How the same generators cleared two different reviews · CEC Docket 24-SPPE-01

Counted as 1. Permitted as 37.

One facility. Two rulebooks.

At the nearest home, the modeled cancer risk from this data center's 37 diesel backup generators already sits at 5.3 in a million, against a legal limit of 6 — and the one scenario the facility exists for, a real blackout with all 37 engines running at once, was never run through the pollution model. How did it clear review? It depends on who was asked. To the energy regulator, the generators are one 92-megawatt power plant — just under the threshold that avoids a stricter review. To the air-pollution regulator, they're 37 separate 3-megawatt engines, each under the emissions limit on its own. Same hardware, two framings, each tuned to what that regulator needs to say yes.

37

backup generators at the site

111 MW92 MW

the engines can produce 111 MW; the review counted the 92 MW the data center draws — just inside the 100 MW ceiling

134 hrs

how long each generator is assumed to run, per year

5.3 in a million

cancer risk at the nearest home — almost at the legal limit

01. The Split

The same hardware, counted two ways

Two agencies each ask a different question about the same 37 generators — and the project's answer changes shape depending on who is asking. As you scroll, watch what happens to the engines.

CECCalifornia Energy Commission01 / 03

“Is this a power plant we have to license?”

Their answer

1

one 92 MW power plant — all 37 added together

How the counting works

The Energy Commission asks one question: how much power can this whole site produce? So it adds all 37 generators together and treats the site as a single power plant.

Result

Counted that way, the project qualifies for a lighter, faster state review instead of the long, full licensing process — and can move forward with local approval alone.

BAAQMDBay Area Air Quality Management District02 / 03

“How much air pollution will it add?”

Their answer

37

thirty-seven separate engines — each its own small source

How the counting works

The Air District asks a different question: how much pollution does each engine add? Only one engine is allowed to run at a time for testing, and each engine's pollution is measured on its own, not added together — so no single source looks large.

Result

Measured that way, every engine stays under the pollution limits. Only the pollution from 34 hours of yearly testing has to be canceled out — and that's been reduced all the way to zero.

RealityWhat the facility actually exists for03 / 03

“What happens when the grid fails?”

The real answer

All 37

every generator running at the same time, at roughly 80% load

What the paperwork leaves out

Here's the catch: in a real, extended power outage, all 37 generators would run together — the moment they'd pollute the most. The paperwork describes that scenario only in words. It was never run through the actual pollution-modeling software, so that pollution was never counted against any limit.

The catch

The one moment this facility exists for — a real outage — is the one moment that was never modeled.

Is it hard to read?

Counted as one 92 MW plant

The mechanism in one sentence

The same 37-generator facility is counted two ways — as one big power plant for the state review that decides whether it needs a full license, and as 37 separate emergency engines for the air permit — and each version is exactly what lets the project clear that agency's bar.

02. 100 MW Line

Which capacity number counts

State regulators only get involved in a power plant of 50 megawatts or more — for scale, that's enough electricity for tens of thousands of homes. But there's a shortcut for plants between 50 and 100 MW: instead of a long, contested public licensing process, they can get a lighter, faster review — as long as regulators find the project won't seriously harm the environment, based on whatever study is put in front of them. This project used that shortcut, and the study behind its finding never modeled the worst-case scenario of all 37 generators running at once.

50 MW — state jurisdiction begins

92 MW — net load, the number counted

100 MW — exemption ceiling

111 MW — installed capacity (37 × 3)

The exemption only covers plants between 50 and 100 MW. These generators can produce 111 MW — over the limit. But the Commission only counted the 92 MW the data center actually draws from them, which fits just inside the range.

Three things had to be true

01

50 MW or more

Below 50 megawatts, the state doesn't get involved at all. To clear that bar, regulators added all 37 generators together and counted them as one single power plant, instead of 37 separate machines.

02

No more than 100 MW

Above 100 megawatts, the project would need the long, full state licensing process instead of the shortcut. To stay under that ceiling, regulators counted not the 111 MW the engines could actually produce, but the smaller 92 MW the data center is expected to draw from them.

