Summary

  • A Wall Street Journal report surfaced at 22:00:36 UTC on 8 August and focused attention on the approval path behind AWS’s now-visible Gilroy construction.
  • City records identify AWS as the applicant for two roughly 218,000-square-foot data-centre buildings and a security building on a 56-acre site, approximately 438,500 square feet in total.
  • The Community Development Director approved the project on 3 July 2025 under existing M2 industrial zoning; no public hearing was required, and a filed appeal was later withdrawn.
  • The project was not absent from public process: a 45-day draft environmental-review period ran in 2024, drew six comment letters and produced a final response record.
  • Phase I requires a 49MW PG&E connection and lists 25 emergency generators rated at 2.5MW plus one 600kW unit; those figures are not current load or normal generation.
  • The city lists the project as under construction and says Phase II industrial water use must shift to Amazon-funded recycled-water infrastructure before that phase can operate.

Construction made an old approval newly visible

The current report does not create a new permit. Gilroy’s Community Development Director approved AS 20-23 in July 2025; grading and underground-utility permits followed in December, and a building permit in March 2026. The city now labels the project under construction.

What changed is public salience. A planning file can remain abstract until earth moves, traffic changes or large structures rise. The 8 August investigation belongs in the current news window because it documents that visibility gap, while the underlying legal milestones retain their original dates.

This distinction also avoids an easy exaggeration. AWS did not receive approval this week, and the project was not entirely hidden. The material issue is how a campus of contemporary infrastructural scale passed through rules that treated it as an allowed industrial building with staff-level site review.

The process was public but did not require a hearing

Gilroy released the draft environmental impact report for 45 days between August and September 2024. The city says it received six comment letters and answered them in the final report. Those are meaningful public records and should not be erased by the language of secrecy.

The decision point was nevertheless narrow. The site already carried General Industrial planning designation and M2 zoning. AWS sought no rezoning, general-plan amendment or subdivision. Under the city code, the required Architectural and Site Review permit could therefore be granted by the Community Development Director without a public hearing.

An appeal was filed on 21 July 2025 and withdrawn on 18 August. Once withdrawn, the project did not go to the Planning Commission or City Council. That is not evidence of an elected vote for or against AWS. It is evidence that the available escalation route closed before an elected-body hearing occurred.

The physical project is larger than the procedural label

The approved plan covers two data-centre buildings of roughly 218,000 square feet each, plus a 2,500-square-foot security building. The total is approximately 438,500 square feet on a 56-acre previously undeveloped parcel. The site also needs a substation or switchyard, off-site PG&E work and other utility connections.

“Architectural and site review” sounds like a question of façades, setbacks and drainage. For a data centre, it also fixes the geography of a long-lived electricity and backup-power system. The consequence of legacy zoning is therefore not merely administrative speed. It is that decisions about grid capacity, generators, water and land conversion can be embedded in a permit category that attracts less political attention than rezoning.

The governance test is proportionality. Rules should not become discretionary solely because a project is controversial; predictability matters to investment. But the trigger for public scrutiny can be tied to measurable infrastructure—load, backup generation, water, land and network effects—rather than to whether a developer happens to request a zoning change.

Forty-nine megawatts is a connection requirement, not an operating result

Gilroy’s project page says Phase I requires a 49MW PG&E connection. That figure sets an infrastructure scale. It does not show that the connection is energised, that AWS is drawing 49MW continuously or that all of it reaches servers.

The phase also lists 25 emergency generators rated at 2.5MW and another 600kW generator for essential functions. Adding those nameplates and treating them as routine plant output would be wrong. Emergency systems serve resilience and testing; their annual operating hours, permit limits, fuel use and emissions require separate evidence.

Phase II is meant to use a different backup architecture: a future battery or fuel-cell system and rack-level lithium-ion modules. “Battery or fuel cell” is not a selected design. Vendor, energy duration, fuel, capacity, safety case and commissioning all remain open. The technology transition should be judged when it becomes an approved, procured and tested system.

Water illustrates how project numbers evolve

The city’s current estimate puts Phase I potable-water use at about 5,480 gallons a day, compared with roughly 7.2m gallons for the city. It says full buildout would shift industrial needs, including cooling, to recycled water and leave about 1,728 gallons a day of potable use for domestic purposes.

These are revised forecasts, not measured operating consumption. Earlier environmental materials used higher estimates, and the newer number reflects claimed technology and efficiency changes. That is exactly why a live project needs versioned assumptions: design evolution can improve performance, but the result should later be compared with meters.

The Phase II condition is more important than the adjective “efficient”. AWS must use recycled water for industrial needs before the second phase operates, and the city says Amazon will fund the necessary infrastructure. The next proof is a completed agreement, design, construction schedule and verified capacity—not merely an intention to build it.

Two impacts remain after mitigation

Gilroy’s environmental record identifies two significant and unavoidable impacts: conversion of Prime Farmland and Farmland of Statewide Importance, and vehicle miles travelled above the applicable threshold. The project’s approval therefore did not mean every impact had been eliminated.

This matters for the public narrative. Revised potable-water estimates and backup-power changes can improve particular dimensions while leaving land-use and transport effects. An accurate assessment must preserve both benefits and residual costs rather than allowing one updated metric to stand for the whole project.

The same principle applies to employment and tax benefits. The city expects development, jobs and property-tax gains, but the packet does not provide realised payroll, occupation mix or tax receipts. Those outcomes should enter the record when they can be observed.

The design problem is notice at infrastructure scale

Gilroy followed the approval mechanism in its code. AWS used the path available to an industrial applicant. The resulting controversy is therefore less about proving a concealed vote than about whether the trigger for deliberation still fits the object being permitted.

A modern data centre combines real estate, high-voltage supply, backup generation, cooling, communications and long-duration land use. If review turns only on zoning change, a very large project on already industrial land can receive less visible deliberation than a much smaller project requesting a map amendment.

The durable response is prospective and measurable: notice radii, public information sessions or hearings keyed to electrical load, generator capacity, acreage or water demand. Changing the rule for future projects is cleaner than pretending that a hearing occurred—or that one was legally required—for this one.

Operations will supply the accountability record

The approval chain is now substantially complete enough for construction. The next evidence belongs to engineering and performance: PG&E upgrades, the 49MW connection, generator permits and testing, the recycled-water system and the Phase II backup decision.

Once the campus operates, the useful denominators will be peak and average load, potable and recycled water, generator runtime, measured emissions, jobs and tax receipts. Those figures can test both the city’s assurances and the concerns that formal review did not settle.

Gilroy’s case will travel because many municipalities have legacy industrial districts. Its policy value lies in showing that a searchable file is not the same thing as an obvious civic decision. Public process has to be legally valid and practically discoverable at the moment meaningful choices can still be made.

Sources