What is OpenAI LP, and what does "capped-profit" mean?

Yesterday, 11 March 2019, the artificial intelligence lab OpenAI announced that it has created a new company, OpenAI LP. Until now OpenAI was a nonprofit research organization. The new company can take money from investors and give employees a share of future profits, but with a limit on how much anyone can earn. OpenAI calls this a "capped-profit" company.

Most of OpenAI's roughly 100 staff now work for OpenAI LP. The original nonprofit, which OpenAI now calls OpenAI Nonprofit, sits above it and keeps control through its board. In OpenAI's own posts, "OpenAI" now means the new company.

Why OpenAI says it needs this

OpenAI says its purpose is to make sure artificial general intelligence, or AGI, benefits everyone. AGI means hypothetical AI systems able to do most intellectual work as well as people; no such system exists today.

The announcement, written by co-founders Greg Brockman and Ilya Sutskever, gives a direct reason for the change: money for computing. The lab says it has seen that the most dramatic AI results use the most computing power, alongside new algorithms, so it has decided to grow much faster than it planned when it launched in 2015. It expects to spend billions of dollars in the coming years on cloud computing, on hiring and keeping top researchers, and on building AI supercomputers.

A nonprofit cannot easily raise that much. Donors get no financial return, and top researchers can earn stock in companies like Google instead. Sutskever put it bluntly in a Reddit comment quoted by TechCrunch: there is no way to stay at the cutting edge "without us massively increasing our compute investment."

OpenAI says no existing legal structure gave it the balance it wanted, so it designed a hybrid.

How the capped-profit structure works

OpenAI LP is a limited partnership, a type of company with general partners, who run it, and limited partners, who invest. Here, the general partner is OpenAI Nonprofit, legally named OpenAI Inc. The limited partners are the investors and employees.

Every limited partner can earn returns only up to a cap, negotiated in advance. For the first round of investors, the cap is 100 times their investment. OpenAI expects caps to be lower in future rounds. Anything earned above the cap goes to the nonprofit.

A worked example, following TechCrunch's illustration, shows how large that cap is:

Step	Amount
An early investor puts in	$10 million
Maximum return at a 100x cap	$1 billion
Profit on that stake beyond $1 billion	Goes to OpenAI Nonprofit

OpenAI argues that if it succeeds at building AGI, the value created will be many times larger than what it owes investors and employees, so most of it would flow to the nonprofit and, through it, to everyone.

Several rules are meant to keep the mission first:

OpenAI LP's main legal duty is to the OpenAI Charter, the lab's published statement of principles, not to investors.
Every investor and employee signs paperwork saying the Charter comes first, even if that costs them some or all of their stake.
The nonprofit's board controls the company. Only a minority of board members may hold a financial stake, and only those without one can vote where investors' interests and the mission might conflict, including decisions about payouts.

Try it: Imagine you are asked to invest in a company with a 100x cap. Write down one reason the cap would make you more comfortable and one reason it would make you hesitate. Then ask yourself: who decides when the mission and the investors disagree, and how would you check that the rule is followed?

Who is involved

The nonprofit board includes three OpenAI employees, Greg Brockman (chairman and chief technology officer), Ilya Sutskever (chief scientist) and Sam Altman (chief executive), plus five outside members: Reid Hoffman, Adam D'Angelo, Tasha McCauley, Holden Karnofsky and Sue Yoon. Elon Musk, an early backer, left the board in February 2018 and has no formal role in OpenAI LP.

Named investors include Reid Hoffman's charitable foundation and the venture firm Khosla Ventures. Altman stepped down as chairman of the startup accelerator Y Combinator only days before the announcement. OpenAI's policy director, Jack Clark, told TechCrunch that the people involved had been discussing the plan for more than a year.

Early reactions

TechCrunch's report captures both sides of the first day's reaction. It accepts that a nonprofit has a hard time outspending giant corporations in a field where budgets matter. It also questions the size of the cap. TechCrunch notes that a 100x limit allows returns large enough that the restriction may feel "limited" in name only, and contrasts the move with OpenAI's 2015 founding statement, which described research free from the need to generate a financial return.

What does this change mean for OpenAI's research and the public?

In the short term, OpenAI says its day-to-day work is not changing. It plans to keep focusing on developing new AI technology rather than commercial products, and to figure out how to earn a return only once it has created safe AGI. Its work continues in three areas: capabilities (what AI can do), safety (keeping systems aligned with human values) and policy (how AI should be governed). The nonprofit keeps running education programs and policy work.

The bigger change is scale. With outside capital, OpenAI can buy far more computing power and compete harder for researchers. Many observers expect larger models and more expensive experiments to follow.

For the public, the key questions are about accountability. The safeguards are written into OpenAI's own legal documents and depend on its board. OpenAI also repeats a pledge from its Charter: to avoid a race that pushes safety aside, it would join forces with an organization that shares its values, even if investors then received little or nothing. Readers who want to follow how the structure works in practice can watch for who joins the board, which investors sign on, and how future caps are set.