OpenAI Bought 7 Teams by May. The Sourceable Pool Is 1,981 People.
OpenAI's April 13 Hiro Finance deal was its 7th acqui-hire of 2026. Here's the pattern, and the ~10-person teams to source before OpenAI calls them.
On April 13, OpenAI bought Hiro Finance. Ten people, a founder with a prior $200M+ exit, a product that was shut down a week later and whose user data was deleted by May 13. It was OpenAI's seventh known acquisition of 2026, and we aren't out of Q2 yet.
If you're sourcing AI talent in 2026, OpenAI is now your most aggressive competitor on small vertical teams. The good news: the pattern is unusually legible. The bad news: the pool is small, and Anthropic isn't bidding against you, which means founder optionality is compressing fast.
The pace: 7 deals in roughly 18 weeks
OpenAI did eight acquisitions in all of 2025. By mid-April 2026 it had matched almost all of that, with Convogo, Torch Health (~$60M), and Crixet in January, OpenClaw in February, Astral and Promptfoo in March, and Hiro in April. The corporate-dev signal arrived in December 2025, when OpenAI hired Google's Albert Lee to run M&A. The capital signal arrived shortly after: $122B in fresh funding at an $852B valuation. That war chest dwarfs any startup acquirer in history, and it's being deployed against teams of 10 to 23 people.
The verticals so far: developer tools (Astral), AI testing (Promptfoo), healthcare records (Torch), media (TBPN), agentic OSS (OpenClaw), LaTeX tooling (Crixet), and personal finance (Hiro, following Roi in 2025). The connective tissue is the ChatGPT "SuperApp" stack: Codex for code, Atlas for browsing, Frontier for evals, and a growing bench of vertical operators who've already solved the parts general LLMs are bad at.
Hiro is the cleanest tell
Hiro was founded in 2024 by Ethan Bloch and Rushabh Doshi, raised $6.3M from Ribbit, General Catalyst, and Restive, and shipped its consumer app about five months before the acquisition. OpenAI bought it, then shut the app down on April 20 and deleted user data by May 13. Nobody on the OpenAI side cared about the product. They cared about Bloch, Doshi, and the eight people behind them.
Bloch's credential is Digit, the automatic-savings neobank he sold to Oportun in 2021 for more than $200M. That's the real signal: a second-time consumer-fintech founder who has already proven he can ship a product millions of people trust with their money. The Hiro app was a vehicle to demonstrate that the team could rebuild that capability with a hybrid AI/deterministic architecture, because LLMs are notoriously bad at multi-step financial math and will cheerfully hallucinate an off-by-one on loan amortization.
If you're trying to identify the next Hiro, you're not looking for a product. You're looking for a founder with a prior $100M+ exit in a regulated vertical, paired with a technical co-founder who's already shipped ML at scale somewhere serious.
The Promptfoo archetype
Promptfoo is the same pattern at the upper bound of team size. Ian Webster led LLM engineering and the developer platform at Discord, scaling AI to 200M users. Michael D'Angelo was VP of Engineering and Head of AI at Smile Identity, where his ML served over 100 million people across hundreds of enterprises. They built a CLI that more than 350,000 developers have used, with 130,000 monthly actives, and adoption inside more than 25% of the Fortune 500. They raised $18.4M Series A from Insight and a16z at an $85.5M post in July 2025. By March 2026 they were inside OpenAI's Frontier team.
Note the headcount: 23 people across engineering, GTM, and ops. That's the upper bound. Anything bigger pulls regulatory attention (the $3B Windsurf deal collapsed for exactly this reason). The sourcing window is Series A or earlier, sub-25 headcount, with open-source distribution doing the work that revenue would normally do.
What "next-Hiro" looks like on a sourcing screen
Here's the explicit screen, built from the seven deals to date:
- Founder has a prior exit in the target vertical. Bloch had Digit. Webster had Discord. Doshi had operating experience. Look for second-time founders, not first-time ones.
- Co-founder shipped ML to 100M+ users somewhere serious. Not "worked on AI at a startup." Operated production ML at Discord, Smile, Stripe, Uber, Robinhood, Plaid, or similar.
- Team is 8 to 23 people, Series A or earlier. Below 8 and there's nothing to acqui-hire. Above 23 and the deal gets messy.
- Open-source wedge with measurable distribution. Astral has uv, Ruff, and ty. Promptfoo has the CLI. Eight of OpenAI's 17 total acquisitions feature open-source components. GitHub stars, PyPI/npm downloads, and weekly active OSS users are leading indicators that ARR isn't.
- The vertical is one where LLMs are unreliable on their own. Personal finance (arithmetic), legal (citation), medical (dosing), tax, CAD, scientific computing. Anywhere a deterministic engine has to plug a known LLM failure mode.
- Backed by Ribbit, General Catalyst, Restive, Insight, a16z, or similar. The cap tables on the 2026 deals rhyme.
The challenge isn't the screen. The challenge is that LinkedIn boolean falls over the moment you try to express "second-time fintech founder whose co-founder ran ML at Smile Identity, currently running a sub-25-person team with a Python OSS wedge." That query lives in five different fields, none of which are searchable in combination, and the second-time-founder signal isn't a LinkedIn field at all. It's a pattern across job history.
