and the rapidly growing list of AI assistants embedded in everyday consumer apps.

Nearest AI-tenanted facilities to Lodi: Prime SMF02 in Sacramento (under construction) and the Vernon site near Los Angeles, with new AI-specific capacity also coming online across the Bay Area.

How the categories stack up

The differences between these operators are easier to see side by side. The table below summarizes the practical distinctions, including the rough power footprint each one carries.

Operator Type	Typical Size	Density (kW/rack)	Primary Workload	Job Footprint
Single-Tenant (Apple, Meta)	10–500+ MW	5–80	Own consumer services	Low (100–400 ongoing per campus)
Hyperscale Cloud (AWS, Azure, Google)	50–500 MW per campus	10–25	Cloud services for many customers	Low to moderate (50–200 ongoing per campus)
Retail / Wholesale Colocation	5–200 MW	5–30	Leased to many tenants	Low (20–100 ongoing)
AI / GPU Facility	30–500+ MW	50–120	AI training and inference	Low (20–80 ongoing); higher construction phase
Edge / Modular	0.5–5 MW	5–20	Low-latency local apps	Very low (under 20)

The headline takeaway is that very few data centers create large numbers of permanent jobs. Construction is a different story — building a large facility involves hundreds or even thousands of trade workers for one to three years — but once a data center is operational, it runs with a small team of facilities engineers, security personnel, and remote-hands technicians. The economic value of a data center, where it exists, comes mainly from utility revenue, property taxes, and lease income rather than from headcount.

A tour of the neighborhood

Explore Further

Lodi411 Northern California Data Center Map — an interactive map of operational and planned data centers across Northern California, including the facilities described in this section.
lodi411.com/ca-datacenters

Lodi is closer to a working data center than most residents realize. The region already hosts several, each illustrating a different model.

Nautilus Stockton (about 12 miles south)

Sitting at the Port of Stockton on the San Joaquin River is one of the more unusual data centers in the country: a floating facility on a barge. Operated by Nautilus Data Technologies, the Stockton site uses river water in a closed-loop heat exchanger to cool its 6.5 megawatts of computing capacity, eliminating the cooling towers and chillers found at conventional sites. It is small by hyperscale standards, but it is a working example of how cooling design can reshape water use. It is also a working example of a multi-tenant colocation facility, meaning multiple businesses lease space in the same building.

Iron Mountain Tracy (about 35 miles south)

In Tracy, Iron Mountain operates a more traditional colocation campus — a facility that leases space and power to many different tenants, from regional businesses to hyperscale cloud providers, with direct on-ramps to AWS, Microsoft Azure, and Google Cloud. This is the most common type of data center in the country: not owned by a tech giant, but rented out, rack by rack or megawatt by megawatt, to whoever needs space.

Prime Data Centers, Sacramento (about 50 miles north)

At McClellan Park in Sacramento, Prime Data Centers operates SMF01 and broke ground on a second building, SMF02, earlier this month. The new building will add 18 megawatts of capacity, explicitly designed for what the operator describes as AI-ready hyperscale workloads. This is the model most likely to expand in Northern California over the next five years: large, purpose-built, ready for liquid-cooled equipment, and leased to one or more anchor tenants.

Apple Newark (about 65 miles southwest)

In Newark, on the southeast edge of the Bay Area, sits a 128,000-square-foot data center that Apple acquired in 2006 and sold to T5 Data Centers in 2020. Apple remains the tenant, using the site to help power iCloud, iMessage, Siri, and the App Store. With expansion plans, the campus is permitted for up to roughly 49 megawatts of critical power, modest by current standards but a useful local reference point for a single-tenant enterprise facility.

AWS, Santa Clara and San Jose (about 80 miles southwest)

Amazon Web Services does not publicly map its facilities, but its US West (Northern California) region, known as us-west-1, is spread across multiple buildings in Santa Clara, San Jose, and surrounding Bay Area cities. These are the buildings serving most California-based businesses, app developers, and consumer services when they select Northern California as their cloud region. The next time a streaming service buffers slightly faster than usual on a Lodi network, there is a reasonable chance the content was delivered from one of these sites.

A closer look at White Slough

For readers wondering why an inland California site might attract a data center operator, it is worth understanding what makes the White Slough area, west of Interstate 5 outside Lodi city limits, distinctive. The site is roughly 1,040 acres and brings together three things data center operators look for and almost never find together in one place: large volumes of non-potable water, a multi-agency electrical generation hub with room to grow, and industrial-buffered land that is poorly suited to residential or agricultural development.

Water

The most water-intensive function inside a data center is cooling. In drought-affected regions, operators are increasingly required to show that they will not draw heavily on potable water supplies. The White Slough Wastewater Treatment Facility produces large volumes of treated water that, by design, is not suitable for drinking. Recycled water of that kind is well-suited to industrial cooling loops, including the closed-loop systems used by modern data center designs — the same approach Nautilus uses at its floating Stockton facility downriver. Treated wastewater in California is increasingly a managed commodity, both a regulatory obligation for the agency that produces it and a sought-after input for any industrial user that can substitute it for potable supply.

