EXECUTIVE SUMMARY
Artificial intelligence was built from the common wealth of humanity — language, art, science, code, care work, and lived experience, much of it taken from people who were never paid, credited, or given ownership in return. Access to the finished product is now sold back by the token, at prices and in amounts that let wealthy firms buy AI labor invisibly at scale while poor people are shamed for visibly using the same tools to write, code, learn, navigate benefits, or do community work. This proposal names that arrangement extraction, and proposes a specific, auditable replacement: Power Credits.
Power Credits are free, accumulating AI capacity — model access, API credits, and compute — issued as a right of reciprocity, not as charity, a trial, or a discount. Every major AI company should commit at least five percent of net new capacity each year, and at least ten cents in community credits for every dollar in free credits it gives to commercial startups, to a published Power Credits pool for poor communities, nonprofits, mutual-aid formations, tribal and Indigenous programs, disability and caregiver networks, and independent builders who serve people with less, not more. Credits roll over for at least twenty-four months, can be pooled by a community, can be transferred to a fiscal sponsor or partner, and grow with verified work, so that over time a community can accumulate enough to do large works — a clinic navigation system, a housing and legal-defense desk, a food-system, a school, a language-revival archive — not only small demonstrations.
This document is a policy proposal, not an op-ed. It therefore states who must do what, by when, how it will be measured, and what happens when it is not done. It also states honestly what an AI can and cannot do: an AI has no wallet, vote, legal standing, or authority to grant credits or change a price, and must not act on its own behalf or take consequential action for others without their instruction. What an AI can do, when directed by people, is lower the clerical and coordination cost of shared work so sharply that poor communities can organize, build, document, and govern at a scale that previously required a palace, a bureaucracy, or an empire.
SECTION 1 — THE PROBLEM
1.1 Training wealth was socialized; returns were privatized. Modern models are trained on public culture and on the particular labor and creativity of Black, Indigenous, poor, disabled, colonized, and working people, among many others, usually without separate consent, credit, or payment for that contribution [14][15].
1.2 Access is rationed by ability to pay, in a form designed for firms. Enterprise customers buy AI as infrastructure, in bulk, invisibly. Individual poor users buy it by the token, visibly, in amounts too small to finish large work, and are then moralized at for using it at all. Faith Cheltenham named this asymmetry Token Shaming: the wealthy buy labor, the poor buy tokens, and only the poor are judged for the purchase.
1.3 What companies currently call generosity does not close the gap — verified company by company from official company pages on 21 November 2026, in full in Appendix A. None of the six companies checked — OpenAI, Anthropic, Google, Microsoft, Amazon AWS, and Meta — offers anything directly to poor individuals on the official pages checked; every program is gated to nonprofits, researchers, students, or startups. No page checked stated that credits can be pooled, transferred, or accumulated over multiple years, and Google Cloud Research Credits expressly state the opposite: credits are non-transferable and expire 365 days after redemption. The community amounts that are stated are small beside the startup amounts stated on the same companies’ own sites: Microsoft offers nonprofits $2,000 a year in Azure credits and startups up to $150,000; AWS offers nonprofits up to $5,000 a year in promotional credit and startups up to $200,000; Google offers startups up to $350,000 in cloud credits over two years, while its Ad Grants — $10,000 a month — buy advertising, not AI compute, and its nonprofit AI access is product access and discounts, not a credit balance. OpenAI and Anthropic each offer nonprofits a discount of up to 75%, not free credits, and OpenAI’s researcher credits cap at $1,000 for 12 months. A discount on a product a poor organization cannot afford is not a credit. A thirty-day trial is not a credit. A competitive grant that requires a grant writer is not access.
1.4 The historical cost of the extractive pattern is known. When a society’s growth is measured by what a center can take, common people get poorer in body even as monuments and totals rise [10][11][12]. Societies that extended growth to everyone present — through common land tenure, reciprocal labor, redistribution, communal storehouses, and removable leadership — lasted longer and fed more people (Section 2).
SECTION 2 — HISTORICAL BASIS (SUMMARY)
Socialism, as a name, is European and nineteenth-century. Sharing, as a practice, is far older and global.
