Open Source as a Response to Antitrust
“Antitrust” refers to laws and regulations that prevent monopolies and promote competition in markets.
My take is that open source has emerged as an important form of response to antitrust actions. Specifically, antitrust has limited the ability of large companies to a) acquire new products, b) bundle current products and c) develop new products internally. Open source uniquely allows a business to commoditise one’s complements, without falling into these three antitrust buckets.
In part, this is how I see Facebook’s open source release of Llama AI models. There are parallels with Google developing Android. I’ll explain step by step.
Do note that I have Facebook shares in my investments, and have for some years now.
The Legal Basis for Antitrust
Some core – and largely agreed upon – bases for antitrust actions are those of “dominant market share” coupled with evidence of “consumer harm”.
Loosely:
“Dominant market share” means the company controlling at least a majority of the market.
“Consumer harm” means customers being charged more than they would be in a competitive market where that one company is not dominant.
Then, there are more nuanced reasons for taking antitrust actions, i.e.:
companies dominating data control (rather than high prices), or
companies stifling innovation and reducing consumer choice (quite a bit more abstract).
There is a debate around whether these more nuanced reasons have a sound legal footing. I am not going to address these questions, but the notion of “consumer harm” will crop up when referencing recent court cases.
Antitrust Actions and Reactions
There are a number of ways regulators might choose to address antitrust. For each action there are follow-on reactions from the market:
Regulators can block big companies from acquiring smaller companies:
Example: In 2023, impending challenges from the EU and UK regulatory commissions led to Adobe breaking off their acquisition of Figma (another design business).
Reactions:
The ability of larger companies to stay relevant is reduced as they cannot acquire new technologies. Unable to acquire smaller companies, firms instead hire their staff – for example, the 2024 Microsoft acquihire/license of Inflection AI’s staff.
With greater restrictions on acquisitions, smaller businesses are limited to IPOs and operation as standalone entities to provide a return to investors. Since acquisitions are a major source of financial return, this acts to reduce the level of investment in new businesses – which reduces long term competition.
2. Regulators can force companies to sell off parts of their business
Example: In 2024, a case is being taken against Google asking that the Chrome browser be split off from their business – as covered here by Alex Tabarrok.
Reactions:
In the short term, the large company is weakened and looses the ability to apply pricing power across products.
It’s less worthwhile for large companies to develop new products. Instead, incentives are shifted towards buying back shares or paying out dividends (often coupled with increased leverage/borrowing) as a means of growing stock price, e.g. Apple.
Now, hold onto these antitrust thoughts. Before I explain why open source is a response, I need to briefly describe the idea of “commoditising your complements”.
Business Strategy and Commoditising your Complements
In a competitive market, the straightforward strategies are to either a) lower your price or b) raise your quality relative to competitors. You can do that in the short run, through operational improvements, or you can do that in the long run – through investment. When you invest, you spend now and may get larger profits later. Investing is a riskier option from an antitrust perspective because – later – you may become a monopoly or earn high profits because of your earlier investments. This is particularly true because of survival bias.
A more nuanced strategy is that of commoditising one’s complements. It is not a “competitive” strategy, per se, because it is not about beating one’s direct competitors. Rather, it’s about making it as cheap and convenient as possible for people to user your products – so you can sell more of them.
If you are selling butter then you would like bread to be free. Why? because then people buy more butter.
“Commoditising your complements” is all about finding ways to make cheap the things that accompany or facilitate the use of your product.
There are many ways to do this. If you sell butter, you may just recommend to customers where they can find good, cheap bread. Or, you may sponsor a bread conference, where the bread industry shares best practises. You want the bread industry to be successful and to be competitive. You want bread to be a commodity.
This is commoditising your complements.
Open Source as a tool for Commoditising one’s Complements
If you sell hardware – say computers – then it can be of benefit to have the software be as free as possible, so people buy more computers. Windows managed the opposite of this, whereby making computers became highly competitive. This drove down computer costs, so Microsoft could sell lots of software.
If we take the example of a company selling hardware, one way they can create a competitive software market is by offering that software for free. This works in two ways:
The presence of a free offering forces other software providers to drop their price.
Developers – seeing that free software, which they may be allowed to build on (e.g. open source) – will often choose to use it as a basis for their software products. This lowers the barrier to entry for new competitors.
