The Federal Trade Commission (FTC) and the U.S. Justice Department (DOJ) have opened a joint public inquiry into whether to update antitrust guidance for collaborations among competitors. That’s good news. Modern markets—especially those shaped by artificial intelligence—need clear rules that distinguish genuinely harmful collusion from productive, welfare-enhancing cooperation.
No one seriously disputes that naked price-fixing and horizontal market-division schemes remain unlawful. But not every agreement among rivals amounts to a cartel. In innovation-driven sectors, collaboration often reduces risk, combines complementary assets, and enables new products and productive capacity that would not emerge nearly as quickly through atomized action. Law & economics scholars have long recognized this point, and it should anchor any new guidance.
AI provides a particularly useful test case. Building and deploying advanced systems requires vast, specialized inputs: semiconductors, cloud capacity, engineering talent, model-evaluation tools, cybersecurity safeguards, privacy-preserving techniques, land, electricity, transmission access, cooling systems, and sometimes shared technical standards. In that environment, antitrust overdeterrence can be as harmful as underenforcement. Guidance that treats coordination with reflexive suspicion will raise costs, slow deployment, and weaken dynamic competition. Sensible safe harbors and administrable rule-of-reason principles, by contrast, can promote innovation without giving cover to true cartel conduct.
That is the core point. Updated competitor-collaboration guidance should make clear—early and often—that collaboration aimed at expanding innovation, infrastructure, interoperability, privacy, and safety usually promotes competition. The law should target agreements that suppress rivalry, not those that make rivalry more effective.
The Danger of Seeing Cartels Everywhere
Law & economics has long emphasized error costs. In the collaboration context, false positives can be especially damaging, because many benefits of cooperation are dynamic and difficult to observe in real time. When the government deters a joint research project, a shared testing environment, or a standards effort, the harm rarely appears as a visible price increase or a smoking gun. It shows up later—as a product never built, a safety protocol that never matured, or a facility that came online months or years too late.
That concern carries particular weight in AI markets, which remain unsettled. Product boundaries are fluid. Entry occurs across multiple layers of the stack, from chips and cloud services to model development, fine tuning, and application-layer deployment. Firms that look like rivals in one dimension may act as complementors in another. A static antitrust lens can easily misread value-creating coordination as competitive harm. Speculative theories should not outweigh real-world evidence, and policy should not discourage the productive use of new technologies merely because they are new. Those principles should guide any serious competitor-collaboration inquiry.
The FTC and DOJ should also avoid vague guidance that recites familiar warnings about spillovers and information exchange. Businesses already understand that naked collusion is illegal. What they need is assurance that enforcers will not second guess procompetitive collaboration simply because independent action seems preferable in theory—even when it would be slower, costlier, and less effective in practice.
A broader institutional point reinforces this caution. Antitrust enforcers are not central planners. They lack the information needed to determine, ex ante, which combinations of technical assets, engineering teams, compute resources, and physical infrastructure are necessary to bring new technologies to market. Guidance should reflect that reality. It should target clear anticompetitive abuse without assuming regulators can outperform decentralized business judgments about how to innovate. Government efforts to “perfect” markets—especially during periods of rapid technological change—risk falling prey to the Nirvana Fallacy and making matters worse.
The Case for Letting Firms Build Together
The strongest case for new safe harbors lies in R&D. Research joint ventures can internalize spillovers, spread risk, reduce duplicative fixed costs, and combine specialized know-how that no single firm possesses in sufficient measure. Those benefits grow in AI and other nascent technologies, where research is expensive, failure rates are high, and the social value of success can far exceed what any one participant can capture.
The FTC and DOJ should say so clearly. Bona fide R&D collaborations should ordinarily receive rule-of-reason treatment and, in appropriate circumstances, safe-harbor protection. A well-designed safe harbor would cover collaborations devoted to research, testing, validation, benchmarking, or precommercial engineering, so long as participants remain free to compete independently downstream and any restraints are reasonably related to the venture’s legitimate objectives.
This would not break new ground. Congress recognized the value of collaborative research in the National Cooperative Research and Production Act, which reduces litigation risk for qualifying ventures and encourages innovation. The agencies should build on that foundation, not treat modern AI collaboration as uniquely suspect.
None of this means every AI partnership is benign. A joint venture can still mask collusion if its restrictions exceed what the project requires, if it facilitates price or wage coordination, or if it excludes firms for reasons unrelated to technical needs. That risk does not justify blanket skepticism. Sound guidance should distinguish sham collaboration from the real thing, not discourage both.
The agencies should also acknowledge a point often overlooked: some cooperation lowers barriers to entry. Shared safety tools, open benchmarking methods, and collaborative validation environments can make it easier for smaller firms to participate. Coordination upstream can intensify competition downstream. That dynamic deserves more weight in agency rhetoric.
