Notes on the recent trend of “Hire and License Out” deals in AI
Halos are made when souls leave their companies and ascend to the Clouds
Over the last year, a new breed of deal structure has emerged in AI: an alternative to acquisitions and hiring that shares traits of both yet isn’t quite either. Companies like Inflection, Character AI, Adept, Covariant and most recently Windsurf have used this new structure in a common pattern.
A core team from the startup–usually including the founders and research team–are hired by a company in tandem with a non-exclusive license to the startup’s IP being inked. In return, the startup receives substantial licensing fees that are distributed as dividends out to their investors and employees. And in the part that most baffles outside observers, the startup continues operating now under new leadership.
These aren’t acquihires. The buyers don’t acquire the company, they hire the people and license the startup’s IP.
And these deals share a few other common characteristics: they happen extremely quickly, have enormous price tags, and are only being used in AI (for now).
Despite their scale, they remain poorly understood. They don’t even have a name. Making it hard to talk about them.
I propose we call them HALO deals.
“Halo” comes from “Hire and license out.”
Besides being a backronym it meets the two other requirements for naming a transaction type: Sounding benign1 and being descriptive. People with halos have left their bodies to live somewhere in the cloud giants.
I will leave the technical details of the deal structure to some other discussion. Though they are very interesting for people who enjoy esoteric legal structures, and I hope one day I’ll be so lucky to get drinks with Matt Levine and Patrick McKenzie and discuss their gory minutiae.
HALOs are a chimeric deal structure. Legally and logistically they are hiring. But to investors and employees of the startup they have the financial return of an acquisition.
This is accomplished by having the license fee be the size of an acquisition and then distributing it out to the investors and employees of the startup.
The mechanics of HALOs have many structural quirks. The biggest is that the legality of HALOs requires that the startup (let’s call it the remain co) continues operating independently. This leads to a lot of external (and often internal) confusion about what is happening when a HALO deal occurs. In practice, it means HALOs must be structured such that everyone would be happy with the financial outcome on its own, and those staying at the startup will feel true ownership and control over it, neither beholden to former coworkers or investors.
From the outside, HALOs look bizarre. And there is a lot of fear, uncertainty, and doubt surrounding them. This makes sense since they are only a year old and still very much being matured. Also many people involved in them have been noticeably silent on explaining them–seemingly feeling it’s not in their interest to discuss them publicly.
Despite worries they are a breakdown of the social contract between founders, investors, and employees, they actually are an effort to preserve the social contract in a changed environment. Albeit one that is still in a messy primordial form closer to Convertible Notes than to SAFEs.
Acquihires revisited
HALOs are a descendant of acquihires. Popularized in the 2010s, acquihires are when a company acquires a startup primarily to hire its employees. Typically these happened because the startup was running out of money and needed to land the plane, but was not very valuable. In these cases the main remnant value to sell was a commitment of the founders and team to to work at an acquirer. These were rarely very profitable transactions, instead it was often founders committing to spending more time than they might otherwise in order to help make investors whole or find jobs for their employees.
Acquihires have been puzzling in some ways to economists and lawyers because on purely financial terms it is not obvious why they happen. Companies seeking to hire the founders or other key employees could spend that same money directly on poaching those employees–cutting out the investors and other employees whom they will not be working with after the transaction.
There are many discussions and hypotheses on this. I won’t go deep into them but this law paper on the topic from a decade ago is a fantastic survey of it. There are a few I want to note in particular.
One that comes up often is this idea of a social contract across the industry. Yes, there is no legal obligation by anyone to do anything. Especially in California. But the iterated nature of silicon valley has created norms around doing right by the investors and employees that contributed to a company such that founders and acquirers default strongly towards taking their interests into consideration when negotiating. This is the same social contract people worry about fraying with HALOs.
Another not mentioned in the legal survey that doesn’t get enough attention is that despite much of the money going to return capital to investors in an acquihire they were still often better offers to employees than they would normally get through a hiring process.