And so, open source software is at the same time i) a public good that benefits many, while also being ii) a valuable strategic tool. Joel Spolsky wrote on this decades ago here – a piece I found in Byrne Hobart’s new book “Bubbles and the end of Stagnation”.
And now, I can put the pieces of this story together.
Open Source? Antitrust Case Closed
To recap:
Large companies are more limited in acquiring smaller companies (Adobe/Figma) and developing their own products in house (Chrome).
Some short term options have emerged to circumvent this – like Microsoft’s investment in OpenAI via a complex minority share in subsidiary of a non-profit – although these options seem bad for everyone. Any new circumvention quickly comes under scrutiny in a regulatory game of whack-a-mole.
Large companies must be extra careful with any strategic partnerships they take on – lest they come under antitrust scrutiny.
Given this context, open source emerges as a response (or, if not a response, one of the few strategic options left) given antitrust strategy:
There is no partnership to scrutinise.
Forcing the company to divest the product is less meaningful, because the product is already free.
If open source software is not a commercial product, it is harder to argue that it is bundled (e.g. like Microsoft bundling Teams into Office 365).
Facebook & Ali Baba
As a concrete example:
Facebook makes money through ads based on the breadth of it’s social networks and messaging apps.
Facebook would like AI chatbots/assistants and recommender systems (which decide what’s in your feed) to be as cheap as possible (they would also like phones to be as cheap as possible, although that is harder given Apple’s position at present – although Android helped a lot).
And so – for now at least – if Facebook is spending tens of billions on AI services, it would like AI to be as cheap as possible. So, their approach is to give out the weights for powerful Large Language Models for free (Llama 2, 3 etc.), which encourages new companies to compete and drags down the price of AI. In doing so, Facebook limits the revenue of OpenAI, Anthropic and Google. In brief, Facebook (and Ali Baba with their Qwen models) are making AI extremely competitive and cheap.
This is both good and bad for consumers because:
i. it means AI is very cheap and competitive, but
ii. it is difficult to make money selling AI products, which kills the incentives for other companies to invest.
As a personal example, for my own business – Trelis.com:
My business would not exist were it not for open source models like Facebook’s, but
Since those models are free, and every other business in AI is forced to be near-free, it’s hard to get paid for products I build. This is a major part of why I sell life-time access to products rather than subscriptions. I don’t have enough pricing power to sell subscriptions.
This is the Theory, next comes the Practise
When Mark Zuckerberg originally doubled down by buying GPUs from Nvidia around 2022, it was because he saw their use within recommender systems for Facebook’s apps. Credit goes to Zuck not for predicting the rise of generative AI, but for having created optionality and made a right-sized early bet. Since then, Facebook have made much bigger bets – my guess is far bigger than one could yet justify by a “commoditising one’s complements” type of argument. (Value for money here is probably moot because Facebook showed they could build something new that was great. A few tens of billions is possibly a good deal for the advertising and talent that comes from that?)
Still, what remains unanswered is the course of pricing within AI – and whether “free” AI can go on. Thrive Capital investing at $100B+ valuation in OpenAI must not think so. Mark Zuckerberg has said that Llama models may or may not continue to be open source. Perhaps those are signals.
Markets are dynamic and difficult to predict. There are important exceptions, but markets – when left alone – are often the antitrust regulator’s best friend in clearing out incumbents and driving competition. When interfered with, antitrust can paradoxically lock-in existing players – either by raising the barrier to entry through complex regulations, or killing off the incentive to invest in new businesses if acquisitions are blocked.
It is complicated whether and how the antitrust landscape has directly led to more emphasis on open source. As I see it, there is some kind of connection. As to whether that in itself is a good thing for progress and for consumers is not entirely clear. If pushed, my moderately confident position is that antitrust has been overzealous – because it is too difficult to foresee the consequences, good or bad, of regulatory actions. While I view open source as positive on the whole for AI and software, it can have the downside of killing investment returns in the space.
If anything – and this is not antitrust related – the biggest issue with open source is that Llama has open source sexy again. Many early businesses pursue an open source strategy believing it can be sustainable – without realising they are the complement. And, although I primarily make proprietary software, the thought that I’m a complement gives me pause for my own business too.