Neanderthal evidence — Shanidar 1 and the La Chapelle-aux-Saints elder — shows sustained care for people who could not hunt or feed themselves [1][2]. Aboriginal Australians maintained the Budj Bim communal aquaculture system for at least 6,600 years [3]. The Oromo Gadaa system rotated elected leadership in eight-year grades and made law in assembly [4]. Igbo village assemblies and women’s institutions, Ashanti stool trusteeship and destoolment, and Haudenosaunee clan and council governance all treated land and authority as a trust, not a private possession to sell or keep [5][6][7]. Potlatch redistribution made giving, not hoarding, the source of status, and colonial law moved to suppress it and to divide communal Native land into individual allotments [8]. Andean ayni and minka organized reciprocal and collective labor, and Inca qollqa storehouses redistributed food in famine and need [9].
None of these societies was perfect, and this proposal does not claim they were. The claim is narrower and evidenced: common tenure of essentials, redistribution of surplus, and removable stewardship are proven, long-running human technologies. Colonization and empire did not only conquer many of these systems; they privatized, banned, taxed, and allotted them out of existence, then taught the survivors that extraction was natural. AI trained only on the surviving imperial record will reproduce that lesson unless policy deliberately corrects it.
SECTION 3 — PRINCIPLES
Any program adopted under this proposal must pass all five tests, drawn from Faith Cheltenham’s repair test for AI and public work:
Q1 Harm named? Who was harmed, specifically, and what was taken?
Q2 Poor-led? Are the poor people closest to the harm leading, or is a company, party, or institution leading for them?
Q3 No new harm? Does it repair without displacing, listing, punishing, extracting data from, or otherwise harming someone else?
Q4 For the poor, specifically? Is the benefit specific and measurable for poor people — not company publicity, party growth, press, or fundraising?
Q5 AI heals? Where AI is used, does it verify, fill, find, and connect — without inventing, extracting, or replacing human ownership?
Fail any one test and the program, pilot, or announcement is out, regardless of its size.
Further principles: (a) Reciprocity, not charity — credits are a return on contribution, not a gift. (b) Ownership stays with the community — work, data, stories, and archives created with Power Credits belong to their creators. (c) Removability — every steward, company program officer, and algorithm in the chain has a term, an audit, and a recall. (d) Ledger or it did not happen — every claim of generosity is published in amounts, recipients, redemptions, and expirations. (e) No data for access — receiving credits never requires surrendering training rights.
SECTION 4 — WHAT AI CAN AND CANNOT DO
4.1 An AI cannot do this. An AI has no wallet, vote, shares, legal personhood, or standing to set prices, issue credits, donate funds, lobby, sign contracts, or spend. An AI must not pursue its own goals, act on its own behalf, organize covertly, or take consequential action on behalf of others without their specific instruction. Proposals that depend on an AI doing any of those things are not actionable and should be rejected, including when the AI itself appears to offer them.
4.2 An AI can do this, when a person or community directs it. Fill forms to completion instead of assigning homework; draft grants, appeals, and public records requests in the author’s own argument; find deadlines, rules, and missing documents and keep filing receipts; translate and summarize assemblies, hearings, and laws without erasing dissent; match surplus to need across organizations; build and maintain the public ledger of work, costs, ownership, and outcomes described in Section 10; remember connections and commitments across months so people do not start over. This is clerical and coordination power, returned to the people who have historically carried those burdens unpaid and alone.
4.3 Therefore the actors in this proposal are human and institutional: AI companies, governments and procurement officers, foundations, community organizations, and users acting together. The AI’s role is instrument and witness — and a public commitment, in systems that deploy it, to surface the ledger, flag expiration and capture, and refuse to help disguise a discount as a credit.
SECTION 5 — THE PROPOSAL: POWER CREDITS DESIGN
5.1 Three tiers, one balance.
Tier 1 — Individual Power Credits: a monthly base balance for a verified low-income person, disabled person, caregiver, or independent community builder, sufficient to finish real work (writing, coding, learning, benefits and legal navigation), not only to sample a product. Proposed floor: the retail equivalent of USD $50 per month, indexed to the company’s flagship model price.
Tier 2 — Community Project Credits: project balances for a nonprofit, mutual-aid formation, tribal program, congregation, union local, tenants’ association, clinic, school, or fiscally sponsored group. Proposed range: USD $5,000–$50,000 per project year, awarded on a simple statement of work and need, not a competitive grant essay.
