Project summary

After the passing of California Senate Bill 53, which spurred the framework for the creation of CalCompute (California Government Code § 11546.8), CalCompute formed as a non-profit organization aimed at advocating for and participating directly in the creation of public compute infrastructure for California. This public infrastructure project gives researchers, the public sector, and startup builders access to an alternative to private compute. CalCompute is currently performing policy analysis and governance research as part of its framework report. Our report is due to the California legislature and is the project behind our proposal. CalCompute is not a government agency and is a coalition dedicated to building and researching safe and accountable public compute infrastructure. 

As AI permeates everyday facets of life, large AI companies are increasingly the sole proprietor of the hardware and the models that run through the large amounts of increasingly sensitive private data. This data is owned by the public and, therefore, its usage ought to be directly accountable to the public. Without a public alternative, the ownership and decisions behind AI will largely stratify towards an increasingly homogenous body of ownership. Additionally, in cases where agency data is expensive to handle, such as HIPAA/FERPA cases, publicly governed compute would let small institutions work in an environment whose terms are stipulated and stabilized by public need rather than costly vendor contracts. 

CalCompute exists to fill a genuine need for greater self-ownership of data; at a time when companies are scrambling to compete for the data of Californians, CalCompute can provide an alternative whose charter includes data sovereignty as a core principle of its operations. Our first movement into this domain, and the emphasis of this project, is to write the comprehensive report that Senate Bill 53 calls for, and deliver it to the California Government Operations Agency. 

What are this project's goals? How will you achieve them?

Current leaders in private AI posit its importance and magnitude in reshaping society. However, they have been inactive in ensuring safe and equitable access to such a purported important resource. Californians have long been upset with the way their data is being used, and this mismatch of power will only worsen in the absence of a public alternative to reinstate rightful ownership to Californians. 

We are seeking funding for the writing and the research tools to deliver our framework report, as called for by Senate Bill 53, which outlines the creation of CalCompute.

This report requires research into understanding what the costs of compute would look like as a public resource, the impact on California, current figures on private compute reliance, and what the lack of a public alternative would mean for California. Our project’s success is measured by the quality of research we produce, the promise of delivering the report before its January 1st, 2027 deadline, and generating momentum on the state legislature’s part to fund and build this infrastructure project. We strongly believe that the creation of a California beholden compute infrastructure will be invaluable in serving our public sector and as an alternative to the current state of compute. 

Our report will focus on measurable research on core topics that we identified as relevant. Please refer to these phases as a guideline for the process of our report. We have set internal deadlines for the completion of each phase to ensure we stay on schedule to deliver our report.

Phase 1: Landscape and Cost & Funding Analysis

The first phase of our report will cover the current landscape of California compute infrastructure. What this means is we will cover the current usage. We need to map UC clusters, national laboratories, and the commercial hyperscalers already provided, and where the real gaps are.

• What current infrastructure does California have?

This will evidently segue into an analysis of the cost of building this infrastructure, cost to maintain, and funding sources. 

• Where are our gaps?

We will identify the current gaps in service. For instance, the UC system compute acts as a way to connect researchers to high performance computing power. However, researchers outside of it are inherently at a disadvantage, and may need to lean on private compute. 

An analysis into our own gaps will lead to a discussion of the gaps that exist in private compute. For example, the recent AWS outages exposed the danger of such a homogenous model of hosting that some people's work days came to a halt, this loss of time represents a significant destruction of resources.

• What funding models sustain it?

Here we will answer: once we are allocated resources, how can we continue funding this project into maintenance, what would that stage of our project look like.

• Where are our gaps?

• Where is the demand for compute going unmet?

• What does it cost to build and maintain this infrastructure?

• Rent versus own: where does the break-even fall?

Phase 2: Governance and Use Parameters

The following phase will cover the governance structure and operation organization of CalCompute. The organizational structure of CalCompute . The distribution of and allocation of resources is inherently limited by the amount of compute we have access to. Determining equitable access to safe and ethical AI. Determining an equitable allocation of compute resources will require research and meeting with stakeholders.

• Who may use the infrastructure, and for what?

We will need to rigorously answer the question “Can our system be abused in a malicious way” and ensure that our policies do not shift the ability to control critical compute resources into the hands of a small group. This will require preliminary research into open-weight models, and exploring the ways they can be misaligned and how we as a platform can ensure they aren’t turned on the public using our own platform. 

• What structure fits, and how does it relate to the University of California?

We’ll explore which organizational structure best fits CalCompute's mission. What will we look like as an entity and what will be our relationship to the UC system, as well as how this relationship may impact oversight access and the platform in the long term. 

• How should compute be allocated, and who decides?

• What structure fits, and how does it relate to the University of California?

• What keeps a public option accountable to the public?

• Who may use the infrastructure, and for what?

• What legal-compliance constraints govern sensitive data?

• What safeguards apply to workloads?

Phase 3: Workforce, Partnerships, and Public-Sector Workforce

Discovering equitable pathways to strengthen the workforce to familiarize them with the tools to run this infrastructure. This acceleration and growth of our workforce will require partnerships to help us build expertise. We must identify these organizations and determine the costs. These partnerships will be incredibly important in driving the future of California workforce development. 

• What open-source tools and vendors do we build on?

What would our platform look like, who would we need to rely on and what expertise would we need to build and communicate among our users. We would need to heavily explore this in our report. 