This is the gap Refolk closes. You describe the operator you want in plain English, and you get a ranked shortlist across GitHub, LinkedIn, and the open web. The query "US-based co-founder of a sub-25-person AI startup, prior exit in fintech or consumer, technical co-founder with prior ML role at a 100M+ user company" returns people, not 800 LinkedIn pages you have to read.
The pool is smaller than you think
Refolk's index shows roughly 1,981 US-based founders, co-founders, and CEOs who list LLM and generative AI as core skills. That's the entire realistic pool for "next-Astral" and "next-Hiro" teams, and it's concentrated heavily in SF Bay and NYC metro.
Filter that 1,981 down to people who have a prior exit in a regulated vertical, run a sub-25-person team, and have an open-source wedge, and you're looking at a few hundred founders. OpenAI is going to call roughly one of them per month for the rest of 2026.
The acqui-hire field is essentially uncontested at the lab level, which means OpenAI sets the price and the timeline.
Anthropic, for context, has done three acquisitions ever: Humanloop and Bun in 2025, and Vercept in 2026. They'd rather spend on compute. Google and Meta are buying, but at the senior-IC level, not the team level. So OpenAI is functionally the only lab paying acqui-hire premiums for vertical operators in 2026. Founders know this, which is why outreach timing has tightened from "this quarter" to "this month."
Where to fish
Three concrete pools to work this quarter.
1. The Frontier customer downstream
OpenAI's enterprise Frontier customers include Uber, State Farm, Intuit, and Thermo Fisher Scientific. The vertical AI vendors these companies have already paid for, signed compliance reviews with, and let into production data are the most natural acqui-hire targets next. State Farm has an insurance-AI vendor stack. Thermo Fisher has lab-software vendors. Intuit has tax/SMB-finance vendors. Walk those vendor lists.
2. The Astral / Promptfoo OSS shadow
The teams behind heavily-used but uncommercialized developer OSS are the cleanest signals. Sort PyPI downloads by growth, cross-reference against teams with a corporate sponsor, and check team size. Anything under 25 people maintaining a tool with millions of weekly installs is on OpenAI's list whether they know it or not. Refolk's GitHub coverage means you can build that list as "maintainers and core committers on the top 50 fastest-growing Python OSS projects of the last 12 months" without writing a single GraphQL query.
3. The "Roi precedent" verticals
OpenAI bought Roi in 2025 and Hiro in 2026. That's a pattern of two in consumer finance. The same pattern-of-two probably exists or is forming in: AI-native legal research (Harvey adjacent), AI-native medical scribing (Abridge adjacent), AI-native tax prep, and AI-native scientific writing. If you find a 10-person team in any of those verticals whose founder has a prior exit, you've found someone who is going to get a call by Q4.
What to actually do this week
If you're a recruiter at a Series B or later AI company, your defensible move is to build the bench OpenAI is about to drain, not to chase the people OpenAI already bought. The 1,981 number is your TAM. The 200 or so people inside it who match the screen above are your ICP. Reach out before the corp-dev call lands.
If you're a founder thinking about your own optionality, the math is brutal but clear: 10-person team, prior exit, open-source wedge, regulated vertical, and Albert Lee will probably find you on his own. The decision is whether you want that to be the only inbound you have when it arrives.
This is the workflow Refolk was built for: turn "the people OpenAI is about to acquire" from a vibe into a ranked list of 40 humans with names, GitHub handles, and cap-table context. The next Hiro is in that list. So is the recruiter who's going to lose them.
FAQ
How many acquisitions has OpenAI done in 2026?
By mid-April 2026, OpenAI had completed seven known acquisitions: Convogo, Torch Health (around $60M), and Crixet in January, OpenClaw in February, Astral and Promptfoo in March, and Hiro Finance on April 13. That nearly matches its full-year 2025 total of eight, and OpenAI's new corp-dev lead Albert Lee (hired from Google in December 2025) has signaled the pace will continue.
What does an OpenAI acqui-hire target look like?
The pattern across the 2026 deals: 8 to 23 people, Series A or earlier, a founder with a prior $100M+ exit in the target vertical, a technical co-founder who shipped ML at 100M+ user scale somewhere serious, an open-source distribution wedge, and a product that solves a known LLM failure mode (arithmetic for Hiro, evals for Promptfoo, Python tooling for Astral). The product itself is often disposable; Hiro was shut down within a week of the deal closing.
Why isn't Anthropic competing for these teams?
Anthropic has done three acquisitions total (Humanloop and Bun in 2025, Vercept in 2026) and is concentrating capital on compute infrastructure rather than acqui-hires. That leaves OpenAI as effectively the only frontier lab paying acqui-hire premiums for 10-person vertical teams, which compresses founder optionality and shortens the window for everyone else to make competing offers.
How do I find vertical AI startup talent before OpenAI does?
Index on founders, not products. The full US pool of founders with LLM/GenAI as core skills is roughly 1,981 people; the subset matching the OpenAI acqui-hire screen (prior exit, sub-25-person team, OSS wedge, regulated vertical) is in the low hundreds. Boolean search across LinkedIn won't express that pattern. A plain-English query against a combined GitHub, LinkedIn, and open-web index will, which is the workflow Refolk is built around.