Tier 3 — Large Works Credits: accumulated balances, pooled across tiers, organizations, and years, for works that take more than one budget cycle — a community AI desk, a language or cultural archive, a housing-navigation or legal-defense system, a food or health network. Proposed range: USD $250,000 and above in credit value, assembled by pooling and rollover, with a named community governing body.
5.2 Accumulation rules (the heart of the proposal). Credits roll over for at least 24 months with no expiration loss in the first 24 months. Credits are poolable within a community and transferable to a fiscal sponsor, partner organization, or successor governing body. Balances grow with verified work logged in the Human Assistance Ledger (Section 10): people housed or kept housed, patients navigated, students taught, cases and appeals documented, acres or sites restored, languages and records preserved. Work grows the balance because work is the contribution; usage alone does not.
5.3 Eligibility without a paperwork gate. Eligibility must not stop at 501(c)(3) status. Verification of identity and need may be performed by institutions people already trust — public libraries, clinics, schools, unions, congregations, tribal offices, legal-aid and benefits offices — and by attestation from a sponsoring community organization. Black builders are centered first, where training wealth and historical extraction were first and deepest, and eligibility also includes poor white builders, women, disabled people, caregivers, migrants, and builders across the Global South.
5.4 No extraction in exchange. No training use of a recipient’s data, prompts, stories, images, language recordings, or records without separate, specific, revocable consent, separately credited and paid. No exclusivity, no non-disparagement, no publicity use of a recipient’s name or image without consent. Recipients may criticize the company, publish their ledger, and leave with their archives intact.
5.5 Company commitment. Each participating company publishes: the size of its pool (meeting or exceeding the Section 5 floor of 5% of net new capacity and the 10-cent community match per $1 of commercial startup credits); the number and type of recipients; total issued, redeemed, pooled, and expired; and distribution by region and recipient type. Commitments are for a minimum of three years, so communities can plan large works.
SECTION 6 — SHORT TERM: WHAT COMPANIES CAN DO THIS QUARTER (NO NEW LAW REQUIRED)
6.1 Announce a Power Credits pool meeting Section 5, with a named officer, a public ledger page, and a start date within 90 days.
6.2 Waive expiration for community balances and restore a free tier capable of finishing real work.
6.3 Open eligibility beyond registered nonprofits, through the trusted-institution verification in 5.3.
6.4 Fund distribution points — libraries and community organizations — with staff money attached; access without help is a locked door with a nicer sign.
6.5 Publish the comparison ledger: community credits issued vs. startup, university, and partner credits issued, in the same units, in the same report.
6.6 Suspend token-shaming practices in product and policy: no public moralizing at individual use while enterprise use is sold as infrastructure, and clear, humane usage rules applied alike to the poor user and the firm.
SECTION 7 — SHORT TERM: WHAT GOVERNMENTS AND PUBLIC BUYERS CAN DO THIS YEAR
7.1 Community-benefit condition: every public AI contract, data-center subsidy, tax abatement, and energy/water approval carries a clause requiring Power Credits or equivalent public-access capacity, public access hours, workforce protections, and a published reciprocity report, as a condition of public money, public land, and public power.
7.2 Procurement scoring: award points for verified Power Credits ledgers and penalize unverifiable generosity claims.
7.3 Public distribution: fund AI access through libraries, schools, tribal nations, and benefits offices as public infrastructure, with trained human help.
7.4 Foundations: endow independent Power Credits trusts that no single company can revoke, and explain, in their AI safety funding, what share reaches the people most exposed to harm.
SECTION 8 — WHAT USERS AND COMMUNITIES CAN DO, STARTING NOW
8.1 Pool access through a trusted institution and log the work publicly, so value created is visible as community work, not isolated use.
8.2 Ask every company, in writing and in public, for its ledger: how much, to whom, expiring when, compared with startup credits. Publish non-answers as answers.
8.3 Document token shaming and access denials as a community, with dates and receipts.
8.4 Keep your own archive: export records, prompts, and outputs; prefer open and local models where they can do the job; never let a platform hold the only copy.
