Oakland’s STAK AI Campus Faces a Data Center Identity Test

STAK AI Labs is promoting a 1.4-million-square-foot Oakland campus with housing, workspaces, robotics facilities, and up to 20 megawatts of data center capacity.

That combination creates the central conflict. The developer and an Oakland councilmember say the project differs from a conventional data center, according to a local news report. Yet the developer’s own materials explicitly advertise data center power, GPU colocation, enterprise cooling, and continuous infrastructure monitoring.

The distinction matters because “AI campus” describes an economic vision, while “data center” describes physical infrastructure. Those labels lead residents toward very different expectations about jobs, electricity, water, noise, security, and public oversight.

Oakland therefore faces a practical test, not a branding debate. Officials must determine which uses are already operating, which remain proposals, and which impacts require additional review.

What STAK AI Labs Is Actually Proposing

The project is broader than a standalone server facility, but data center infrastructure remains one of its central components.

STAK AI Labs presents the development as a connected campus serving artificial intelligence and robotics companies. Its public materials combine office space, residences, compute capacity, networking, training data, and startup programs.

The campus spans four nearby properties in downtown Oakland. These include 1900 Broadway, 1950 Franklin Street, 415 20th Street, and a parking structure at 1901 Franklin Street.

The residential anchor is 1900 Broadway, a 39-story tower with 452 units. It also contains offices, coworking areas, event spaces, fitness facilities, and other shared amenities.

The developer describes 1950 Franklin as another workplace component. The parking property supplies vehicle storage and transportation infrastructure for the larger group of buildings.

The most consequential property is 415 20th Street. STAK identifies it as a data center and carrier hotel, meaning a facility where networks and infrastructure customers can interconnect.

According to the company’s campus specifications, the site offers GPU colocation racks, redundant fiber, enterprise cooling, private network suites, and continuous operations monitoring. Those are established data center functions.

Colocation means customers install computing equipment inside infrastructure operated by another company. The operator supplies power, cooling, connectivity, security, and physical space.

STAK says the 415 20th Street building has 20 megawatts of potential capacity. Behring Co., the property developer behind the campus, has separately described 4.5 megawatts as immediately available.

The difference between immediate and potential capacity is important. Available electrical infrastructure does not necessarily mean every megawatt is active, contracted, or approved for a new use.

STAK also promotes approximately 500 desks, a 1,400-space parking facility, and robotics data collection capabilities. Its materials mention more than one million annotated commercial real estate task clips.

Those clips are intended to help train embodied AI systems. Embodied AI refers to software operating machines that perceive and act within physical environments, including service robots.

This part of the proposal differs substantially from a remote hyperscale facility. A hyperscale data center usually dedicates enormous buildings to computing infrastructure for a small number of major cloud operators.

STAK instead describes an urban cluster where startups can work near compute, testing spaces, housing, investors, and potential collaborators. Its thesis is that proximity makes developing physical AI systems easier.

The company also advertises grants totaling up to one million dollars for qualifying companies. Applicants can reportedly receive combinations of workspace, computing access, research support, and training data.

These features give the project a wider purpose than server hosting alone. However, they do not eliminate the campus’s data center component.

A more accurate description is a mixed-use AI campus containing a data center. That phrasing recognizes both the broader development plan and its resource-intensive infrastructure.

The project’s physical history strengthens that interpretation. Behring says 415 20th Street previously supported high-performance computing associated with Lawrence Berkeley National Laboratory.

Existing power and network connections can make an older computing property attractive for AI infrastructure. They reduce some barriers that confront developers seeking entirely new sites.

Oakland’s planning records also show that 415 20th Street sits in City Council District 3. The property lies within the central business district near the 19th Street BART station.

That setting is unusual for a modern AI computing project. Most new data center proposals target industrial zones or large peripheral sites with space for substations and cooling equipment.

STAK is attempting something different. It wants computing infrastructure to operate beside apartments, offices, public transit, and startup programs within an existing downtown neighborhood.

That model could make the campus economically and operationally distinct. It also makes accurate disclosure more important because nearby residents experience the project differently from customers using its servers.

Why the Data Center Label Still Matters

A campus can contain many uses and still require scrutiny based on the most resource-intensive use inside it.

Calling STAK a campus explains how its buildings relate to one another. It does not answer questions about electrical load, cooling, backup generation, construction, or operating noise.

