In the race to dominate generative AI, big tech firms haven't just been building, they've been buying. But there's something strange about most of the deals that they've struck. Companies like Meta, Microsoft, Amazon, Google and NVIDIA are embedding themselves deep within the AI ecosystem through strategic investments, exclusive partnerships, and talent acquisitions with deals that stop just short of formal takeovers.
But the economic impact of these deals, it turns out, is indistinguishable from full control. This new approach, which some analysts are describing as non acquisition acquisitions, is designed to work around antitrust law. The idea is to avoid drawing regulatory scrutiny with deals that allow tech giants to secure privileged access to foundational by models, critical data pipelines, and elite research talent.
Despite all the talk of slashing regulation, and despite the support that Trump has drawn from Silicon Valley so far, the new administration is showing a surprising degree of continuity with the Biden era approach to antitrust enforcement, particularly when it comes to Big Tech and AI related deals.
According to the law firm Goodwin Proctor, US enforcers are still willing to challenge deals they view as anti competitive under traditional horizontal theories of harm, particularly mergers between direct competitors with leading market positions. So far, there's been no sign of reduced FTC enforcement. The tactics used by today's tech giants to quietly consolidate power in the AI sector are not
without precedent. More than a century ago, John D Rockefeller pioneered similar strategies during the rise of Standard Oil, where he often acquired companies without formal legal documentation, using silent partnerships, verbal agreements, and trusted intermediaries to quietly absorb
competitors. According to his biographer, Standard Oil executives worried that this secrecy could cause problems for the firm, as if a seller died unexpectedly, his heirs might mistakenly claim ownership of a refinery that Standard Oil had bought and paid for. To further obscure control of the businesses he had purchased, Rockefeller created the Standard Oil Trust in 1882, a legal innovation that allowed him to centralized authority over dozens of companies without technically owning them
outright. These tactics eventually led to the 1911 Supreme Court decision that broke up Standard Oil under the Sherman Antitrust Act. Today, Big Tech's use of non acquisition acquisitions, exclusive infrastructure deals and strategic talent consolidation echoes Rockefeller's playbook. Meadows recently announced $14.8 billion investment in Scale AI, which involved hiring.
The data labeling startup CEO will test how the Trump administration views Aqua hire deals, where a company buys a startup as a means of hiring its most important staff, which many have criticized as an attempt to evade regulatory scrutiny. Meta executed a $14.3 billion investment in Scale AIA leading data labeling firm, earlier this month. On paper, Meta bought a 49% non voting stake in Scale, but in practice they gained exclusive access to scale's Human in the
Loop data pipeline. The migration of Scale founder Alexander Wang and key staff to Meta's new superintelligence unit, and a web of contractual agreements that effectively embedded Scales capabilities within Meta's AI infrastructure. The structure of this deal meant that Meta could avoid the formalities of a merger while achieving many of its strategic
benefits. In an excellent sub stack article, which I'll link to in the description, Drayton de Silva described the deal as being like the way medieval Masons would move entire cathedrals from one side to another, brick by brick, by numbering each brick, transporting it to a new site, and then rebuilding the entire cathedral exactly as before. The reconstructed cathedral, he says, is in a new location, but in a spiritual and metaphysical sense, He argues, it's the same cathedral scale.
AI, which was once an independent supplier to the entire AI ecosystem, once this deal is done, now functions as a de facto internal division of Mehta. De Silva argues that there are multiple elements in the deal structure that make it vulnerable to FTC scrutiny. From an antitrust perspective, the deal raises red flags across multiple dimensions. De facto control input foreclosure, talent consolidation, coordination risk, and dominance entrenchment.