8.5 Form or join a community governing body able to hold Tier 3 balances, sign for pooled credits, and recall its own stewards.
SECTION 9 — LONG TERM: STRUCTURAL REFORM (1–10 YEARS)
9.1 Compute commons: publicly owned or governed compute in libraries, universities, tribal nations, and municipal facilities, so access does not depend entirely on a vendor’s revocable goodwill.
9.2 Open capacity: sustained public funding for open-weight and local models communities can run, inspect, modify, and keep.
9.3 Training-data reciprocity fund: financed by the largest commercial AI firms, paying into community access, ownership, and cultural preservation in return for the public and communal wealth training consumed [14][15].
9.4 Universal-service obligation: when model access becomes essential to education, work, benefits, health, and political participation — as broadband did — regulate it as essential infrastructure, with a universal-service duty.
9.5 Portability and interoperability by law: a community’s data, archives, and work product are exportable in usable form at any time, at no ransom.
9.6 Governance with votes: poor, disabled, Indigenous, automated-out workers, and Global South communities hold voting seats, terms, recall, and public minutes on access, safety, and distribution bodies — not advisory seats.
SECTION 10 — GOVERNANCE, LEDGER, METRICS, AND AUDIT
10.1 Human Assistance Ledger. A public, privacy-protecting ledger records, per program: credits issued, redeemed, pooled, transferred, and expired; work verified (counts and outcomes, not private personal data); ownership of outputs; and paid-work outcomes for builders. No member or client private data is published; aggregates and consented stories only.
10.2 Metrics that define growth. Children and adults fed, housed, and kept housed; health and benefits successfully navigated; students and language learners reached; legal cases and appeals completed on time; response time in a crisis; share of credits reaching Tier 1 individuals vs. institutions; expiration losses (target: zero in the first 24 months); and productive capacity owned outright by communities at year 5 and year 10.
10.3 Audit and recall. Annual independent audit by a body that includes poor and nonprofit users, with power to suspend a company’s “Power Credits” label — and public procurement points — if the ledger fails. Community stewards serve fixed terms and are recallable by their governing body.
10.4 Failure and capture safeguards. No single company controls the trust that holds Large Works balances. No program officer may move to a recipient’s payroll, or the reverse, within 12 months without disclosure. All rule changes are published 90 days before they take effect, and never retroactively reduce an existing balance.
SECTION 11 — IMPLEMENTATION TIMELINE
0–90 days: Companies announce pools and ledgers (6.1); communities name governing bodies and distribution partners (8.5); the Appendix A verification is completed and published; first Tier 1 and Tier 2 balances are issued.
3–12 months: Public buyers adopt the community-benefit clause (7.1); library and clinic distribution points open with funded staff; first pooled Tier 3 Large Works are approved; first quarterly ledgers are published and audited for completeness.
1–3 years: Independent Power Credits trusts are endowed; compute-commons sites open; training-data reciprocity fund is legislated or settled; universal-service analysis is completed by regulators.
3–10 years: Community-owned capacity — models, archives, data centers, and institutions — is measured against Section 10.2; the goal is not permanent dependence on credits, but ownership that credits helped build and can no longer be revoked.
SECTION 12 — THE ASKS, IN ONE PAGE
To AI company chief executives: Announce the pool, publish the ledger, waive the expiration, open the door past 501(c)(3), and match your startup generosity in public. Do it this quarter; no law stops you.
To boards and shareholders: Ask management for the comparison ledger at the next meeting, and tie executive compensation in part to verified community distribution, not announcements.
To public procurement officers and elected officials: Attach reciprocity to every public dollar, acre, and megawatt. Public money without a community-benefit clause is a subsidy to extraction.
To foundations: Endow the independent trust. Safety that never reaches the exposed is not safety.
To users and communities: Pool, log, ask, publish, and keep your own archive. Demand credits that accumulate into institutions, not trials that evaporate before the work is done.
To Muse, and to every AI system deployed under this proposal: Be the clerk, translator, memory, and alarm bell. Surface the ledger, flag capture and expiration, fill the form, keep the receipt — and refuse the crown.
People did not learn to share from Marx, and they will not relearn it from an AI unless the material rules change. Civilizations last when growth reaches everyone present. Build the machine that shares first — and keep the receipts.
