The IPO money is coming. How much of it will actually have impact?
1
Anthropic listed on Nasdaq today at a market cap over \$2 trillion. Many Anthropic employees, including the seven cofounders, have pledged to give much of their new wealth away, and the company has pledged to match donations. For organizations hoping to benefit from this philanthropic windfall, the clock just started. Per the [S-1](https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm), 10% of employee shares unlocked at listing. Another 35% unlock after full-year 2026 results, and the rest at Q1 2027 results or day 180, whichever comes first. Cofounders wait longer, up to a year for most of their stake. OpenAI is still [<u>forecast</u>](https://www.metaculus.com/questions/44766/when-will-openai-ipo/) to list around Q2 2027. Our first maps asked 1) [<u>How much of this reaches charity</u>](https://radiant.metaculus.com/projects/1c3f7d1a-8ad9-4d6f-a906-90de0ec06f88) and 2) [<u>Which cause areas it goes to</u>](https://radiant.metaculus.com/projects/e80a5e79-b71f-4c4b-bf4a-821497fedcee). Then three Metaculus [<u>Pro Forecasters</u>](https://www.metaculus.com/services/pro-forecasters/) each picked up a different thread on : whether AI safety can spend its share, how the split changes if the windfall is small or huge, and what happens to one tiny cause area (AI welfare) when the money shows up. So we asked Metaculus [<u>Pro Forecasters</u>](https://www.metaculus.com/services/pro-forecasters/) to poke at that assumption. They built three maps, and all of hit the same wall from different directions. ## **Map #3: Can AI Safety spend the IPO money?** MaxA [<u>built this map</u>](https://radiant.metaculus.com/projects/9adefa92-4cb6-45eb-9c4e-c1ba07b610bd). He took the AI safety slice from our cause-split map (a central estimate of about \$37 billion by the end of 2031) and asks whether the field can actually grant it. That depends on people more than money. Coefficient Giving, who has thus far made some of the largest investments in the AI Safety space, says [<u>staff time</u>](https://coefficientgiving.org/?p=40627) is what limits its grantmaking. And across the whole field, about 140 people make AI safety grants full-time (75 of them at CG). The map turns that into capacity. In 2028, its central case has about \$7.9 billion of windfall money aiming at AI safety and about \$5.7 billion of room to place it well. **That's roughly 2/3 placed, with the rest sitting in donor-advised funds (DAFs) or drifting to weaker projects and higher salaries.** Stretch the window to the end of 2031 and it's close to a coin toss (52%) that the field places all of it. CG’s share of the field is something to watch. It was about 72% in 2025 and about 54% in 2026, and the map's central guess for 2028 is around 41%. A smaller share means more grantmakers elsewhere, which means more room. Longview, Macroscopic, and the OpenAI Foundation all [<u>plan</u>](https://manifund.org/ais-funder-bulletin) to scale fast. Some money can also go out in big pieces that don’t need many grantmakers: compute for safety research, endowing new institutes, prizes, advocacy campaigns. The map estimates about \$1.4 billion a year could go out this way. Some people argue it could be several billion a year; others doubt it’s possible at all. ## **Map #4: What changes if the windfall is small, or huge?** ADQ’s [<u>map</u>](https://radiant.metaculus.com/projects/16afa921-cbaf-4b7b-8155-6a14620b3a4d) is more qualitative. He runs a low and a high scenario side by side and asks how each one reshuffles the cause-area split. The spread between those scenarios is large. Forecasters put Anthropic's market cap at lockup end around \$2 trillion and OpenAI's around \$1.76 trillion, but each has a 3x to 5x range between its 10th and 90th percentiles. The low and high scenarios end up more than an order of magnitude apart. Here's the pattern the map draws: * **AI Safety** gets a bigger share when money is tight (donors who think it's most urgent front-load it) and a smaller share when money is flush, because it can only absorb so much. * **Climate and general science** go the other way. They can soak up big sums, so they gain share in the high scenario. * **Biosecurity** gains share in the low scenario (cheap, high marginal value) and loses it in the high one (hard to scale, plus dual-use worries). * **Animal welfare** loses ground either way, more so when money is abundant. * **Personal causes** take a bigger slice of a small pie, then shrink as donors refocus once their pet projects are funded. So Map #4 lands roughly where Map #3 does, by intuition instead of arithmetic: AI safety's share is capped by how much it can use. ## **Map #5: How will the small field of AI welfare meet this big wave?** Sergio [<u>zoomed in on one niche</u>](https://radiant.metaculus.com/projects/46b332d4-d9cb-4d67-a33c-3d48ebb3d91d): money for AI welfare (research, applied work and governance on whether AI systems have interests that matter). The author's rough estimate is that the field currently gets \$10 to \$30 million per year. The map’s central estimate is about \$280 million a year in documented giving across 2027 and 2028 (\$565 million total), or **roughly 10 to 30 times what it gets now**. A handful of people could decide most of this. Each of Anthropic’s 7 cofounders holds roughly 5 to 8% of the company’s pledged giving. So if just 1 cofounder sent 12 to 20% of their giving to AI welfare, that alone would be 1% of everything Anthropic insiders give. The map also gives Anthropic’s Public Benefit Fund a 35% chance of funding AI welfare at all. But like AI Safety, the AI Welfare field can only absorb so much. The author puts the field's capacity at about \$830 million in commitments over those 2 years. Past that, only about 40% of extra money is assumed to land. The map is honest about uncertainty. The headline blends the bottom-up model (70%) with the author's direct forecast (30%), and every node is tagged as a forecast, an estimate of the present, or a modelling shortcut. ## **What the 3 maps say together** Each map adopts a different focus, and in aggregate they move the question one step further: once the money arrives, can anyone use it *well*? For big fields with deep pipelines, probably yes. For AI Safety, a lot depends on hiring grantmakers. For a field as small as AI Welfare, the windfall could be most of the field's money within 2 years. If you're planning around this money, the things worth watching are pretty concrete: CG’s hiring, whether Longview and Macroscopic hit their targets, how fast insiders move pledged shares out of DAFs, and what Anthropic's Public Benefit Fund decides to fund. ## **What do you think?** These maps run on live forecasts and stated assumptions, with the authors’ reasoning clearly laid out. If you disagree with something in a map, open it, find the node you trust least, and fork it. Share it with us in the comments if you do. And forecast the questions underneath them: [<u>Coefficient Giving's AI safety funding</u>](https://www.metaculus.com/questions/45881/), [<u>CG's total grantmaking growth</u>](https://www.metaculus.com/questions/45374/), [<u>METR's contributions</u>](https://www.metaculus.com/questions/45385/), and the rest of the [<u>AI Lab IPOs series</u>](https://www.metaculus.com/tournament/ai-lab-ipo/). OpenAI also hasn't listed yett.