• What expertise is needed to run and use the infrastructure?

• How do we bring on hardware and distributed-compute expertise?

• How do we train the next generation of public-compute researchers?

• Who are the academic, public, and private partners?

• How do we structure partnerships without recreating private lock-in?

• What compute capacity do public agencies actually need?

• How do we equip the public sector to use it?

• What internal skills must the state build to remain independent?

We have set guidelines to meet each of our phases in a timely manner. Our working groups meet daily, our full compute group meets weekly, and progress on each phase is reviewed at those meetings. We plan on keeping our stakeholders and the public informed along each step of the process. CalCompute will publish a progress report for each phase, enumerating our current progress, our future goals, and what we have accomplished. Phase progress is reviewed at our coalition meetings. 

This communication will come primarily in two forms:

1. Our website, the repository of everything CalCompute.

   • Each phase has its own progress report, covering the relevant timekeeping information for that phase.

   • Our deadlines are posted publicly not just for our own tracking, but as a statement of our responsibility to deliver.

2. Direct public updates, issued quarterly.

   • A newsletter summarizing the quarter's milestones, delays, and next steps.

   • Distribution through our social channels so updates reach people who aren't checking the site.

Milestone

Target Date

Phase 1 Landscape and cost & funding analysis complete - September 2026

Phase 2  Governance and use parameters complete - October 2026

Phase 3 Workforce, partnerships, and public-sector workforce recommendations complete -December 2026

Final report delivered to the Legislature - January 1, 2027

How will this funding be used?

LLMs

We plan on using this funding to accelerate our report writing and research pace. With these LLMs we can quickly digest information and better communicate our needs. Without these tools, we stand at a disadvantage and cannot move at the same pace with which this industry calls for. We are organizing to expand access to these tools and we will be using them.

Each of our phases will require significant research, writing, and in some cases the development of bespoke tools. The creation, upkeep, and maintenance of each phase will be supported and accelerated using LLMs. 

Our LLM Use Cases:

• Supporting our Writing and Research

  • Bouncing ideas to clarify and/or restructure prose

  • Stress-testing our claims

  • Ensuring consistent ethos across documents

  • Researching the current landscape

• Content Summarization

  • Summarization of documents

  • Content finding (with human verification, always)

Funding Ask: 3.7k USD for LLMs, find our breakdown.

Compute

We are requesting funding to rent computational resources to support both external and internal researchers who lack access to it. This is ultimately the goal of CalCompute, we want to fill the need gap and accelerate our community by allocating access to a public infrastructure platform, as they need it. 

An interesting question we will need to answer in our report is what will public compute infrastructure look like. What use cases will people have for the platform and what tools will they share, use, maintain, and develop. Currently closed-weight models dominate the field, however open-weight models aren’t far behind. As stewards of a platform granting access to run these models, we need to create a discussion around open-weight model security as it points to the underlying inherent capabilities of these models to resist attacks. Having access to greater computational power would lend itself to reporting on and researching LLMs in the context of the public.  

Our Compute Use Cases:

• Evaluation

  • Benchmarking LLMs

  • New techniques and our use cases for them

• Replication

  • Reproducing published findings and analyzing how we can employ them

  • Realistic cost baseline through findings from papers

• Adaption

  • Tuning inference models for research

  • Adversarial testing open-weight model adaptation to resist misuse

  • AI Safety

    • What use cases exist from misalignment

    • Ensuring we mitigate these tools from being used

Funding Ask: 10k USD for Compute, find our breakdown.

We acknowledge that as we continue operating, these numbers may change to allocate costs towards different aspects of compute related research.  

Who is on your team? What's your track record on similar projects?

Find our team over on: californiacompute.org/people/, we are a diverse set of professionals coming from different sets of expertise and backgrounds. From academic to private sector experts, we are confident that our report will produce significant change for California. CalCompute’s current goal is to produce a report that defines the expectations of this project, its feasibility, and its impact. CalCompute is uniquely positioned in that we have both the expertise and the mandate to change a largely entrenched field. This report is the first of its kind for us and the state, we are confident in our ability to deliver. 

What are the most likely causes and outcomes if this project fails?

Without a public alternative for the infrastructure that currently serves Californians, our private counterparts will be largely in charge of making the important decisions around our data, without looping in the people it represents. CalCompute being unable to fulfill its role would mean a significant blow towards creating public compute infrastructure that serves the needs of California and establishes sovereignty over our data.

Current overreliance on private sector infrastructure has shown its cracks and merely hoping for the benevolence of non-public aligned private stakeholders will evidently leave the public burdening the costs. In principle, regardless of outcome, private companies should not have the de facto power to unilaterally make decisions for the public. We have seen the impact of the historical outcomes of oligopolies, though rarely so when the market is positioned to be so inextricably linked not just to daily life but in the manner in which we move and operate our most important modern commodity: data. 

CalCompute has an opportunity to create a blueprint for other states to follow and create a conversation around how critical ownership of said data is to our modern climate. Just as California is a global leader in producing some of the best in the domain of computation, CalCompute will produce the framework for not just other states to base their future on, but we will also act as a model for global communities. Our partnership with UN members demonstrates that. 

Without CalCompute’s report, the legislature is brutally uninformed and less likely to take action during this crucial period. Most importantly, the American people will be without a representative fighting for them. 

For additional information and insight into our organization, please visit us at californiacompute.org