2026-10-14 SB-53 Assembly Floor analysis

Citation

California State Assembly, Assembly Floor Analysis. “SB 53 (Wiener), Senate Third Reading, As Amended September 5, 2025.” With Assembly Appropriations Committee fiscal comments. Consultant: John Bennett, Assembly Committee on Privacy and Consumer Protection. Dated Sept. 8, 2025. https://billtexts.s3.amazonaws.com/ca/ca-analysishttps-leginfo-legislature-ca-gov-faces-billAnalysisClient-xhtml-bill-id-202520260SB53-ca-analysis-392761.pdf. Full copy: raw/shared/2026-10-14 SB-53 Assembly Floor analysis.md (original: .pdf beside it), fetched 2026-10-14. Read 2026-10-15.

This page covers the full text. The CalCompute passages, word for word, are in the excerpt raw/legislative/SB-53 Assembly Floor analysis 2025-09-08.md. Its source page is SB-53 Assembly Floor analysis 2025-09-08.

Bottom line

• The analysis gives a cost figure for the consortium’s own work. The Government Operations Agency (GovOps) estimated “total costs of $2.5 million for expert contractors, infrastructure planning, and staffing” (p. 7).
• The Appropriations Committee expected “Possible cost pressures … of an unknown but potentially significant amount to the UC to operate CalCompute” (p. 7). Private donations may offset them “to some extent” (p. 8). This is the Legislature’s view of subdivisions (c) and (j).

Background

The Assembly floor analysis summarises SB-53 before the final Assembly vote. Most of it is about frontier AI developers: safety frameworks, transparency reports, incident reporting and whistleblowers. CalCompute is one of six major provisions. (pp. 1 to 2)

Key claims

• Major provision 5: “Upon appropriation, establishes in the Government Operations Agency a consortium required to develop a framework for the creation of a public cloud computing cluster to be known as ″CalCompute″ that advances the development and deployment of AI, as prescribed.” (p. 1)
• The bill “establishes a consortium within the Government Operations Agency to create a public computing cluster, known as CalCompute, to support AI research and safety testing.” (p. 3)
• The author says CalCompute is “a public AI research cluster that will provide startups and researchers with access to the resources needed to develop large-scale AI systems.” (p. 4)
• Supporters (Secure AI Project and others) say CalCompute “would make those resources more accessible to California′s top universities and startups” and “mirrors a similar computing cluster that is already being established in New York state.” (p. 5)
• Fiscal comment 2: costs to GovOps “until January 1, 2027, possibly in the high hundreds of thousands of dollars to low millions of dollars. GovOps estimates total costs of $2.5 million for expert contractors, infrastructure planning, and staffing”. GovOps “was not able to provide a breakdown” and expected “contract workers due to the short timeframe”. Member expenses “may be in the thousands to low tens of thousands of dollars”. (p. 7)
• Fiscal comment 3: “Possible cost pressures (General Fund) of an unknown but potentially significant amount to the UC to operate CalCompute, should it be established within the UC. State costs may be offset to some extent by private donations, which the bill authorizes the UC to receive to implement CalCompute.” (pp. 7 to 8)
• Votes: Senate Floor 37-0-3; Assembly Judiciary 12-0-0; Assembly Privacy and Consumer Protection 10-0-5; Assembly Appropriations 11-1-3. (p. 8)
• The opposition (Chamber of Progress) does not mention CalCompute. (pp. 6 to 7)

Evidence and method

A committee staff analysis. The fiscal comments rely on department estimates. GovOps gave no breakdown of its $2.5 million figure.

Limits

• Written for the September 2025 bill.
• The analysis is inconsistent on two names. It says “Office of Emergency Management” in the summary and “Office of Emergency Services” in major provision 3. It says incidents go to the Attorney General in one place and to the Office of Emergency Services in another. (pp. 1, 3)
• The copy prints the compute threshold as “1026” (p. 3). The supporters’ letter gives “10^26” (p. 5). The PDF lost the superscript.

⚠️ Contradiction: the analysis frames the consortium as created “Upon appropriation” inside GovOps and run by GovOps contractors (pp. 1, 7).