If I Did It: How Meta Should Have Structured Its Non-Acquisition of Scale AI

Minority Stakes and Major Outcomes

Disclaimer: I only have publicly reported information about the deal reported in the news so all the analysis below has this limitation. Not financial advice. Not legal advice. Consult specialists. Do your own research. I may have a position in the companies discussed, etc. etc.

tl;dr: The Meta-Scale AI deal is exposed to an FTC antitrust challenge and there are cleaner, lower-risk means to achieving the same ends.

There are stories of medieval masons moving entire cathedrals from one site to another, brick by brick, to a less vulnerable location¹. They would number every block, cart each brick to a new site, and rebuild the structure exactly as before. The reconstructed cathedral is in a new location but in a spiritual and metaphysical sense, it is the same cathedral.

Less poetic, but the still very useful Nexus of Contracts theory proposes that a company arises from a group of contracts, both formal and informal. This bundle of contracts states the relationships, duties, and exchanges between shareholders, bondholders, managers, suppliers, customers, employees, contractors, regulators and the government. If a party accumulates enough power and rights across this nexus, the party can wield the same power as if it were a majority owner.

Meta’s non-acquisition of Scale AI has elements of this reassembled cathedral strategy. Instead of outright acquisition that would trigger an FTC review, Meta split a Big Beautiful Deal into multiple smaller components that disassembled Scale, the independent company, and then reassembled Scale inside Meta’s empire. However, there are multiple elements in this deal structure that make this deal vulnerable to FTC scrutiny. I’ll do some armchair quarterbacking and tell you how the deal should have been done.

There are 11 Merger Guidelines but I focused on the ones that correspond directly to announced deal terms and fact patterns. And I am aware that the UK’s Competition and Markets Authority and the EU’s Directorate-General for Competition exist, but we’ll cross those bridges when we get to them.

Zuck Wins Either Way
Even if the FTC or the courts block the deal, Zuck already has a win. OpenAI, xAI, Google, and Microsoft have already scaled back or paused work with Scale AI, citing data-privacy risks and conflict of interest issues. Anthropic has not 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. I estimate setting up and announcing this deal cost Meta approximately $40 million across external advisers and internal resources over 6 months, which is a very effective asymmetric attack.

Dissecting The Deal

Deal Rationale: Meta instantly gets proven AI leadership and [speculative] secures priority access to the expert-curated datasets that will power next generation foundational models. Meta also disrupted rival labs’ data labelling workflows, and gets insights into their proprietary training methods.

Meta executed at least four distinct but interlocking agreements where I think it has exposure to the FTC.

1. Equity Investment and Partial-Ownership Control Bypasses Guideline 11: When an Acquisition Involves Partial Ownership or Minority Interests, the Agencies Examine Its Impact on Competition

Meta paid $14.3-14.8 billion for 49% of Scale AI’s non-voting shares and “a deeper partnership”.

Where This Component Has Exposure
Guideline 11 directs the FTC to scrutinize minority stakes if this confers de facto control through board seats, veto rights, influence over capital budgets, or access to sensitive data. Meta’s combination of economic interest, and if it has any other special rights that could let it steer Scale’s R&D priorities, customer access, and its product roadmap, it would mirror full control (such input rights have not been reported but it’s a big risk if these exist). The TC Group precedent illustrates this risk:

TC Group (Carlyle & Riverstone)/Kinder Morgan (2007): Carlyle and Riverstone held minority stakes in Magellan Midstream and Kinder Morgan, along with board seats and veto & information rights. FTC said the combination of minority stakes, board rights, and information flow gave Carlyle/Riverstone de facto control over 2 rival pipeline systems. A consent order forced them to make the Magellan stake completely passive and build firewalls.

Revised Structuring
The size of the equity stake is not as important as the additional provisions such as special voting rights, the ability to nominate directors, or information rights, so Meta should avoid taking a board observer role, not take any information rights to preserve alignment without the ability to influence Scale’s strategic decisions.

2. Exclusive Data-Pipeline License to Bypass Guideline 5: Mergers Can Violate the Law When They Create a Firm that May Limit Access to Products or Services That Its Rivals Use to Compete

Foreclosure in antitrust can be read as ‘cornering a strategic resource’ or creating a chokepoint. There are 2 kinds of foreclosure: Input Foreclosure, applicable here, where a company corners a critical resource so competitors (data, technology, raw materials, or components) and refuses to sell it to competitors, or sells it only on unfavorable terms; and Customer Foreclosure where a company controls a distribution channel and channel and refuses to carry competitors’ products, or demands exclusivity cutting rivals off from important sales channels.