Even if regulators ultimately blocked the deal, Mehta may have already achieved A strategic victory by disrupting rival operations and gaining insight into scales operations. In a funny piece about Mark Zuckerberg's new bro look in the FT, Hannah Murphy argues that Facebook feels desperately dated today. Instagram has been losing ground to TikTok for quite some time, and the metaverse, after which Mark Zuckerberg renamed the entire company, is a ghost town,
but without even ghosts. The FTC is already pursuing an antitrust case against the firm, saying that it's acquisitions of Instagram and WhatsApp were anti competitive and aimed at monopolizing the social media market. The case is currently in trial, with arguments focused on whether Meta's acquisitions were part of a broader strategy to eliminate competition and maintain its dominance.
Meta has struggled from the very beginning to innovate internally and has mostly had to rely on acquisitions for growth. If Meta loses this antitrust case, the company could be forced to sell off Instagram and WhatsApp. They would most likely still be allowed to hold on to the metaverse. And those creepy AI glasses that Zuckerberg wears with the built in cameras just because nobody else actually uses them? Meta has been very aggressive in its attempts to grow in generative AI.
And the new investments that Meta and almost all of the other big tech firms have been making in AI startups are not just financial investments. They're strategic and designed to secure early access to cutting edge models, proprietary data, and elite research talent. These deals, it could be argued, are driven by a belief amongst Big Tech CE OS that the next wave of AI breakthroughs will be shaped not just by algorithms, but by control over the infrastructure and inputs that
power them. Data labeling pipelines, cloud compute access, and model training workflows have become the new battlegrounds. Microsoft's relationship with Open AI is a master class in contractual leverage, but the deal still has its flaws. Since 2019, Microsoft has invested over $13 billion into Open A is for profit subsidiary, securing exclusive commercial rights to its models and integrating them deeply into its Azure cloud platform and
enterprise software products. Yet despite the scale of this investment, Microsoft holds no formal equity stake in Open AIS nonprofit parent and no voting control over its board. This structure was carefully designed to avoid the appearance of a merger. Microsoft didn't report the transaction to the FTC at the time because the investment didn't amount to control of the company. Under U.S. law, Open AI remains an independent entity, at least
on paper. But in practice, Microsoft has a level of influence that rivals are exceeds that of a majority shareholder. The FTC is now investigating whether this partnership constitutes a de facto acquisition. A year and a half ago, when Open A IS board of directors ousted Sam Altman over concerns about AI safety, Microsoft received just minutes notice of what was going on and its executives were not consulted in the decision.
At the time, Microsoft offered Altman and the core research team roles within Microsoft, which would have likely drawn regulatory scrutiny. The bizarre structure of Open AI, which includes A nonprofit entity and a capped profit subsidiary, has caused all sorts of conflicts.
Now that the company is trying to convert into a for profit company, the FT reports that Microsoft has considered halting discussions if the two sides can't agree on critical issues like the size of Microsoft's future stake in Open AI. If Microsoft was to walk away from these discussions, it could still rely on the existing commercial contract to retain access to Open AI's technology until 2030.
The Wall Street Journal recently reported that Open AI executives have discussed leveling antitrust accusations against Microsoft as a last ditch option for loosening the firm's control under their deal. So while Microsoft's deal may have been structured to avoid FTC scrutiny, the fact that they don't have full control is still causing them plenty of problems. Amazon and Google have each invested billions of dollars in Anthropic, the AI startup behind the Claude family of models.
In return, Claude is hosted on AWS and Google Cloud, raising concerns about cloud dependency and preferential access. NVIDIA, meanwhile, has taken equity stakes in dozens of AI startups, bundling those investments with early access to hardware developer support and Co marketing opportunities. In one of the stranger deals, NVIDIA invested in Core Weave, a business that buys and rents out access to NVIDIA graphics processing units for training artificial intelligence models.