This was the core that made the math of acquihires math. Companies were willing to pay a premium when hiring via acquisition than they would hiring directly. And that created surplus to make investors whole and pay employees who weren’t being hired.
Like acquihires, HALOs are also about hiring the team. However, in these deals the acquirer places high strategic value on hiring the team so there is flush cash to make the deals work. It is this abundance of demand that both gives a lot of room to maneuver in finding a structure that works.
And like acquihires, HALOs have so far been structured to do right by everyone. Despite the fears that the social contract is breaking, that is not what we see with HALOs. Instead they have been deliberately structured to make sure that investors and employees in the remain co are participating in the upside like the company was acquired. There are definitely things to improve in the HALO structure to make this default embedded legally, but it’s worth noting that so far we have seen this converge to the same intent as acquihires.
It’s true HALOs would not exist if not for the current antitrust regime. Under both the Biden and Trump administrations acquisitions have become politicized and uncertain. This has had a chilling effect on acquisitions which can often become existentially devastating to startups if they fail to close in the eleventh hour. And undoubtedly avoiding this was an impetus for the creation of HALOs.
However, it is a mistake to think of HALOs as otherwise normal acquisitions contorted into a bizarre form. They are in fact hiring.
In some ways, acquihires are actually the odd legal fiction. Companies were in fact actually trying to hire the employees. And framing it as an acquisition was the legal fiction that best helped everyone.
Acquihires are a good legal fiction. And the industry would probably use acquihires for these giant deals if they could. But if hiring key talent is a casualty of the government blocking acquisitions, then it is cleaner to drop the fiction.
HALOs are more honest in many ways. What is happening is hiring. And also it is still important to make sure that everyone on the cap table has a good outcome.
It may still turn out that it ends up being more advantageous to use acquihires or some other method to accomplish this goal. Or that only in the peak of the AI hype cycle they temporarily exist. But I think it is wrong to think that HALOs are disingenuous. Today they are messy because they are less understood and developed. And I’m sympathetic to the worry that it is a bad sign when focus is on deal structures not the actual core work of building products. But I think they are market attempts to solve for a real thing and once figured out could be significantly simplified. And they have real advantages like their certainty and speed in closing.
It is very difficult to stop people from deciding where they want to work. And unclear something collectively we should want to try to stop at all. And compared to the alternative of being poached, HALOs are significantly better in communalizing the returns to everyone.
It is true that collectively we should want startups to go for building real enduring products and staying independent. But that is about what founders should want, not HALOs. They are a solution to make things work even as there is increasingly an inversion between the value of companies and the talent they are composed of.
The market, especially in AI, increasingly views the value of companies to be their people not their customers, product, or IP.
The machine that builds the machine
In AI especially we can see how insane the fight over talent has become. And it’s not just the HALOs that have been done. Funding rounds for companies that have just formed like Thinking Machines Lab at $10B. OpenAI’s acquisition of io and Jony Ive’s support at $6.5B. And not least of all Mark Zuckerberg’s recruiting / buyout spree to build out an AI lab.
There are a number of reasons for this. Large companies winning or losing in AI will be worth trillions to them, making any amount of money paid to someone inconsequential if it can move the needle. This is really a consequence of the size of large tech giants relative to the startup ecosystem today. Over the decades large cloud giants have grown to be the majority of the tech ecosystem by market capitalization–and even more so by revenue. Correspondingly, the amount they can pay for someone who can affect their business has grown significantly.
AI is also particularly competitive because there is a relatively small set of people who have built the experience and intuitions from building at the frontiers of scale that are sought by top companies who all share similar needs.
But we’re also seeing that perhaps value itself is not held in the hard assets of companies.
