⇓ More from ICTworks

Anthropic’s IPO Is Dependency Event Disguised as a Funding Event

By Wayan Vota on November 19, 2026

anthropic ipo global health

Digital development practitioners are about to treat a dependency as a donation.

Anthropic is expected to begin trading as ANT on Nasdaq today. For anyone running Claude inside a ministry of health or a district hospital, the actor that matters most is the U.S. Commerce Department.

On June 12, Commerce barred foreign nationals from Anthropic’s two newest models, and Anthropic switched both off for every customer until July 1. Anthropic tells investors in its prospectus that it could happen again.

Our sector has a different view. The Chronicle of Philanthropy counted roughly 20 percent of Anthropic’s shares headed for “the public good,” worth $300 billion to $400 billion at the valuations floated in October. Add the $200 million Anthropic and Gates Foundation partnership for health and education in sub-Saharan Africa and India, and we see a new mega-donor.

I think the windfall story is compelling, yet the part nobody is discussing is already happening in Kampala.

Sign Up Now for more digital development insights

Let’s Talk About the Money First

The optimistic case deserves a fair hearing. Global health has a seat close to the money.

Neil Buddy Shah, CEO of the Clinton Health Access Initiative, chairs Anthropic’s Long-Term Benefit Trust, and the Chronicle reports he is one of four interim managers of the new Anthropic Public Benefit Fund. He previously ran GiveWell as managing director. Few people in AI governance have spent their careers on malaria and child mortality, and Anthropic chose one of them to chair its trust.

The Gates partnership adds substance. It pays to label African language data and release it publicly. Janet Zhou of the Gates Foundation told Reuters the design responds to governments’ worries about “proprietary lock-in and sovereignty.”

Those are real assets. They’re also relationships, and relationships aren’t a mandate. Nothing in Anthropic’s filings obliges anyone to spend a dollar in a low-income country.

Five Concerns We Should Talk About

1. Washington Holds the Off Switch

Anthropic’s prospectus describes a hard year with its own government. The Pentagon named the company a supply-chain risk. The president told federal agencies to stop using its models in February. Then Commerce imposed the June export controls, and Anthropic complied by turning off its frontier models in every country.

June 12 to July 1, 2026: Anthropic’s two best models were dark for every customer on Earth.

Older models, including Claude Opus 4.8, kept running, so most production systems probably survived. That misses the point. Martin Chorzempa of the Peterson Institute told Axios that export controls always “encourage the development of alternative suppliers,” and that uncertainty about U.S. access is “a big advantage to open models.”

African developers had already chosen the alternative.

Sunbird AI, a Kampala nonprofit, built its Sunflower models on Alibaba’s Qwen 3 to cover the roughly 40 languages spoken in Uganda, and released them as open source. Lawrence Nderu of Jomo Kenyatta University of Agriculture and Technology argued in July that open-weight models give African teams control that closed systems do not.

My bet is that we’ll remember June 12 as the week African ministries began defaulting to Chinese open weights for public-sector AI. Anthropic’s IPO capital can buy more GPUs. It can’t buy back the certainty Commerce destroyed.

I made the same sovereignty argument in stop fighting Starlink restrictions and support digital sovereignty. Governments will choose the model they can host, and ministries should write that choice into procurement by testing every AI tool on at least two models before they sign.

That choice brings a worry of its own.

A ministry that downloads Qwen escapes the U.S. kill switch and takes on a model trained under Chinese content rules. When Alex Colville of the China Media Project tested Sunflower in Luganda, asking each question three times, the model called China a democracy “with Chinese characteristics” and said Ugandan MPs earn their pay because they work hard.

That’s the only audit of its kind I’ve found. Somebody should fund the next ten.

2. The Public Benefit Fund Has No Development Mandate

According to Anthropic’s S-1 filing, the co-founders’ pledge to give away 80 percent of their shares is not binding. Moving 5 percent of shares into the Public Benefit Fund needs shareholder approval, and the fund’s permanent governance comes later.

The priorities Anthropic has named are labor-market disruption, possible direct payments to displaced workers and existential AI risk. Global health isn’t on the list. International development isn’t mentioned.

The numbers have shrunk too.

The $300 billion figure assumed a $1.5 trillion valuation. The filed price range values Anthropic at about $916 billion to $1.02 trillion on shares outstanding. By my arithmetic, 20 percent of that is roughly $183 billion to $204 billion, and the 5 percent fund is about $46 billion to $51 billion.

Here is my prediction.

The fund’s first large commitments will go to U.S. workforce transition, because that’s where Anthropic’s political risk lives. A company the Pentagon labeled a supply-chain risk needs friends in Ohio more than friends in Ugnada.

Shah brings a global health perspective to the fund’s interim leadership, and the co-founders’ stated priorities point elsewhere. Which way the fund leans will be decided while the bylaws are still blank.

