Algorithms remain, governance splits: TikTok’s U.S. service pivots from bans to management

Operations delegated to an Oracle-led joint venture
Algorithm handling emerges as the core fault line
Regulatory debate is spreading across the platform industry

The long-running controversy over the potential sale of TikTok’s U.S. business—once a flashpoint of U.S.-China technology tensions—is moving toward a provisional resolution through the creation of a joint venture and a recalibration of ownership stakes. By separating ownership from operational control, the arrangement lowers political exposure while establishing a U.S.-centric governance framework. In the process, questions surrounding recommendation algorithms and data control have emerged as decisive variables, extending even to a new experiment in country-by-country algorithm separation.

Changes in ownership and governance to keep the service running

According to Bloomberg on the 22nd (local time), TikTok operator ByteDance said in a recent internal notice that it had signed agreements with investors related to a TikTok U.S. joint venture, adding that the U.S. business would be operated as an independent entity with authority over local data protection, algorithm security, content moderation, and software assurance. Under the deal, TikTok’s U.S. operations will be entrusted to “TikTok USDS Joint Venture LLC,” established by Oracle, global private equity firm Silver Lake, and Abu Dhabi state-backed investment firm MGX. The joint venture is set to hold a 45% stake in the U.S. business, while ByteDance’s stake will be reduced to 19.9%.

The transaction suggests that the debate over banning TikTok—once emblematic of U.S.-China technology rivalry—has moved beyond outright confrontation into a phase of managed oversight. At its core is the crystallization of “Americanized operations” as the prerequisite for keeping the service available in the United States. Oracle, whose largest shareholder is Larry Ellison, a known ally of former U.S. President Donald Trump, is positioned front and center as both a technology partner and an equity participant, with infrastructure and control frameworks reoriented around U.S. companies. The direction points less toward a feared “service shutdown” scenario and more toward conditional continuity.

Under the proposed structure, Oracle, Silver Lake, and MGX would each hold 15%. Factoring in ByteDance’s reduced stake and the remaining 30.1% held by existing investors, ownership and control are expected to remain in a measured balance. The approach effectively formalizes a framework long operated under the banner of “U.S. data security,” converting it into a corporate entity and embedding regulatory demands into the design of operations. From the user’s perspective, the criteria for service continuity now hinge on who controls the data and how algorithms are managed.

Debate over TikTok’s U.S. operations began in earnest in 2020. During his first term, President Trump’s administration argued that the Chinese government could use TikTok to collect data on U.S. citizens or inject specific narratives into American society via the app’s recommendation algorithm, pressing for a sale of the U.S. business. In April last year, Congress escalated pressure by passing legislation to ban the app unless the U.S. operations were divested. The law was originally set to take effect this January, but after Trump’s return to office for a second term, enforcement was delayed multiple times, pushing the deadline to January 20 next year.

Separate operation and training of algorithms for the U.S. market

Uncertainty surrounding TikTok has found a turning point in the most contentious issue of all—the recommendation algorithm—through a proposed solution of national separation. The U.S. government’s core concern has consistently been control over the algorithm that determines what information is shown and how it is surfaced. The algorithm is viewed not merely as a filter for harmful content but as a mechanism that shapes public opinion, information diffusion, and the visibility of political messages. For this reason, U.S. authorities have maintained that even if data are stored on U.S. servers, national security risks persist as long as the algorithm remains under the control of the Chinese parent.

Industry sources say the newly formed TikTok USDS Joint Venture LLC will license the recommendation algorithm from ByteDance but retrain a separate artificial intelligence system using U.S. user data. While ownership of the algorithm remains with the parent company, operational control, training, and oversight would be managed by the U.S. entity. The structure is widely interpreted as a compromise that preserves Beijing’s red line on retaining ultimate algorithm ownership while partially accommodating Washington’s demands for tighter control.

Bloomberg described the shift to oversight and retraining without a full transfer of the algorithm as the essence of the deal, while noting doubts over whether the joint venture possesses the expertise to meaningfully monitor and verify a large-scale social media recommendation system. How code-level adjustments translate into user experience remains difficult to assess from the outside. Sarah Kreps, director of the Tech Policy Institute at Cornell University, similarly cautioned that while the arrangement may limit direct data access, it remains unclear how effectively it can curb more subtle forms of influence.

A political and institutional challenge, not a technical one

Within the industry, however, a prevailing view holds that the prolonged TikTok dispute stems less from technical hurdles than from entangled political and diplomatic interests. Most global platforms already operate regionally segmented servers, access controls, and audit logs at the data storage and processing stages, making it relatively straightforward to confine inputs—such as watch time, replays, likes, comments, and shares—to specific regions. The trade-offs lie in potential performance degradation and higher costs.

What troubled U.S. authorities was not the precision or efficiency of recommendation models, but who controls the data used for training and who ultimately governs the outcomes. Suspicion that the Chinese parent could influence the system, combined with a national security narrative that algorithms might amplify certain messages, pushed the issue beyond the reach of simple technical fixes. In this light, algorithm separation appears less a technical preference than a minimum condition for rebuilding political trust between the two sides.

The implications are rippling across the platform industry. As concerns grow over the role of recommendation algorithms in shaping public opinion, political polarization, and social conflict, governments are increasingly pulling algorithms into the sphere of public regulation. The European Union and China have already codified requirements around transparency, user choice, and risk assessment for recommendation systems, and legislative reviews are under way domestically over political polarization and social media accountability. Against this backdrop, the TikTok case underscores a broader message: what is technically feasible but politically sensitive ultimately converges on questions of ownership and governance.

Over the longer term, the foundations of platform businesses that rely on recommendation algorithms as a core competitive advantage may also be tested. Training models on integrated global data has favored rapid growth and scale, but country-by-country separation entails higher costs and reduced efficiency. Yet in a regulatory environment, “controllable algorithms” may take precedence over “high-performing algorithms.” Algorithms are no longer operating solely as technical systems; they are entering a phase where training and deployment are bounded by political and institutional constraints.