Tech, and especially AI, is increasingly deflationary. Every year the advances in AI are obsoleting the last year’s models. The knowledge gleaned from training the last generation of models or from building products that best utilized them might be essential for working with the latest models–but the actual old models or products will be outdated fast. Conversely, as it gets easier to build software every year, the value of owning the legacy codebase falls or can even go negative.
Windsurf’s recent acquisition is an interesting example. It had a real product and business with $80M+ ARR. And OpenAI’s original acquisition would have been for the entire business. So what exactly is Google buying by doing a HALO? The thing they care most about capturing is not the current customers or revenue. It’s a team of people who have proven they can identify and organize their efforts in a way that is able to repeatedly translate these models into products customers want. Google, despite its superior models, has not yet shown they can do that–whatever one thinks is the thing getting in the way. The belief that the Windsurf team can do this is the core thing Google is trying to ingest. The details of rebuilding (or building anew) the product or customer base it believes are simpler to be redone.
This is not universally true of course. Despite the fact that “OpenAI is nothing without its people,” ChatGPT is perhaps the clearest example of traditional defensibility in AI right now and it would be incredibly difficult for its team to leave and supplant it.
Reading the Meta
If importance is shifting from valuing assets to people, the person who most aggressively understands the meta is Mark Zuckerberg.
He is not the first company to realize this. After all, AI salaries have had an order of magnitude of dynamic range in them for years now. But he is taking the current market realization that if talent is truly what matters the market is very inefficient in both pricing it and taking advantage of the relative ease in structuring ways to hire talent vs acquire companies.
And while he has not used HALOs, each of his moves has shown us more of this new meta we occupy. If it is important and valuable, there are many ways to make things work. His hiring of Alex Wang, Nat Friedman, and Daniel Gross were structured as minority buyouts of their respective company and firm (also a clever deal structure worth analyzing). And for those labs he could not acquire he has been very aggressively and publicly poaching key talent at very high compensation.
Once again, like with Google, what Zuckerberg is buying is not a product or lab, but people he thinks understand how to manifest the functioning AI lab he feels Meta lacks.
An interesting note on the hiring done by Zuckerberg is how much of his hiring is of a profile that maps closer to being founders than AI researchers. Across the industry we are increasingly seeing companies figuring out how to create setups that work well hiring “founders.” And there’s a lot to unpack in this blurring by the market of the founder role.
Underwriting people
It should not be surprising that the market is increasingly finding people can be more valuable than the company’s assets.
Venture has implicitly understood this for some time. One underappreciated wrinkle in valuing companies is that this can really be done based on a number of different methods such as their revenue, technology, or team. And at any given point a company’s ability to raise ends up being whichever of those is strongest. This in fact is why companies sometimes don’t feel comfortable raising at too high a valuation. Because they feel like the thing that justifies their current valuation (like their pedigree) may not be the same as what they will need to raise their next round (like say revenue), and they don’t want to raise so much that they won’t be able to support the valuation on that new metric. Another example of this is when tech companies slow down in growth and decide they need to sell to firms that judge them on EBITDA not their growth rate.
New AI labs have raised at very high valuations. Valuations that certainly cannot be justified by their revenue since they don’t have any. Nor their product, since they don’t have that either. What has carried these valuations is the confidence by investors that these collections of talent understand how to build at the frontier of AI and as importantly that they would be highly sought after by large cloud giants and AI labs in the event the company needed to sell. And that this sets some floor on the company’s value that lets investors invest at what would otherwise be too large sums for them.2
While most prominent in AI labs, this is not unique to them. Much early stage fundraising is because VCs think the founding team even prior to having product or revenue is valuable. They believe that the team of people will be able to understand what to build and be able to do so over the coming years in a way that will generate outsized returns. And they believe people will want to hire the team.
It should be no wonder that the rest of the market converged on this outlook too. Especially since part of the venture industry’s calculations on their downside protection was literally that companies would feel this way.
Parting thoughts
HALOs aren’t destroying Silicon Valley. They are a leading signal of the shift towards valuing people more than their companies. And a hacked together attempt to preserve tech’s social contract amid this wave.
