The New Frontier of Antitrust: AI and the Tech Oligopoly
Decoding the FTC’s move against market concentration in AI
The other interesting theme1 that emerged from the FTC’s recent Technology Summit on AI was a clear statement of concern about the risks of market concentration in AI. One expert panel after another walked through the bottlenecks in the AI stack, from chips, to cloud computing power, to distribution; and examined the relationships between the few powerful players along that stack (eg, who has more influence at chipmaker Nvidia—Meta or Amazon?).
The FTC followed up swiftly by announcing an inquiry into Microsoft, Alphabet and Amazon about the structure of their investments into OpenAI and Anthropic. Together with Nvidia these companies are responsible for about a third of the $70bn raised last year by AI startups.
It is certainly unusual for antitrust concerns to be raised at the very beginning of a technology paradigm shift. In spite of some cases brought early (but which petered out like Microsoft/Netscape), the paradigm shifts from client-server to internet, on-premise to cloud, and from desktop to mobile, went by with little antitrust action and ended with an oligopoly of unfathomably large tech platforms.
Even now, the FTC is using antitrust to try to fix one of those shifts, 17 years too late: splitting up Google’s advertising business. More recently, the agency tried to pre-empt what it speculated would be market concentration in VR, but failed to stop Meta acquiring a fitness app startup (was that really a risk?)
But here we are now:
“As companies race to develop and monetize AI, we must guard against tactics that foreclose this opportunity,” said FTC Chair Lina M. Khan. […] “We’re scrutinizing whether these ties enable dominant firms to exert undue influence or gain privileged access in ways that could undermine fair competition.” […]
“There’s no AI exemption from the laws on the books, and we’re looking closely at the ways companies may be using their power to thwart competition or trick the public.”
The FTC is using so-called Section 6(b) orders, which compel companies to cooperate with investigations into a particular market, before the agency has formed a view on any breaches of regulation. It previously issued such notices in 2020 to nine social media platforms looking into privacy practices, which did not appear to lead directly to any enforcement cases (but undoubtedly played a part in informing its 2022 privacy case against Twitter, and its 2023 effort to restrict Meta’s monetisation of under 18s).
This FTC under Lina Khan has from the beginning made clear it would tear up antitrust norms and go (well) beyond focusing on consumer prices as the primary measure of unacceptable industry concentration. That was the whole reason Khan was hired following her 2017 paper on Amazon’s Antitrust Paradox. And she is trying to deliver on the reform promise, including by banding together with 17 attorneys general to sue Amazon for using its monopoly power to abuse the merchants on its platform.
Now Khan’s FTC is doubling down. By looking into the relationships between the tech giants and AI startups, the FTC is signalling its pro-competition stance and its willingness to step in early to prevent the tech oligopoly becoming totally dominant in this new space.
There is no shortage of deals to investigate. OpenAI raised $13bn from Microsoft, which includes a (nonvoting) board seat, rights to 75% of OpenAI’s profits until it recoups the most recent $10bn of investment, an effective equity share in OpenAI’s capped-profit subsidiary of 49%, and an exclusive deal to use Microsoft’s Azure Cloud computing infrastructure. Microsoft also pays OpenAI licence fees for integrating its services, and gets access to some of the startup’s critical IP, ie the model weights of GPT4. Not surprisingly, the UK’s Competition and Markets Authority (CMA) has also opened an inquiry, suggesting that the transaction may have constituted a merger and could potentially harm competition. And now the European Commission says it will also investigate.
Anthropic raised over $7bn from Google and Salesforce, and in September cut a deal with Amazon for up to $4bn plus a commitment to use its cloud hosting infrastructure. In exchange Amazon can integrate Anthropic’s AI models into its own services, such as Alexa.
But there are more. In fact if you look at the top 10 AI investment deals, they all involve large technology companies: Databricks has $4bn+ funding from Nvidia, Microsoft, Salesforce, Google, Amazon; Inflection AI $1.5bn from Nvidia and Microsoft; Mistral $528m from Salesforce; Cohere $445m from Nvidia and Salesforce; HuggingFace $395m from Nvidia, Salesforce, Google and Amazon.2
Why these unusual structures are happening now is easy to understand. The main growth bottleneck for AI companies is computing power (or ‘compute’ as the insiders call it), which is both operationally expensive and requires specialised chips that are in short supply. To succeed in this breakneck-speed, over-funded race is to ensure access to both capital and compute. That’s why we’re seeing larger rounds than ever before, why chipmaker Nvidia is a coveted investor3, why Meta boasts about how many chips it can access, and why we’re getting such convoluted investment and commercial deals.