Scale granted Meta an priority, multiyear license to its human-labeling pipeline so rival AI labs face delays and/or higher costs if they lack the same priority access. Data labeling is an essential input for model training and exclusive access forecloses competitors or forces them to duplicate expensive data pipelines. Based on some estimates and how ‘the market’ is defined, Scale AI’s market share is 5-15%, but is likely much higher if you only consider the top 6-7 AI Labs that actually matter.

Antitrust Vulnerability
By giving Meta sole priority, rivals face higher costs or must rebuild parallel pipelines.

Surescripts (2019): The FTC charged Surescripts with using exclusive network‐access and message‐routing agreements to lock out rival e-prescribing intermediaries, blocking competitors from a critical digital channel. FTC mandated that Surescripts had to offer fully non‐exclusive, equivalent access to any qualifying intermediary, showing that the FTC will unwind priority-access deals that “may substantially lessen competition” even when no merger is involved.

Revised Structure
Using a non-exclusive data license with with equal access and no special priority would defeat the point of such an investment, so the real alternative would simply to have it as an informal/handshake agreement with the actual deal terms hidden, like pretty much every other enterprise deal. Meta could pay 3-10x more for Scale’s services and is included in design partnerships and as an alpha customer so Meta stays at the front of the line to ingest new data accompanied by more professional service arrangements and the better forward deployed engineers & engagement managers.

3. Talent Acquisitions (‘AcquiHires’) to Bypass Guideline 4: Mergers Can Violate the Law When They Eliminate a Potential Entrant in a Concentrated Market

Alexandr Wang, Scale’s co-founder/CEO, and some employees will join Meta. These acquihire arrangements removed Scale’s key employees from serving/working with Meta’s competitors, precisely the kind of talent consolidation the FTC views as eliminating potential future competitors before they can emerge. Guideline 4 bars transactions that eliminate “potential entrants” in concentrated markets acquisitions of nascent rivals or their talent.

Antitrust Vulnerability
FTC has never challenged a pure labor transfer (acquihire) so I expect some novel legal theory (and possibly new precedents) to be used if this is challenged. Current antitrust precedents are of and from the Industrial Age, and are anachronistic in the present Digital/Cognitive Age.

The NVIDIA/Arm and Lockheed/Aerojet precedents described below were about acquiring design, fabrication, and manufacturing capabilities, while with Scale AI, Meta is after processes, and know-how i.e. the intellectual capital. Poaching core talent hinders Scale’s independent ability to service Meta’s rivals and preempts future competition.

1. NVIDIA/Arm (2021): The FTC blocked NVIDIA’s $40 billion acquisition of Arm. The FTC found that giving NVIDIA control over Arm’s architecture would risk “raising rivals’ costs” or degrading access to the foundational technology downstream. NVIDIA walked away from the deal in 2022.

2. Lockheed Martin/Aerojet Rocketdyne (2022): The FTC challenged Lockheed’s proposed acquisition of Aerojet, then the last significant independent US supplier of critical missile propulsion technologies, arguing that vertical integration would allow Lockheed to withhold inputs or hike prices on inputs essential to rival contractors. Lockheed abandoned the deal.

Revised Structuring
Avoid orchestrated talent transfers tied to the investment, and simply let talent change their own employment to Meta using ordinary labor-market channels. The key employees could simply quit Scale and join Meta after their notice period under standard employment contracts and California bars non-competes so they can move away regardless of investment.

4. Governance Side Letters and Special Rights to Bypass Guideline 3: Mergers Can Violate the Law When They Increase the Risk of Coordination

Alexandr Wang will stay on as a director on the board of Scale AI while leading Meta’s superintelligence lab. Guideline 3 prohibits mergers that “increase the risk of anticompetitive coordination” (tacit or explicit collusion) including through shared board roles or privileged information sharing.

Antitrust Vulnerability
Access to Scale’s strategic deliberations facilitates tacit coordination.

ExxonMobil/Pioneer Natural Resources (2024): When Exxon agreed to buy Pioneer for $64.5 billion, Pioneer’s former CEO was slated for an Exxon board seat but he FTC barred that appointment, concluding the interlocking directorate would create a conduit for sharing competitively sensitive production and pricing information, thus raising coordination risks under Guideline 3.

Revised Structuring
Wang should resign from he board of Scale AI. Eliminate any governance side letters (if any) that provide additional information rights or input for Scale’s roadmap or strategy, and instead rely on customer-vendor feedback loops. Meta can pay super-premium rates, and its feedback can be incorporated through normal business methods.