03

No substantial adverse impact

Regulators also have to conclude the project won't seriously harm the environment. That conclusion came from a study — but the study pictures the generators running one at a time, for testing, not all 37 running together.

The exemption was granted on December 8, 2025.

This is the step where measuring the hardware differently changes the answer. Measured at what the engines could actually produce — 111 MW — the facility is too large for the shortcut and would need the long, full licensing process the shortcut is meant to avoid. Which number is the right one to use is a judgment call; what the record shows is that the shortcut only works because of which one was chosen.

03. Paper vs. Reality

What the filing assumes. What the audit found.

AVAIO's analysis assumes each engine runs about 134 hours a year — 34 hours of testing plus a 100-hour emergency placeholder. The Air District's own audit of comparable South Bay data centers, self-reported by the operators themselves, shows real operation running far higher.

On paper

“Each engine runs roughly 134 hours per year — 34 hours of testing plus a 100-hour emergency placeholder.”

— AVAIO air-quality analysis

What the Air District measured

In a 13-month audit of data centers in Santa Clara, San Jose, and Sunnyvale, nearly half the identified facilities ran their backup diesels for reasons other than routine testing and maintenance.

— BAAQMD audit, Sept 2019 – Sept 2020

On paper

“The air-quality model assumes a single engine running at a time; on that basis every pollutant clears federal and state limits.”

— Modeling basis, Ramboll report

What the Air District measured

One facility logged roughly 400 hours of non-testing operation across its generators in 13 months. At two facilities, 40 or more generators ran at the same time.

— BAAQMD audit, self-reported operator logs

On paper

“Emergency operation is treated as rare — the review leans on a 1.6%-per-year outage probability.”

— IS/MND energy analysis

What the Air District measured

The largest single driver of real generator use was the August 2020 statewide rolling blackouts — a recurring grid event, not a rare one.

— BAAQMD audit findings

On paper

“134 hours per engine per year is the assumed ceiling for the pollution math.”

— Potential-to-emit basis

What the Air District measured

A comparable Bay Area facility — the San Jose Data Center — is authorized to run its generators up to 459 hours per year per engine for PG&E demand response.

— BAAQMD Authority to Construct A/N 31953

Read carefully: the ~400-hour figure is the aggregate for one facility's generators over 13 months, and the San Jose units are natural-gas-fired, not diesel. The comparison is the order of magnitude — hundreds of hours of real non-testing operation versus the dozens AVAIO assumes. The regional air regulator itself flagged this gap and asked for a more realistic model; AVAIO's application didn't provide one.

04. The Numbers

No room to spare

These are the applicant's own figures, as filed — and every one clears its limit. But the health-risk number has a history: the first modeling exceeded the threshold at nearly every load level, and even the developer's revised figures still do.

Cancer risk at the nearest home

As filed in the IS/MND — single-engine testing only; all emergency operation excluded

5.3 in a million

District limit: 6 in a million

Smog-forming NOx, “potential to emit”

The legal worst-case per engine: 134 hours/year each

12 tons/yr

Offsets trigger at 10 (marked) · self-purchase ceiling at 35

NOx from testing & maintenance

The only slice that must be offset — offset to zero

3.0 tons/yr

Offset via the District's Small Facility Banking Account

Net operational NOx after offsets

As filed

0.23 tons/yr

Significance threshold: 10 tons/yr

Greenhouse gas from testing (CO₂e)

Testing operation only

2,862 MT/yr

Threshold: 10,000 MT/yr

How the cancer-risk number moved

The 5.3 figure above is the as-filed snapshot. The docket tells a longer story: the health-risk modeling started well above the threshold, and even the developer's own fix didn't bring it under. One note before the numbers: two different ceilings apply. The as-filed testing figure is judged against the District's 6-in-a-million limit, while the load-specific modeling below is judged against a 9-in-a-million threshold. Different scenarios, different limits — same story: at the line, or over it.