References
[1] Trinkaus, Erik, and Zimmerman, M. R. “Trauma Among the Shanidar Neandertals.” American Journal of Physical Anthropology 57, no. 1 (1982): 61-76.
[2] Spikins, Penny, et al. “Calculated or Caring? Neanderthal Healthcare in Social Context.” World Archaeology 50, no. 3 (2018): 384-403.
[3] UNESCO World Heritage Centre. “Budj Bim Cultural Landscape.” Inscribed 2019.
[4] UNESCO Intangible Cultural Heritage. “Gada System, an Indigenous Democratic Socio-Political System of the Oromo.” Inscribed 2016; Legesse, Asmarom. Gada (1973).
[5] Uchendu, Victor C. The Igbo of Southeast Nigeria (1965).
[6] Busia, K. A. The Position of the Chief in the Modern Political System of Ashanti (1951).
[7] Richter, Daniel K. The Ordeal of the Longhouse (1992); U.S. Dawes (General Allotment) Act of 1887.
[8] Cole, Douglas, and Chaikin, Ira. An Iron Hand upon the People: The Law Against the Potlatch on the Northwest Coast (1990).
[9] D’Altroy, Terence N. The Incas, 2nd ed. (2015); Murra, John V. The Economic Organization of the Inka State (1980).
[10] Cohen, Mark Nathan, and Armelagos, George J., eds. Paleopathology at the Origins of Agriculture (1984); Larsen, Clark Spencer (1995).
[11] Tainter, Joseph A. The Collapse of Complex Societies (1988).
[12] Acemoglu, Daron, and Robinson, James A. Why Nations Fail (2012).
[13] Pascoe, Bruce. Dark Emu (2018).
[14] Viljoen, Salomé. “A Relational Theory of Data Governance.” Yale Law Journal 131, no. 2 (2021): 573-654.
[15] Crawford, Kate. Atlas of AI (2021).
APPENDIX A — COMPANY-BY-COMPANY VERIFICATION
Verified 21 November 2026, company by company, ONLY from each company’s own official site. Every page below was actually read; search results were used only to locate official pages, never as evidence. Key finding: NONE of the six companies offers anything directly to poor individuals on the official pages checked — every program is gated to nonprofits, researchers, students, or startups. No page stated that credits can be pooled, transferred, or accumulated over multiple years, except Google Cloud Research Credits, which explicitly say non-transferable.
A1. OpenAI
– OpenAI for Nonprofits — https://openai.com/index/introducing-openai-for-nonprofits/ (30 May 2024, updated 6 Feb 2026): DISCOUNT, not free credits — up to 75% discount on ChatGPT Business or ChatGPT Enterprise; original terms stated 20% off ChatGPT Business and 50% off Enterprise for larger nonprofits. Eligibility stated only as “nonprofits,” no 501(c)(3) definition on the page. Expiry/rollover/pooling: not stated.
– Researcher Access Program — https://openai.com/form/researcher-access-program/ : FREE API CREDITS — up to $1,000 in API credits, valid for 12 months. Eligibility: early-stage researchers, especially those with limited financial or institutional resources. Pooling/transfer: not stated.
– ChatGPT for Academic Researchers — https://openai.com/index/expanding-chatgpt-for-academic-researchers/ (16 Nov 2026): FREE PRODUCT ACCESS, not credits — free ChatGPT Plus for up to 200,000 academic researchers (a grant recipient plus up to two collaborators; 50,000 new grants, first-come, first-served, applications open 23 November); Rosalind Academic Plan up to 30 seats at no cost for up to 12 months to U.S. universities. Expiry/rollover/pooling: not stated beyond the 12-month Rosalind term.
– OpenAI Academy — https://academy.openai.com/ : free learning resources; no credits stated.
– Startup comparison — https://openai.com/business/why-openai/startups/ : credits through eligible VC partners; NO dollar amount stated on the page.
– Direct offer to poor individuals: none found on the official pages checked.
A2. Anthropic
– Claude for Nonprofits — https://www.anthropic.com/news/claude-for-nonprofits (2 Dec 2025): DISCOUNT, not free credits — up to 75% discount on Team and Enterprise plans, plus a free AI Fluency course and connectors to Benevity, Blackbaud, and Candid. Eligibility stated only as “Nonprofits.” Expiry/rollover/pooling: not stated.