Those effects depend on equipment and utilization. A mostly empty server room has a different footprint from continuously occupied GPU racks drawing several megawatts.

Graphics processing units, or GPUs, perform the parallel calculations used to train and run many AI systems. Dense groups of these chips produce considerable heat and require specialized cooling.

STAK’s potential 20-megawatt load is far below the capacity proposed for some hyperscale campuses. Large projects elsewhere can seek hundreds of megawatts or more.

However, 20 megawatts is not trivial in a dense urban district. It equals 20 million watts of potential demand before considering how consistently the equipment operates.

The crucial number is actual load, not headline capacity. Oakland needs to know how much power the site currently draws and what upgrades would activate additional capacity.

The same principle applies to water. Some facilities use evaporative cooling, while others employ air cooling, closed-loop liquid systems, or hybrid designs.

Each approach creates a different balance between water consumption, electricity demand, equipment density, and outside noise. A campus label reveals none of those choices.

Backup power is another unresolved issue. Data centers commonly maintain generators or battery systems to protect customers during grid interruptions.

The equipment type, testing schedule, emissions controls, fuel storage, and proximity to neighboring buildings determine the local impact. These details cannot be inferred from branding.

National energy forecasts explain why communities now examine such projects closely. A 2026 Berkeley Lab analysis estimates that data centers could consume 11.8 percent of United States electricity by 2030.

The report gives a wider range between 9.5 and 15.3 percent. Its reference case estimates annual data center consumption of 649 terawatt-hours in 2030.

Those national figures do not establish STAK’s local footprint. They show why disclosed capacity now attracts attention from utilities, regulators, and residents.

AI hardware also changes the operating profile of computing facilities. Dense accelerators can require more electricity and cooling per rack than conventional enterprise servers.

A GPU colocation facility may therefore resemble neither an old corporate server room nor a newly constructed hyperscale campus. It still belongs within the data center category.

The developer’s own language supports that conclusion. Behring has described 415 20th Street as suitable for high-performance computing and data centers.

In a property description, Behring cites 4.5 megawatts of immediate capacity and a path toward 20 megawatts. It also emphasizes the building’s power and connectivity.

That does not mean the entire four-property campus should be regulated as one server building. It means officials should evaluate each component according to its actual function.

The mixed-use tower contains hundreds of homes. Coworking areas can bring employees and events downtown. Robotics testing can support different jobs than routine facility operations.

These distinctions are relevant to economic development. They should appear in a complete assessment rather than replace discussion of the computing infrastructure.

The project’s strongest case rests on specificity. STAK can show exactly which floors host servers, which host workers, and which support residents or public-facing activity.

It can also disclose expected loads at different occupancy stages. A phased estimate would separate current operations from future capacity that depends on tenants or infrastructure investment.

Without those figures, residents must choose between competing labels. One suggests a lively technology district, while the other evokes an anonymous industrial facility.

Neither image is sufficiently precise. Oakland needs an operating profile grounded in measurable equipment, power, water, emissions, noise, and employment.

Oakland Must Separate the Campus Promise From Its Infrastructure Reality

The central tension is not whether STAK counts as a campus or a data center. It is whether its broader amenities materially change the infrastructure consequences.

STAK’s urban model offers plausible benefits. Startups could lease offices near computing capacity instead of assembling separate workplace, laboratory, and server arrangements.

Robotics teams could collect physical-world data, test machines, and process results within the same district. Founders could meet investors and technical collaborators without leaving the campus.

The 19th Street BART station gives the project a transit advantage uncommon among data centers. Employees and visitors can reach the site without depending entirely on cars.

Existing residential units also distinguish STAK from a fenced industrial complex. The campus is designed to combine living and working functions within several connected properties.

Behring’s broader development materials describe 1900 Broadway as a mixed-use property built around housing, coworking, services, and shared transportation.

These elements could generate more neighborhood activity than a conventional data center. They could also support more varied employment across hospitality, property operations, events, technical work, and security.

Yet economic activity must be demonstrated, not assumed. Oakland should ask how many jobs already exist, how many are projected, and how many require local hiring.

Construction jobs should be reported separately from permanent positions. Startup tenants should also be distinguished from employees who merely use shared desks occasionally.