01 · As first modeled

17.0 in a million

at 25% engine load

The pre-mitigation figure: modeled cancer risk assuming testing could occur at any hour of the day or year, before any mitigation was applied. Initial modeling exceeded thresholds at every load level except 100%.

02 · After the daytime fix

10.45 in a million

at 25% load · 10.0 at 50% load

AVAIO then committed to daytime-only testing (7 AM–6 PM). Its own revised modeling still exceeds the 9-in-a-million threshold that applies to these load scenarios.

03 · The remaining fix

17 hr/yr cap

not independently modeled

A further proposed mitigation — capping operation at or below 50% load to 17 hours per year — has not been independently modeled or verified in the public record.

And outside the framework entirely

No health risk assessment has been conducted for emergency (blackout) operation — the scenario where all 37 engines run at once, and a recurring one in a region subject to PSPS power shutoffs. Air District Regulation 2-5 exempts emergency use of standby generators from health-risk-assessment requirements. This is a documented analytical gap in the record, not an unstated risk number.

Figures from the applicant's filings: Ramboll Air Quality, Public Health & GHG Technical Report (Tables 10, 11, 18) and the IS/MND Health Risk Assessment (Table 5.3-9). The cancer figure is the residential maximum and counts only single-engine testing; emergency operation is excluded. Pre-mitigation and revised figures are from the Responses to Data Requests AQ-8 (Table 1C) and AQ-9 (Table 1D).

05. Why It Matters

Counted as one unit, it likely would not have been approved

Both the state exemption and the air permit rest on the same assumption: one engine at a time, for testing. The docket's own numbers show how hard that framing has to work — and what likely fails if the 37 generators are judged as a single operating source.

  1. The numbers are already at the edge

    10.45 in a million

    Revised modeled cancer risk, against a 9-in-a-million threshold

    The entire approval turns on the 37-individual framing — and the docket's own numbers show how hard that framing has to work. Before mitigation, the modeled cancer risk reached 17 in a million at 25% engine load, exceeding thresholds at every load level except full power. After the developer restricted testing to daytime hours, its own revised modeling still showed 10.45 and 10.0 in a million. And every one of those numbers counts only testing — emergency operation is left out entirely.

  2. Both approvals rest on the one-engine picture

    0 model runs

    Of the real outage scenario — all 37 engines running together

    Both approvals this project needs can only be granted if regulators find the project won't seriously harm the environment — and that finding is based entirely on the picture of one engine running at a time for testing. The scenario where all 37 run together in a real outage is only described in words. It was never run through the actual pollution-modeling software, and the rules don't count emergency emissions against the limit anyway.

  3. Run it as one source, and the findings fail

    Judge the same engines as one integrated source operating together — the way they run in an outage, and the way the Air District found comparable South Bay facilities actually ran — and the record contains no analysis showing the health-risk and pollution limits could still be met. With the single-engine numbers already at the threshold, the project likely could not have made the findings required for approval had it been treated as one unit rather than 37.

What this is — and is not

This is not a claim that the project is improperly permitted. It is that the reassuring picture and the real operating picture rest on different assumptions — and only the reassuring one is modeled, against a limit it nearly touches.

The Commission has heard this before

This is not the first data center where the Air District flagged the gap. In the Great Oaks South proceeding, the District put its 13-month audit in front of the Energy Commission and specifically asked it to model emergency operation. An intervenor's testimony records that CEC staff “still refuses to evaluate emergency operations” — resting instead on the low statistical probability of any single outage. The AVAIO review takes the same path. The result is a project approved on assumptions the District's own measurements contradict — a recurring practice, not a one-off.

The window is open — and on the record

The Small Power Plant Exemption is already granted. But the Air District's Authority to Construct — the document that sets the binding, enforceable per-engine limits — has not yet issued. How the all-37 case is bounded is still an open question, and decisions made now go on the public record.

06. What to Do

The permit isn't final. You can still act

The air permit that will set the binding, enforceable limits on these generators has not been issued yet. Comments made now still shape it — by residents directly, and through two offices that can press the regulators. Nothing here requires reopening anything already decided.