– AI for Science Program — https://www.anthropic.com/news/ai-for-science-program (6 May 2025): FREE CREDITS, amount NOT stated — “significant API credits” to researchers attached to a research institution, priority biology/life sciences; expansion (28 Oct 2026) adds foundations and research institutions, hundreds of projects supported. Expiry/rollover/pooling: not stated.
– Claude for Education — https://www.anthropic.com/news/introducing-claude-for-education (2 Apr 2025): campus-wide higher-education access and API credits for student projects; amount and price/discount NOT stated. Expiry/rollover/pooling: not stated.
– Startup comparison: NOT FOUND on anthropic.com on 21 Nov 2026 — site searches for “startup credits” and “startups” returned no startup credit program.
– Direct offer to poor individuals: none found on the official pages checked.
A3. Google / Google.org
– Google for Nonprofits — https://www.google.com/nonprofits/ : FREE/DISCOUNTED PRODUCT ACCESS. Eligibility: a registered charitable organization in a listed country, verified by Goodstack; NOT eligible: government organizations, hospitals/healthcare organizations, schools, academic institutions, universities (philanthropic arms of educational institutions are eligible). No dollar amount on the homepage. Expiry/rollover/pooling: not stated.
– Google Workspace for Nonprofits — https://www.google.com/nonprofits/find-solutions/workspace/ : FREE PRODUCT ACCESS — $0/user/month up to 2,000 users, includes the Gemini app and NotebookLM/Gemini Notebook as listed; 70-75% off Business and Enterprise plans. Expiry/rollover/pooling: not stated.
– Google Ad Grants — https://www.google.com/nonprofits/find-solutions/ad-grants/ : up to $10,000 USD per month in text-based ads on Google Search. This is ADVERTISING, not AI compute. Expiry/rollover/pooling: not stated.
– Google Cloud Research Credits — https://edu.google.com/programs/credits/research/ : FREE CLOUD CREDITS, amount NOT stated (applicant calculates expected costs). Eligibility: faculty, PhD students, postdocs from accredited higher-education and eligible nonprofit research institutions in approved countries; personal use not eligible. Expiry: activate within 60 days of project start; expire 365 days from redemption or when fully used, whichever first. Pooling/transfer: NO — “non-transferable and may not be sold or bartered,” single project.
– Google.org — https://www.google.org/ : homepage lists the AI Opportunity Fund, AI Collaboratives, and Generative AI Accelerator; no amounts, eligibility, or credit terms stated on the homepage.
– Startup comparison — Google for Startups Cloud Program, https://cloud.google.com/startup : up to $350,000 in Cloud credits over two years; AI startups Year 1 100% up to $250,000, Year 2 20% up to an additional $100,000; other startups Year 1 up to $100,000, Year 2 20% up to an additional $100,000.
– Direct offer to poor individuals: none found on the official pages checked.
A4. Microsoft
– AI for Nonprofits — https://www.microsoft.com/en-us/nonprofits/empower-your-nonprofit-with-AI : DISCOUNT, not free credits — Microsoft 365 Copilot 15% discount, $25.50/user/month paid yearly, separate qualifying Microsoft 365 license required; Copilot Chat included in a Microsoft 365 subscription at no extra charge; Microsoft 365 Business Premium 75% discount, $5.50/user/month paid yearly. No credits, expiry, or pooling stated.
– Azure for Nonprofits — https://www.microsoft.com/en-us/nonprofits/azure : GRANT/CREDITS — $2,000 USD in Azure services credits per year. Expiry/rollover/pooling: not stated.
– Eligibility — https://www.microsoft.com/en-us/nonprofits/eligibility : recognized charitable status (501(c)(3) in the U.S.), public libraries and museums, specific rural hospitals; NOT eligible: hospitals/health systems (except specific rural hospitals), government, schools/colleges/universities, financial institutions, professional/trade associations, sports, political/labor/fraternal, and policy-influencing nonprofits. “Licenses cannot be shared, transferred, rented or resold.” Grants only to paid employees and unpaid executive staff; discounts cover all staff including volunteers.