5. The Collective Deal to Bypass Guideline 6: Mergers Can Violate the Law When They Entrench or Extend a Dominant Position

Guideline 6 targets transactions that “entrench or extend a dominant position,” by raising rivals’ costs, and depriving rivals of economics of scale. Meta has effectively deprived its rivals of Scale by using its equity investment and talent transfer to sow sufficient doubt in Scale’s perception as an independent company. The large AI labs and tech firms have already announced they will stop using Scale AI to avoid leaking sensitive information to Meta.

The FTC’s “substance over form” doctrine considers economic reality, not just contractual labels. By unbundling Meta’s deal into smaller transactions, it looks like Meta is attempting to avoid scrutiny but the FTC still has a really strong case.

1. Equity + Information = Control (Guideline 11)

2. Exclusive License = Foreclosure (Guideline 5)

3. Talent Moves = Eliminates Potential Entrants (Guideline 4, novel legal theory)

4. Wang On Scale’s Board + Talent Transfer = Coordinated Effects (Guideline 3)

5. All The Above Collectively = Dominance Entrenchment (Guideline 6)

Antitrust Vulnerability
Meta’s nexus of transactions with Scale replicates many of the outcomes of a conventional acquisition. The FTC needs to only show that the aggregate effect of those agreements substantially reduces competition, and it can seek to unwind the entire deal.

Amgen/Horizon Therapeutics (2023): Amgen’s $27.8 billion acquisition of Horizon threatened to let Amgen use its rebate toolkit to coerce insurers and PBMs into favoring Horizon’s two monopoly rare disease drugs (Tepezza and Krystexxa). By bundling or cross-rebating, Amgen could have raised costs for payers, which in turn drives up patient premiums, deductibles, and copays. The FTC sued on a Guideline 6 “dominance-extension” theory and reached a settlement that barred Amgen from bundling, cross-rebating, or conditioning contracts that tie the two Horizon drugs to any other Amgen product, and subjects Amgen’s future rare disease buys to FTC approval (the NVIDIA/Arm (2021) deal described earlier is also a precedent).

Which shows that the FTC will challenge cross-portfolio leverage; Meta risks using its social-media dominance plus Scale’s data chokepoint to lock rival AI labs out of critical inputs.

---

The Playbook For Non-Acquisitions

A company, BigCo, seeking deep alignment with a partner, SmallCo, while steering clear of the FTC could follow the blueprint below. Each measure resembles a normal commercial practice without exposing either party to antitrust enforcement scrutiny.

1. A Joint Venture with Equal Stakes

Form a new, standalone entity, NewCo. Draft the operating agreement or corporate charter to require unanimous board approvals for any major decision including budgets, R&D projects, capital calls, or executive appointments. Ensure NewCo’s charter is silent on any observer seats, veto rights, or standalone information-sharing agreements beyond standard fiduciary communications. Limit inter-company communications to routine financial and operational reports that every JV partner gets equally so do not use side letters granting extra access to roadmaps or R&D plans.

2. A Platinum Tier With Non-Exclusive Licensing/Leasing

Secure access to critical inputs under a non-exclusive, multi-year license and priced on a super premium basis. This approach neutralizes foreclosure concerns, since rivals cannot credibly claim BigCo has “locked them out” of vital resources.

3. Passive Minority Equity

Acquire only a small stake, say 30% of total equity, and waive any special rights such as board seats, veto powers, or exclusive information access. Without formal governance rights and standard financial reporting, this stake remains purely financial. It aligns incentives without giving the investor any ability to steer strategic decisions and keeps the investment comfortably outside de facto control.

4. Ordinary Change in Employment

When talent transfer is critical, do the usual hiring process of public job postings, and then use simple individual employment agreements. Avoid coordinated talent moves or acquihire contracts tied to the investment. That way, each hire is a normal labor-market transaction, not a disguised management takeover.

5. Informal Priority

If technical integration or service capacity matters, the SmallCo grants the BigCo a “first look” at technology/product/data and not through any written exclusivity covenant. Because nothing in the contract bars others from the same privilege, there is no foreclosure.

This playbook is a bit basic because I am keeping my best stuff for more lucrative opportunities rather than sharing it on a free blog. So if you’re looking to make big acquisitions, or you’re at Meta thinking about remedies, or you’re anyone else who wants to discuss the optimal deal structure, you can get in touch by DMing me here, or on LinkedIn. I’m sure we can negotiate a deal.

1. Segovia Cathedral was dismantled and moved by 500 meters in the 1500s, and Old West Kirk was dismantled brick by brick and transported by horse and cart in the 1920s to a new site a mile away.