3. Anthropic’s Employees Will Fund Pill Counting Software

The second pool of money belongs to employees. It may be the more certain of the two, because it leaves the company one donor at a time. Dario Amodei, Amanda Askell and Jack Clark have signed the Giving What We Can pledge. Transformer reported that the co-founders’ pledges alone are nearly ten times everything Coefficient Giving has ever given away.

That money will move through effective altruism channels that already concentrate on a few big givers. GiveWell directed $427 million in 2025. An EA Forum analysis of GiveWell’s data found that 39 donors giving $1 million or more supplied $359.7 million of it. I expect a few hundred Anthropic employees with liquid stock to double GiveWell’s annual giving within three years.

GiveWell does fund digital health, on one condition.

In June 2024 it recommended $1,000,186 to Dimagi to pilot CommCare Connect in Nigeria. The app pays community health workers about $2 for each verified door-to-door visit that delivers vitamin A, deworming pills, oral rehydration salts and zinc.

GiveWell then paid Innovations for Poverty Action $83,079 to visit 360 households and confirm the visits happened. It is also funding CHAI, with mSupply Foundation and UC Berkeley, to put an electronic supply chain system into health facilities in Laos and test it with a randomized trial.

The rule is clear. GiveWell funds software that measures commodity delivery and can be priced by a randomized trial.

Most of what our sector builds has no single commodity to count: the national health information system, the district hospital’s EMR, the telemedicine platform that routes a nurse’s question to a doctor. Searches of GiveWell’s published grants found none for that kind of shared infrastructure.

My prediction is that AI money flowing through GiveWell will make Dimagi’s model the default for digital health. Implementers will redesign products to count pills, because counting pills is what gets funded. Meanwhile the shared systems underneath will keep running on aid budgets that no longer exist.

4. Sixty Days’ Notice Doesn’t Fit a Clinical Validation Cycle

Anthropic promises at least 60 days’ notice before it retires a public model. In the first eight months of 2026 it retired seven, from Claude 3 Opus in January to Claude Opus 4.1 in August. Claude Sonnet 4 lasted about 13 months. Claude Sonnet 4.5 has a tentative retirement date of November 30.

A decision-support tool validated against one model version has to be validated again on the next. In my experience, that takes months in a public health system. Anthropic has committed to preserving model weights, but preserved weights don’t answer API calls from Kathmandu.

I expect a public Anthropic to retire models faster, not slower. Its documentation says it retires models to free capacity for new releases, and shareholders reward new releases. Anthropic could close this gap cheaply by offering public health and education buyers one model version guaranteed for 24 months.

5. Public Markets Will Pull Claude Toward Rich Customers

In the first half of 2026, 72 percent of Anthropic’s revenue came from customers billed in the U.S., up from 70 percent a year earlier. Anthropic’s own economists found that GDP per capita explains most cross-country variation in use, with no sign that low-use countries are catching up. Anthropic has also signed $518 billion in compute commitments, about 80 percent non-cancelable. A company with that bill will chase the customers who pay the most.

I’ll predict that the international share of revenue falls below 25 percent by the end of 2028.

The good news is that we can check, every quarter. Starting Thursday, Anthropic owes the SEC a 10-Q that breaks out U.S. and international revenue. The split goes by billing address, so a Nairobi startup paying through a U.S. cloud account counts as American.

Even so, these are audited numbers, which no corporate social responsibility report has ever given us.

Why I’m Still Hopeful

Every risk in this post already has someone working against it.

  • Sunbird AI built models for Ugandan languages that big tech treated as uneconomic, and put them on Hugging Face for anyone to use.
  • GiveWell, the most demanding evaluator in global health, wrote a million-dollar check for a mobile app and then paid to check 360 front doors to see if it worked. That’s the evidence culture our sector has claimed for twenty years and rarely practiced.
  • Anthropic published research showing its own product follows national wealth, which few companies would volunteer.
  • The Public Benefit Fund’s rules are still unwritten, with a global health leader among the four people who will write them.

Turning that into success takes a few concrete changes within the next year.

  • The fund needs a development mandate in writing, paid in cash, before shareholders vote on the transfer.
  • Digital health implementers need to bring GiveWell the kind of evidence it can price, which means running the rigorous trials many of us have avoided.
  • Ministries need models they can host and audit, whether the weights come from Hangzhou or San Francisco.

None of that requires Anthropic to become a development agency. It requires our sector to show up while the decisions are still open.

Anthropic will spend Thursday answering to Wall Street. We have about a year to make sure it also answers to Kampala.

Filed Under: Opportunity
More About: , ,

Written by
Wayan Vota co-founded ICTworks. He also co-founded Technology Salon, Career Pivot, MERL Tech, ICTforAg, ICT4Djobs, ICT4Drinks, JadedAid, Kurante, OLPC News and a few other things. Opinions expressed here are his own and do not reflect the position of his employer, any of its entities, or any ICTWorks sponsor.
Stay Current with ICTworksGet Regular Updates via Email

Leave a Reply

*

*