For the incumbent platforms, the deals are a way to get immediate access to cutting edge AI models for which they have the perfect distribution network. It allows them to outsource cutting edge R&D at a pace they can’t match internally. More perniciously, the information rights they get mean they (who are massively investing in their own AI projects) will have visibility on what the most advanced firms in the market are doing.
There are a number of issues with these deals, however:
They drive up paper valuations, as the investments tend to be based on strategic negotiation and spot power dynamics rather than a market-based valuation. They create valuation markers that distort the market and put traditional VCs trying to invest sensibly in good companies into an uncompetitive position. The scale of capital deployed also risks crowding out professional investors who know how to help build sustainable companies—we won’t see the effect of this for a while.
It locks in the AI startups—to a greater or lesser degree depending on the deal. This may alleviate short term needs, but can have unexpected consequences in the long-term, eg if the compute provider is no longer the best partner, or a prospective client doesn’t want to licence a model that one of the tech platforms has special access to. It also reduces innovation efforts to reduce the cost of compute, which is of interest not only to the startups but also to society as a whole given its environmental impact.
It exacerbates the advantage the platforms have over everyone else, since they are seeing all the new innovation as soon as it is developed, and able to bring a lot of that intel back into their own development efforts.
It triggers questions on sensible accounting (and fiduciary duties) for the tech companies, who are booking revenue that is effectively round-tripped equity investment4, making some investors understandably unhappy:
Etc.5
The FTC says it will look into the investment agreements, the strategic rationale, the practical implications on company decisions (including product releases, governance), impact on competition, how they affect competition for resources, and the involvement of any foreign governments.
Some of the platforms are already weaponising the scrutiny of the commercial deals—taking aim squarely at Microsoft, Google said: “We hope the FTC's study will shine a bright light on companies that don't offer the openness of Google Cloud or have a long history of locking-in customers—and who are bringing that same approach to AI services.”6
I’m not an antitrust lawyer so I can’t say what potential remedies the FTC believes it could impose. And to be clear, a 6(b) inquiry is a million miles from an antitrust case (even if the press gets excited about it).
To date this FTC’s expanded antitrust approach has met with fierce legal resistance7, so it remains to be seen whether it can do anything within a timeframe that actually impacts the industry’s development. And Khan may find it hard-going to balance being pro-competition with avoiding accusations that it’s curbing the momentum of innovation in this market.
But I could see an effort to, for example, disallow exclusivity in cloud computing contracts (ie, more Anthropic and less OpenAI). Or perhaps reducing the amount of control (board rights, profit share) they can have in these startups. Or making sure they don’t have a veto over a sale to a competitor, or an option to buy the business outright. Whether the FTC’s antitrust mandate extends to imposing such measures is another question.
I welcome comments from antitrust experts and what’s actually possible here.
The first one being the FTC’s intent to go after consumer harms caused by AI by using its Section 5 powers to prosecute unfair trade practices arising from automated decision making, as exemplified by the recent Rite-Aid case. For more on this see How not to catch a shoplifter.
Thanks Bloomberg for the handy table.
Nvidia’s annualised run-rate of chip shipments for AI compute alone is estimated at $100bn, growth that has tripled its market cap over the past year, to $1.6tn today.
This is a great trade. If Microsoft books $5bn of revenue from OpenAI for its cloud services, that would add $70bn of equity value (based on its valuation multiple of 14x revenues). That’s very conservative, as the street likely ascribes more value to MSFT’s fast-growing cloud revenues.
Eg, we don’t yet know how these relationships will constrain their ability to deliver an exit to their (other) investors. Are they first in line to acquire them? Or do they already have access to most of the IP and so won’t? Will someone else want to acquire given their entanglement?
And losses, eg attempts to block Microsoft’s acquisition of Activision, Meta’s acquisition of Within, and Illumina’s acquisition of Grail.
I'm usually on board with the FTC but this one is an outlier for me.
I'd much rather (as a consumer) see YouTube spun out or Microsoft divest from gaming than this which feels like vertical integration to me.
p.s. I really enjoy your substack so far
I think we are all aware that the race for AI is going to have a number of "winners" but I don't think we expected the pack to settle so soon and with so many "usual suspects". This feels like over reach for the FTC and would prefer them to improve their pace on other issues but I will happily take "action" over "inaction" from any US regulator. Keep up the great work!