The same logic extends beyond core model development. AI progress depends on complementary R&D in privacy-enhancing tools, model-evaluation systems, cybersecurity testing, provenance and auditability methods, and domain-specific applications in health care, manufacturing, and logistics. Firms often have strong reasons to collaborate in these adjacent areas because the resulting knowledge is nonrival, diffuses quickly, or must operate within shared technical architectures. Antitrust guidance should treat this kind of precommercial experimentation as part of the innovation process—not as a suspicious deviation from it.
Infrastructure Isn’t Collusion
America’s AI future will depend as much on physical infrastructure as on software. Advanced models require massive data centers, reliable energy, transmission access, networking, cooling, and developable land. These are not marginal inputs. They are essential complements to innovation.
Antitrust law should not criminalize efficiency in this setting. Joint purchasing, pooled site development, shared infrastructure projects, and coordinated efforts to secure power or interconnection can reduce transaction costs and solve real bottlenecks. They can also make otherwise uneconomic projects viable. When firms work together to assemble sites, finance improvements, or secure long-lead infrastructure inputs, the likely effect is not reduced output, but expanded capacity.
The FTC and DOJ should reflect that reality. They should adopt a safe harbor—or at least a strong presumption of legality—for collaborations aimed at bringing new AI-related infrastructure online. The key questions are straightforward: Does the arrangement expand productive capacity? Does it avoid spillovers into downstream price or customer coordination? Is any exclusivity reasonably tied to investment incentives rather than strategic foreclosure?
These questions have real-world stakes. Building AI data centers at speed may require coordination on land acquisition, transmission upgrades, specialized construction inputs, backup generation, and regional power contracting. A legal regime that treats these efforts with suspicion will slow deployment and may push investment elsewhere. Clear guardrails, by contrast, can accelerate domestic buildout while preserving the ability to prosecute naked buyer cartels or exclusionary conduct.
Competition policy should complement—not undermine—efforts to promote innovation and economic growth. Competitor-collaboration guidance offers a concrete opportunity to do so. Antitrust should not stand in the way of rapid, lawful, output-expanding AI infrastructure development.
The underlying economics reinforce the point. Data-center projects often involve indivisible investments, long lead times, and sequential approvals. Delay is costly. So is fragmentation. If firms cannot coordinate to aggregate demand, secure infrastructure financing, or share development risk, some projects will not proceed on commercially viable timelines. Guidance that ignores these constraints risks functioning as an unintended anti-investment policy.
Don’t Fear the Standard-Setters
The FTC and DOJ should also say more about standards. Properly structured standards efforts typically lower transaction costs, improve compatibility, reduce uncertainty, and open markets to new entry. That is not a loophole in antitrust law—it is why standard-setting organizations have long received rule-of-reason treatment.
AI presents a strong case for this approach. Firms may need to coordinate on testing protocols, incident-reporting formats, red-team practices, provenance tools, security baselines, and interoperability interfaces. These forms of coordination can improve market performance by reducing lock-in, enhancing comparability, and helping users and complementors connect across systems.
The agencies should also make clear that cooperation aimed at preventing dangerous model outputs, malicious misuse, or national-security harms is ordinarily legitimate. Collaboration around abuse indicators, secure deployment practices, or shared technical baselines can help internalize externalities that no single firm can address alone. Antitrust should not penalize firms for reducing serious risks—so long as those efforts do not serve as a pretext to exclude disruptive rivals.
Administrability matters here. Guidance should emphasize practical safeguards: objective technical criteria, transparent procedures, reasonable access rules, and limits on sharing competitively sensitive information unrelated to the standards effort. Where those features are present, enforcers should say plainly that the activity is unlikely to raise serious concern.
Clarity would do more than protect incumbents. Interoperability and standards can lower switching costs and make it easier for new firms to plug into larger ecosystems. In other words, coordination can expand the field of competition. That point deserves a central place in the agencies’ analysis.
The same logic applies to safety-testing consortia and secure model-evaluation arrangements. When firms jointly develop testing suites, secure sandboxes, or reporting templates that improve confidence in model performance, users can compare offerings more effectively and deploy them with lower risk. Those are classic competitive benefits. The agencies should say so directly, rather than leaving firms to infer legality from silence.
When Sharing Data Protects Competition
Modern antitrust debates too often treat information sharing as inherently suspicious. It is not. Whether information exchange is harmful depends on what firms share, why they share it, and the market effects that follow. In AI and digital markets, some forms of information sharing can improve privacy, security, and competition at the same time.