Nvidia's investment strategy creates a self reinforcing loop. It funds the companies most likely to buy its products and in doing so, ensures continued demand for its hardware. This dual role raises questions about conflicts of interest and fair market practices. Preferential treatment can distort competition, not because NVIDIA is denying access outright to certain competitors, but because it's subtly shaping the playing field through capital and supply chain
influence. The term Aqua hire is used in finance to describe the acquisition of a company primarily to hire its employees rather than for the operating business. It's a strategy used by larger companies to quickly acquire specific talent, particularly in industries where skill teams can be hard to find through
traditional recruitment. The Information reported last week that Meta is in advance talks to hire the prominent AI investor, not Friedman and Daniel Gross, to help lead it's AI efforts, and that as part of those talks, Meta might partially buy out Friedman and Gross's venture capital fund, which holds stakes in top AI startups and is worth billions
of dollars on paper. If the talks are successful, Gross would leave Safe Super Intelligence, which he Co founded with the former Open AI chief scientist last year, and work mostly on AI products at Meta. As part of the deal, Meta could end up owning minority stakes in the start-ups that the venture capital firm has invested in, but allegedly won't get information about and control over these start-ups, according to the article.
Drayton da Silva's Sub Stack article shows how Meta broke its non acquisition of Scale AI into 4 main parts, each of which could raise red flags under US antitrust law. Meta bought 49% of Scale's shares, but with no voting rights, they got exclusive access to Scale's data labeling pipeline, a key resource for training AI models. They hired Scale's CEO and key staff, meaning that competitors could no longer access them, and they kept Scale CEO on Scales board while he leads Metas AI Lab.
Each of these moves on its own might seem harmless, but together they give Meta almost complete control without calling the deal a merger. Drayton points out the traditional antitrust enforcement was built around clear ownership and control, but in the world of technology, influence can be exerted through contracts, infrastructure and
informal arrangements. The FTC and the Department of Justice have recently started to adopt A substance over form approach, focusing on the economic reality of deals rather than their legal structure. The FTC has published 11 merger guidelines, and Drayton focuses on Guidelines 3456 and 11 in his piece, which are most relevant to this structured deal.
Guideline 11 is about how minority stakes can still give control, and under this deal, Meta owns 49% of scale and might be able to influence decisions with that leverage. Guideline 5 is about how blocking rivals from access to key resources is bad, and under this deal Meta gets exclusive access to scales data. Guideline 4 is about how hiring away key people can kill competition, and in this deal Meta is hiring away scale CEO and other top talent.
Guideline 3 is about how shared board members can lead to collusion, and in this deal scale CEO is on both sides. Guideline 6 is about how deals that make a dominant firm even stronger are risky, and Meta weakens its AI rivals by disrupting their access to scale. Drayton argues that even if the FTC or the courts blocked the deal, Zuckerberg already has a win.
Open AIXAI. Google and Microsoft have already scaled back or paused their work with Scale AI, citing data privacy risks and conflict of interest issues. Anthropic has not yet announced anything publicly, but it also works with scale. Together, these five labs plan to spend over $250 billion on AI related CapEx in 2025. Meta has already induced months long delays and disruptions to their critical workflows, likely costing them billions.
Drayton estimates that setting up and announcing this deal cost Meta approximately $40 million across external advisors and internal resources over six months, which is a very effective asymmetric attack on their rivals. He finishes up his piece with an amusing playbook for non acquisitions for Tech CE OS Hoping to avoid regulatory scrutiny. I'll leave a link to a sub stack
in the video description. It's worth checking out the deals being struck by Meta. Microsoft's Open AI partnership and Nvidia's customer investment loop all challenge regulators to look beyond formal mergers in assessing competition within an industry. While the deals that are being struck might not trigger traditional review thresholds, their cumulative effect can without a doubt be expected to reshape competition within an industry.
Big Tech CE OS don't seem to just expect AI to be a technological shift, but a structural shift too. They believe that as foundational models become the engines of the next digital economy, the companies that control their development, deployment and distribution could end up wielding extraordinary influence. Unlike past industrial consolidations, this one is being shaped less through traditional mergers and more
through strategic entanglements. While regulators may not have noticed this in the past, the sheer volume of these deals has brought this issue to their attention. At present, it would appear that there is plenty of competition in the AI space. There are a number of competitors and they're all racing to develop better and
better models. Some of the huge investment in the space has quite likely been driven by a belief that AI will have winner take all economics like Internet search did, where Google built the best search engine many years ago and then dominated that sector for the next few decades.