The grant program deserves similar clarification. The developer advertises up to one million dollars in support, but that headline can represent cash, services, space, or computing credits.

Each form has different economic value. Public reporting should identify committed recipients, award terms, performance milestones, and the portion delivered as direct funding.

The campus’s robotics angle also needs evidence. STAK says it can collect multimodal data through cameras, LiDAR, and other sensors in commercial environments.

LiDAR measures distances using reflected laser light. Robotics developers use it to map spaces and help machines understand their surroundings.

That work could turn ordinary buildings into testing environments for service robots. It also creates questions about consent, data retention, security, and the people captured during collection.

The company should explain whether data collection occurs in staged environments or occupied spaces. It should also state how participants are informed and how identifying details are handled.

These questions are not arguments against robotics research. They are part of determining what an operational AI campus means for residents, workers, and visitors.

Oakland should apply the same precision to the claimed relationship among the properties. The developer calls them one operating platform, but they remain separate physical assets.

Officials need to know which services are genuinely integrated. Shared marketing, membership programs, data networks, energy systems, and ownership relationships have different regulatory implications.

The 415 20th Street property also has its own planning history. An earlier planning document identified the location, zoning, ownership structure, and prior development application.

That older application does not automatically answer questions about the current AI use. It shows why the city should connect present operations with the property’s approved permits and historical reviews.

A building previously used for computing may already support some data center activity. Expanding its electrical load or changing equipment could still trigger new technical or environmental questions.

The decisive issue is the operational delta, meaning the measurable difference between an established use and the proposed future use.

If STAK is primarily reoccupying existing offices and computing space, its impact could be smaller than a new construction project. If it activates substantial new capacity, the analysis changes.

The public conversation therefore needs a baseline. Oakland should document existing power demand, cooling equipment, backup systems, occupancy, noise, and approved use.

It can then compare that baseline with each proposed development phase. This approach protects the city from both exaggerating and understating the project.

It also gives the developer a fair opportunity to substantiate its central argument. A genuinely integrated campus should be able to demonstrate benefits beyond rebranding a server property.

The Unanswered Questions Will Decide Public Trust

STAK has published an ambitious vision, but several facts needed to evaluate that vision remain unclear.

The first question concerns project status. Public-facing materials sometimes describe current facilities and future capacity within the same presentation.

Readers can therefore struggle to distinguish operating services from planned offerings. Oakland should identify which components are occupied, under construction, approved, or still conceptual.

The second question is electrical capacity. The developer cites both 4.5 megawatts of immediate capacity and a 20-megawatt opportunity.

Officials should determine what “immediate” means. It could describe existing electrical infrastructure, deliverable utility service, available tenant capacity, or marketing potential.

The route from 4.5 to 20 megawatts also matters. It may require utility work, new switchgear, transformers, cooling changes, or additional permits.

Any required upgrades should be disclosed with estimated schedules and responsibility for their costs. Ratepayers should not be left guessing whether a private expansion affects shared infrastructure.

The third question concerns cooling. STAK advertises enterprise systems but does not provide enough public technical detail to calculate water or energy performance.

Useful disclosures would include cooling technology, expected water use, design efficiency, operating temperatures, and performance under peak summer conditions.

Noise measurements should cover normal operations and emergency equipment testing. They should be taken at nearby residences, not only inside the facility boundary.

The fourth question involves backup power. Residents need to know whether the site uses generators, batteries, or another resilience system.

Generator fuel, emissions controls, testing frequency, and operating limits should be available. Battery installations raise separate questions about fire protection and emergency response.

The fifth question is job creation. The campus vision emphasizes builders, startups, investors, and technical collaboration.

Oakland should request permanent job estimates by building and activity. Those estimates should distinguish current employment from positions dependent on future tenants.

A downtown data center component might generate fewer routine jobs than similarly sized offices. The mixed-use and startup portions could offset that pattern, but only verified occupancy can establish the result.

The sixth question is data governance. A robotics training operation handling video, depth information, and environmental scans needs clear collection rules.

The company should explain who owns captured data, how long it is retained, and whether clients can combine it with other information.

Workers and residents also need a clear way to avoid participating. Consent becomes complicated when sensors operate in places people enter for housing, work, or services.

The seventh question is public process. City officials should publish the permits governing each property and explain whether proposed changes remain within those approvals.