Consider privacy-enhancing technologies. Collaborative work on privacy-preserving machine learning, federated systems, secure multiparty computation, and related tools can help firms extract insights from data while reducing exposure of sensitive information. Joint benchmarking and best-practice development can also build trust and accelerate adoption. The National Institute of Standards and Technology’s (NIST) work n trustworthy AI and privacy frameworks underscores how shared methods and technical validation can improve outcomes.
The same logic applies to data portability. Common principles and technical tools for secure data transfer can reduce switching costs, facilitate multi-homing, and weaken lock-in. The FTC’s earlier work on data portability recognized these competitive benefits when efforts are designed with privacy and security in mind. Updated guidance should reinforce that lesson
A sensible safe harbor here would cover collaborations aimed at benchmarking privacy tools, developing trusted privacy standards, sharing privacy-protective best practices, and creating secure portability mechanisms. The usual guardrails would apply: firms should avoid unnecessary exchanges of current pricing, output, or customer-specific strategic information; use aggregation, anonymization, clean rooms, or independent administrators where appropriate; and ensure the collaboration is genuinely tied to privacy, portability, or security objectives.
Those caveats should not swallow the rule. The better rule is straightforward: collaboration focused on privacy and portability often promotes competition by lowering switching costs, building trust, and enabling rivalry on dimensions beyond price.
This is another area where antitrust should look forward, not backward. If firms fear that discussions of technical best practices will later be recast as improper information exchange, they will underinvest in shared solutions to privacy and security problems. That chilling effect would harm consumers and likely entrench larger incumbents that can address these issues internally. Smaller and midsized firms often depend on shared learning and common tools to compete effectively.
Competition Policy Meets Geopolitics
None of this means antitrust should morph into industrial policy. It does mean the FTC and DOJ should account for the institutional context in which AI competition unfolds. Advanced AI now intersects with national security, supply-chain resilience, and global technological rivalry. Guidance that needlessly discourages lawful collaboration in strategic sectors will have consequences that extend beyond antitrust doctrine.
That point has limits. It should not excuse genuine anticompetitive conduct. Still, sound economics recognizes that domestic infrastructure buildout, shared safety efforts, and efficient collaborative R&D can generate public benefits that firms cannot fully capture alone. When collaboration expands U.S. productive capacity, improves secure deployment, or accelerates innovation amid foreign competition, the case for avoiding false positives grows stronger.
The agencies need not announce a sweeping geopolitical program. They need only make clear that antitrust law accommodates lawful coordination that accelerates American innovation. That message would matter. It would signal that enforcers can distinguish between cartels that choke off competition and collaborations that strengthen it.
Clarity Beats Cautionary Boilerplate
If the FTC and DOJ want this guidance to matter, they should move beyond abstraction. They should offer concrete examples drawn from AI development, infrastructure buildout, standards, privacy engineering, and portability tools. Businesses need practical signals, not reminders that some collaborations fall under the rule of reason.
At a minimum, the guidance should state that the agencies ordinarily will not challenge bona fide AI R&D joint ventures, collaborative efforts to build or procure data-center-related infrastructure, standards-development activity focused on interoperability, testing, or safety, and information-sharing arrangements reasonably necessary to develop privacy-enhancing technologies or secure portability tools.
The agencies should also identify clear red flags that remove a collaboration from any safe harbor: naked agreements on price or output, labor-market collusion, restrictions unrelated to the project’s aims, exclusionary access rules without technical justification, or exchanges of competitively sensitive information that are not necessary to the venture. That kind of line drawing would do far more work than high-level warnings about vigilance.
Guidance should also include examples of permissible safeguards. Firms should know that firewalls, clean teams, independent administrators, time-lagged or aggregated data, and objective access criteria weigh in their favor. If the agencies want compliance, they should reward well-structured compliance architecture.
The bottom line is straightforward. In fast-moving technology markets, collaboration often forms part of the competitive process. When firms coordinate to innovate, build capacity, improve interoperability, protect privacy, or reduce safety risks, they often make markets more effective and more contestable. Antitrust should leave room for those gains.
The Right Guidance at the Right Time
The competitor-collaboration inquiry presents the FTC and DOJ with a choice. They can issue a document that restates familiar cautions, preserves uncertainty, and invites overdeterrence. Or they can produce modern guidance that addresses the institutional realities of AI-era competition.
The better course is clear. Safe harbors for R&D collaboration, practical assurances for infrastructure buildout, clarity on standards and interoperability, and protection for privacy- and portability-related information sharing would strengthen dynamic competition. None of this would legalize cartels. It would reduce the risk that antitrust policy—through vagueness and overbreadth—slows the innovation and market expansion it is meant to protect.
Markets work best when firms can experiment, invest, and cooperate to create value, subject to clear limits on conduct that actually suppresses rivalry. Updated competitor-collaboration guidance should reflect that principle. If it does, the AI economy will be better for it.