Since the DeepSeek breakthrough earlier this year, there's been growing concern amongst investors that while it might be extremely expensive to achieve AI breakthroughs, the new models once built might be quite easy to replicate. Meaning that there may not be a dominant AI company in the near future and that the industry could instead remain extremely competitive with very little pricing power.
Meta, Microsoft, Amazon, Google and NVIDIA are without a doubt trying to build moats around AI1 contract at a time, but regulators do appear to already be on to what's going on and could shut these deals down if they appear to be harming
competition in the industry. We've all seen the sudden transformation of Mark Zuckerberg from the left-leaning tech nerd who wrote an open letter to his child about his goal of building inclusive and welcoming communities while pouring millions into social justice causes into the gold chain wearing MMA bro who told Joe Rogan that corporate America had been culturally neutered and workplaces needed more masculine
energy. This all happened right around the time of Trump's election victory, and this metamorphosis could be explained as much by Zuckerberg's need to align himself with the new administration as an actual change. In his beliefs, should any of these deals be deemed anti competitive, being aligned with the current administration, along with the purchase of the correct meme coins could solve a lot of Mark Zuckerberg's
problems. As an example, the US Supreme Court unanimously upheld a federal law requiring TikTok to either be sold by its Chinese parent company or be banned in the United States more than six months ago. But enforcement and has been repeatedly delayed by executive orders from the US president, who even invited its CEO to his inauguration. It's not obvious that any of the big tech firms have anything like a monopoly on AI. At present, it appears to be an extremely competitive space.
While the technology is driving massive investment and strategic growth, profitability remains elusive for most AI companies other than NVIDIA. According to the Wall Street Journal, Microsoft loses around $20 per user per month on GitHub Copilot. The expectation for these companies, whose core businesses are very profitable, appears to be that long term gains will come as AI becomes more efficient, cheaper to run, and more deeply integrated into
enterprise workflows. At present they have a lot of money and are spending some of it on generative AI to make sure that they are still competitive in the future. They're also spending this money as a moon shot investment where maybe they'll develop a dominant product which will matter as much to the future Internet as search did to the Internet 25 years ago. The Economist made a very good argument in a recent article that the FTC case calling for the breakup of Mehta is without merit.
They say that the FTC's argument against Mehta wasn't compelling when it was launched in 2020, and in the years since the complaint was filed, it's only become less persuasive. the FT CS claim in the case is that network effects create insurmountable barriers to entry. But The Economist points out that the last few years have seen a rush of new successful social media start-ups. Some have fizzled, like Clubhouse and Be Real. Others remain small, like Blue Sky and Truth Social.
But the arrival of TikTok made a mockery of the idea that the market was closed to new entrants. They point out that partially because of TikTok, social media apps now revolve around short form video, which has put metas, Instagram and Facebook in competition with YouTube, Shorts and Twitter, who are all pursuing the same short form video audience that TikTok drew in. They highlight that Spotify, which was originally a music service, is pushing into video too.
Roblox, a gaming platform, acts as a social network for many teenagers. They say that as competition spreads across previously separate categories, the FT CS narrow market definition looks ever sillier. There's no doubt that all of these AI deals are being specifically designed to work around the FTC rulebook. And it does appear that some of the deals are anti competitive in nature in that they're being done to block access to key AI services, slowing down competitors.
But this scramble to invest and build better and better models while keeping prices low doesn't appear to be harming consumers and reflects how competitive the space is at present. While these deals are worth keeping an eye on, it's not obvious to me that they'll be stopped anytime soon. Thanks for tuning into this week's podcast. If you found it interesting, don't forget to subscribe and hit the bell icon so that you're alerted to new episodes.
A special thanks to my supporters on Patreon whose support makes this podcast happen. If you want to support the podcast, there's a link in the show notes. Have a great week and talk to you again soon. Bye.