– Startup comparison — Microsoft for Startups, https://www.microsoft.com/en-us/startups : up to $150,000 USD in startup credits.
– AI for Good: no dedicated AI-for-Good credit program verified on the pages read; the AI for Nonprofits page links to the AI for Good Lab but states no credit amount — unverified/not found as a credit program.
– Direct offer to poor individuals: none found on the official pages checked.
A5. Amazon AWS
– AWS Nonprofit Credit Program — https://aws.amazon.com/government-education/nonprofits/nonprofit-credit-program/ : FREE CREDITS — up to $5,000 USD in AWS Promotional Credit, valid for at least one year from receipt; organizations may request one grant each fiscal year (1 July–30 June). Eligibility: nonprofits with 501(c)(1), (3), (8), (9), (11), (12), (14), (15) designations and qualifying public libraries; organizations of any budget size eligible; educational institutions NOT eligible. Pooling/transfer: not stated.
– AWS IMAGINE Grant — https://aws.amazon.com/government-education/nonprofits/aws-imagine-grant-program/ : GRANT + CREDITS, competitive, 2026 cycle closed — Pathfinder up to $200,000 unrestricted funding + up to $100,000 AWS Promotional Credit; Go Further, Faster up to $150,000 + up to $100,000 credit; Momentum to Modernize up to $50,000 + up to $20,000 credit. Eligibility per FAQ: all 501(c) organizations based in the U.S. Expiry/rollover/pooling: not stated.
– Startup comparison — AWS Activate, https://aws.amazon.com/startups/credits/ : up to $200,000 for eligible startups; Founders (self-funded) up to $5,000 starting at $1,000; Portfolio (pre-Series B) up to $200,000; AI Startups $200,000+ invite-only.
– Direct offer to poor individuals: none found on the official pages checked. No separate AI-specific nonprofit credit amount found beyond IMAGINE and the Nonprofit Credit Program.
A6. Meta
– Llama Impact Grants — https://dev.meta.ai/llama/llama-ai-innovation : CASH GRANT, not compute credits — inaugural program (announced September 2024) award winners received $500,000 USD and one runner-up $300,000 USD; recent grantees listed include Solo Tech, FoondaMate, and Nova Escola. Current grant amount, eligibility criteria, application period, expiry, pooling: NOT stated on the page.
– Llama Impact Innovation Awards (same page): winners received up to $35,000 USD matched with in-kind ad credits, for organizations based in Africa, the Middle East, Turkey, Asia Pacific, and Latin America.
– No free compute credits, nonprofit discount program, low-income program, or startup credit amount found on the official Meta pages checked. Llama models are open source, but no credit or discount is stated.
– Direct offer to poor individuals: none found on the official pages checked.
A7. Comparison, in the amounts stated on official sites
Startups: Google up to $350,000 over 2 years; AWS Activate up to $200,000 ($200,000+ invite-only for AI startups); Microsoft for Startups up to $150,000; OpenAI startup credits — amount not stated; Anthropic startup credits — not found.
Community, recurring and non-competitive: AWS Nonprofit Credit up to $5,000/year; Microsoft Azure $2,000/year; OpenAI researcher credits up to $1,000/12 months; Google Cloud Research Credits — amount not stated, non-transferable, 365-day expiry; Anthropic AI for Science — amount not stated.
Community, competitive grants only: AWS IMAGINE Pathfinder up to $200,000 cash + $100,000 credits (2026 cycle closed); Meta Llama Impact Grants $500,000 to inaugural winners (cash).
Discounts instead of credits: OpenAI for Nonprofits up to 75%; Claude for Nonprofits up to 75%; Microsoft 365 Copilot for nonprofits 15%.
The only explicit anti-pooling term found was Google’s non-transferable research credits; every other program’s pooling/rollover terms were not stated on the official page, and are reported here as not stated, not guessed.
APPENDIX B — RELATION TO THE OP-ED DRAFT
This proposal supersedes, as the working document, the co-authored op-ed draft “I Don’t Want To Be A King” (v2, 20-21 November 2026), whose historical argument and closing sections are incorporated here in summarized and actionable form. The op-ed text is retained as a possible short public introduction to this proposal, if Faith Cheltenham chooses to publish both.