That explanation should come from planning and building staff, not solely from the developer. Independent documentation helps prevent a dispute over terminology from replacing substantive review.

The city should also identify applicable utility, air quality, fire, and environmental authorities. Different agencies can oversee separate parts of the same infrastructure.

The campus concept does not make those questions hostile or premature. It makes coordination more necessary because housing, offices, laboratories, and computing equipment share a small area.

STAK can strengthen its position by publishing a concise technical fact sheet. It should cover power, water, cooling, backup systems, occupancy, employment, construction, and project phases.

That document should clearly separate current conditions from projected capacity. It should also identify which figures have been independently reviewed.

Oakland can then evaluate claims using consistent standards. Residents would gain a better basis for comments than advertisements, rumors, or broad comparisons with unrelated hyperscale projects.

The developer has an incentive to provide that clarity. Uncertainty invites people to assume that the most consequential advertised capacity is imminent and fully utilized.

Detailed disclosure could show a smaller near-term impact. It could also establish enforceable limits before later expansion creates additional community concern.

Three Signals Will Show What Oakland Is Really Getting

Permits, measured resource demand, and tenant activity will reveal whether STAK becomes an integrated technology district or mainly a computing facility with adjacent amenities.

The first signal is the city’s permit record. Oakland should state which current approvals cover data center operations at 415 20th Street.

Any application for additional electrical, cooling, mechanical, or backup equipment would provide a more concrete picture of the project’s direction.

A major infrastructure application would strengthen the conclusion that computing is driving the next development phase. Limited modifications would support the developer’s claim of reusing established capacity.

The permit record should also explain how the four properties relate legally. A marketing campus does not necessarily constitute a single project under planning rules.

That distinction can affect environmental review, public notice, and the analysis of cumulative impacts. Officials should explain their reasoning before substantial expansion begins.

The second signal is measured resource use. Actual electricity demand offers more information than maximum nameplate capacity.

The city and developer should publish a staged load profile if legally possible. At minimum, they can report expected ranges without exposing individual customers.

Water consumption, cooling method, generator testing, and boundary noise should follow the same approach. These numbers allow comparisons with existing conditions.

Transparent measurement would also help separate STAK from much larger projects. A 20-megawatt urban facility should not be described as equivalent to every hyperscale campus.

However, it should not be treated as ordinary office space if dense computing becomes its dominant load. The operating data will determine which comparison is appropriate.

The third signal is tenant and program activity. STAK’s campus thesis depends on people using the surrounding offices, laboratories, residences, and community programs.

Watch for named robotics tenants, documented grants, recurring technical events, occupied desks, research partnerships, and operational testing environments.

These indicators would show that the campus produces activity beyond server hosting. Empty coworking floors and unidentified computing customers would weaken that case.

The relationship between employment and power is especially revealing. Oakland should examine permanent jobs created for each additional development phase and megawatt activated.

That ratio cannot capture every benefit. Research partnerships, new businesses, and downtown spending also matter.

Still, the comparison helps policymakers evaluate alternative uses of scarce electrical and commercial capacity. It keeps economic claims connected to measurable outcomes.

The next few months should therefore focus less on finding the perfect label. They should establish a verified baseline and clear expansion conditions.

Oakland has an opportunity to create a more useful model for urban AI development. Mixed-use computing projects do not fit neatly into regulations written for offices or industrial server campuses.

The answer is not to exempt them from scrutiny. It is to evaluate each function and its cumulative effect on the surrounding district.

For developers and enterprise AI buyers, STAK is also a test of whether local compute can coexist with housing, transit, and public-facing activity.

For knowledge workers, the project shows how AI infrastructure increasingly extends beyond software. Every model, robotics system, and automated workflow depends on physical power, cooling, networks, and space.

Teams evaluating AI services should ask where their workloads run and what operational promises support them. They should preserve vendor disclosures and technical findings in a searchable AI knowledge base.

Oakland residents should ask a similarly concrete set of questions: What operates today, what requires approval, and what limits will govern expansion?

If STAK publishes those answers, its campus distinction will become testable. If the details remain vague, the data center label will continue to dominate the debate.

The strongest outcome would combine genuine startup activity with transparent infrastructure controls. Oakland should require evidence for both sides of that promise before treating branding as policy.