HALOs are underspecified and will be improved upon. They are closer to a predecessor of convertible notes than they are to safes. They would benefit from having clear norms established and understood by everyone involved. And better yet to have these defaults codified into standard terms. If they really stick around I suspect they will also need to have a better structure.
I do think there is a place though for HALOs or something like them that is optimized for closing speed and certainty when the people are what’s important. Much of its problems are things that can be improved through better norms, defaults, and comfort with the structure. Whereas the things it does well like its speed are hard to replicate.
Some other random thoughts on improving HALOs.
HALOs should have their flow of returns mirror those of acquisitions. That means that the bulk of consideration should flow through pro rata to the entire cap table. All vested shareholders should be included in the dividends.
One grey area today seems to be whether unvested shareholders should have acceleration. I strongly think that double trigger acceleration should be triggered by this licensing event, just as it should exist for employees when there is a normal acquisition. If HALOs continue to exist the double trigger should be written into employment agreements.
It’s likely good to put terms into standard employment and investing agreements to make sure rights in HALOs match those in acquisitions.
Standardization of a “clean” HALO will be very useful. Agreed upon fair defaults would help everyone feel more confident in these deals. Companies are welcome to add bespoke terms to their specific deals, but these should be clear deviations from the norm.
HALOs are currently extremely tax inefficient, due to being taxed as income versus capital gains (or better yet QSBS). This makes them worse for employees than is properly appreciated yet. The top line numbers of HALO prices matching those of traditional acquisitions does not quite reflect the post-tax reality. Not sure if this is solvable but likely deserves more scrutiny than it has today.
Very little is understood yet on what is possible and how to think about the remain cos. And due to deal structure they are left on their own post deal. We’ve already seen a bunch of different configurations of how they have been handled. I expect there’ll be better understanding over time of what shapes they can have–and what is the best default. It’s probably bad dynamic to have the terms of this be negotiated in the eleventh hour by a new CEO.
Appendix: M&A is broken
One unexpected discovery of HALOs has been how fast they close. Part of this is not having governmental review, but that is just one aspect. They are significantly faster across the board, and can take an order of magnitude less time than a traditional acquisition with much higher certainty of closing.
Companies that go through HALOs are often shocked at just how fast they close. And there is adjustment to typical timelines of announcing deals that will need to be done when the time between deal closing and employees starting at a new employer can be days. There are things to fix in this.
But this is very material. Sometimes it takes actually seeing how things can be, to realize how broken things are. This is true universally, for example, without seeing China or Elon doing manufacturing it would be hard to realize how slow normal manufacturing is in the United States. But with a comparison, it can be discerned how much is fundamental and how much could be improved.
HALOs show us how broken the normal M&A markets are. Partially this is lack of clarity from the government on antitrust regulations. But this isn’t even what I primarily mean.
All acquisitions, even those nowhere close to being subject to HSR review, are so broken. In fact, it is the smaller acquisitions that often seem the most broken.
M&A can be the most stressful experience for founders. They must run the highest stakes process with very long closing time and high uncertainty and costs. And when those fall apart it can destroy the startups and seriously hurt the finances of employees’ families. But no one is improving this process.
Whether HALOs can be the answer to this is unclear. But they really point at the need for a set of SAFE-like defaults for acquisitions that can improve the main kinds of acquisitions we see like acquihires. Universally agreed upon norms for fair terms so that as much of the drawn out process can be removed.
This post was written at SHV and Padlet offices.
Endnotes
[1] YC truly took this requirement literally when they named SAFEs.
[2] This fundraising tactic can be recursive. Thinking Machines Lab raise created a very clever dynamic where their potential valuation rose as they recruited more founding researchers. And they could recruit more founding researchers as that potential valuation rose. And this flywheel could be repeated multiple times before actually locking in their fundraise.