Start here · Any community member

What you can do directly

  1. 1.Send a pre-written letter — contact details and ready-to-send letters for every office are on the take-action page.
  2. 2.Submit these findings as a formal public comment in the Air District's permit proceeding, and ask for a public comment period and community notice on the permit.
  3. 3.File a Public Records Act request with the Air District for the engineering evaluation and draft permit conditions once they're available.

Ask your State Senate district office

Request a closer look from the CEC

  1. 1.Write to the California Energy Commission asking it to re-examine the AVAIO Pittsburg Backup Generating Facility Small Power Plant Exemption (24-SPPE-01) and the adequacy of its Mitigated Negative Declaration, because the environmental finding relied on a single-engine, testing-only operating assumption that the Air District's own data on comparable South Bay data centers contradicts.
  2. 2.Ask the CEC to explain why it declined the Air District's specific, repeated recommendation to quantitatively model emergency and non-testing operation — including concurrent operation of multiple generators — and to apply that model to this project.
  3. 3.Request that, because the exemption is granted but the air permit is still pending, the CEC coordinate with BAAQMD so realistic operating scenarios and enforceable runtime limits are in place before construction is authorized.

Ask the Bay Area Air District

Strengthen the Authority to Construct before it issues

  1. 1.Require enforceable permit conditions that cap non-emergency, non-testing operation; define “emergency” narrowly so demand-response and grid-support runs count against the limit; and require real-time runtime metering with public reporting of actual generator hours.
  2. 2.Require a project-specific Health Risk Assessment that models realistic and concurrent operation — not single-engine testing alone — given the residential cancer risk already sits at 5.3 against the District's 6-in-a-million ceiling.
  3. 3.Open a public comment period and community notice on the permit, given the site sits in a community in the upper 90th percentile statewide for diesel particulate exposure.
  4. 4.Use the District's own 2020 South Bay audit (TN 235803) as the evidentiary basis for these conditions.

Ready to send something?

Pre-written letters and contact details for every office are on the take-action page.

Take action

07. Sources

Every figure, filed and public

CEC / Technical Filings

T1Ramboll Air Quality, Public Health & GHG Technical Report (SPPE Application, Appendix B), Tables 9–12, 18, 31–32 — CEC TN 254729. ↗
T2Per-engine permit commitments, Data Request responses AQ-2 / AQ-3 / AQ-4 — CEC TN 259481. ↗
T3Initial Study / Mitigated Negative Declaration: §5.3 Air Quality, §5.6 Energy, Appendix A Jurisdictional Analysis — CEC TN 266581. ↗
T4Executive Director's Recommendation on the SPPE application — CEC TN 267720, docket 24-SPPE-01. ↗
T5San Jose Data Center (Microsoft), Santa Clara County — BAAQMD Authority to Construct A/N 31953, authorizing up to 459 hr/yr per engine for PG&E demand response. Natural-gas-fired units.
T6BAAQMD audit of South Bay data-center backup-generator operation, Sept 2019 – Sept 2020 (one facility ~400 hr non-testing; 40+ generators concurrent at two sites) — Attachment 1 to BAAQMD's comment letter on Great Oaks South, CEC TN 235803. ↗
T7Intervenor (Sarvey) reply testimony on CEC declining to model emergency operation, Great Oaks South Data Center — CEC TN 239449. ↗
T8AVAIO Response to Data Request AQ-8, Table 1C — pre-mitigation health-risk modeling at 25% engine load, testing allowed at any hour — CEC docket 24-SPPE-01.
T9AVAIO Response to Data Request AQ-9, Table 1D — revised health-risk modeling after the daytime-only (7 AM–6 PM) testing restriction — CEC docket 24-SPPE-01.
T10BAAQMD Regulation 2-5 (New Source Review of Toxic Air Contaminants) — exempts emergency use of standby generators from health-risk-assessment requirements.

This is a plain-language summary of public regulatory filings, prepared for community reference. It is not legal advice. Emission figures are the applicant's estimates as filed; the binding air-permit conditions will be set by the Bay Area Air Quality Management District's Authority to Construct, which had not been published at the time this was compiled.