Power

The same site hosts the Lodi Energy Center, a natural gas plant co-owned by the City of Lodi and the Northern California Power Authority, a joint-action agency representing multiple Northern California municipal utilities. That co-ownership is unusual: most Central Valley sites are served by a single investor-owned utility, often with multi-year wait times to connect new industrial loads. The White Slough complex, by contrast, generates power on-site and has reported headroom in current operations.

The more consequential point for any future large industrial user is that White Slough has the physical and regulatory footprint to expand generation capacity further, independently of two other Lodi initiatives already in motion. The city has a planned 230 kV transmission project that will strengthen its broader grid connections, and a separate substation buildout sized to serve projected residential and industrial growth. Adding generation at the Lodi Energy Center site is a different matter from either: additional turbines, or future turbines specified for hydrogen co-firing, would not depend on, nor compete with, those parallel transmission and distribution projects. Modern combined-cycle natural gas turbines are increasingly being designed or retrofitted to run on hydrogen blends, a pathway that would extend the carbon profile of the site to match the procurement standards that many cloud and AI operators are now required to meet. The combination — on-site generation already operating, room to add more independent of Lodi’s other plans, and a credible decarbonization pathway — is rare in California.

Land

Approximately 220 acres at White Slough remain available for development, with an estimated 130 of those potentially designated for a proposed PG&E research and development site, leaving the remainder for other industrial uses. The acreage that is not the treatment plant or the power plant sits in the buffer zone around them — land that is industrial-adjacent, outside city limits, and poorly suited to residential or commercial growth because of its proximity to a wastewater facility. From an operator’s perspective, that kind of buffer parcel is actively desirable: it reduces the likelihood of neighbor conflicts over noise, generator testing, and 24-hour lighting, and it is typically available at lower per-acre prices than land closer to existing development. The same buffer characteristics that make the site less attractive for housing or farming are what make it more attractive for an industrial user looking for room to operate.

Those three things — water that is already non-potable, power that already generates and has room to grow, and land that sits outside the residential and agricultural development pattern — explain why operators searching for inland California sites might view a location like White Slough as structurally interesting. The same three considerations, in different proportions, apply to every data center siting decision being made in the state right now.

Why the physical building matters

It is easy to think of digital services as something abstract, a process that happens “in the cloud.” The cloud is a useful metaphor, but it is also a misleading one. Every photo backed up to iCloud sits on a physical disk in a physical building. Every Google search travels down a physical fiber cable. Every minute of a Netflix movie is streamed from a server in a specific city. The buildings these services run from are real, they sit on real land, they draw real water and real electricity from real utilities, and the choices their operators make — about where to put them, how to cool them, and what to power them with — have real effects on the places that host them.

Explore Further

OpenGridWorks Data Center and Infrastructure Map — a national view showing every US data center alongside the electric transmission grid and network infrastructure they depend on. A useful way to see how the physical layers fit together.
opengridworks.com

That is the part of the story that is now becoming more visible in California. The state has roughly 288 operating data centers, with the heaviest concentrations in Santa Clara County, the broader Bay Area, and a fast-growing cluster in Sacramento. As demand for AI computing accelerates and the Bay Area approaches the limits of available electricity, operators are beginning to look at sites further inland, where land is cheaper and power, in some cases, is more available. That is the trend driving much of the discussion about future data center development in the Central Valley. None of which changes what the buildings do. They run the apps, the searches, the photos, the streams, and the AI assistants that have quietly become a constant part of daily life — mostly out of sight, until they are not.

About This Report

LodiEye is a civic-data and analysis project. It does not replicate the primary reporting of the Lodi News-Sentinel, Stocktonia, the Sacramento Bee, or CalMatters. It synthesizes regional and statewide reporting, adds public-record context, and explains technical and policy material that local readers may not otherwise encounter in one place.

This explainer drew on five capacities:

• Source Discovery — locating public disclosures, regulatory filings, and trade-press coverage on data center capacity, locations, and which consumer services run from which operators.
• Credibility Validation — cross-checking operator capacity figures, regional cloud-region assignments, and customer-service relationships against company sources and reputable industry references.
• Analysis and Synthesis — assembling the comparative framework across operator types, density, workload, and job footprint, and tying each to the consumer services they host.
• Presentation — structuring the piece for general readers, producing the comparative summary table, and drafting the operator profiles.
• Final Review — editorial review by LodiEye’s founder and editor.

Research and drafting used Claude (Anthropic) and Perplexity as analytic tools. All factual claims were checked against named sources, and all editorial decisions were made by the editor.

Corrections, additions, or source tips: editor@lodi411.com