CADE ruled that Amazon’s 2023 minority investment in Anthropic was not subject to mandatory notification.
The Brazil competition authority Tribunal voted unanimously Wednesday to close an administrative merger proceeding concerning Amazon’s multibillion-dollar investment and commercial partnership with artificial intelligence developer Anthropic.
The case, opened by the Administrative Council for Economic Defense (CADE) in October 2024, examined Amazon’s acquisition of minority stakes in Anthropic and agreements involving computing infrastructure, technological development and the distribution of artificial intelligence models.
The Tribunal concluded that Amazon’s 2023 acquisition of a minority stake wasn't subject to mandatory notification. Anthropic’s Brazilian revenue in the year preceding the transaction did not meet the minimum statutory threshold.
Levi separately found that the companies’ collaboration agreements did not constitute an associative agreement under CADE Resolution No. 17/2016. Although the contracts created significant and increasingly integrated commercial links, the arrangements did not establish a joint undertaking or provide for the sharing of risks and results from the economic activity covered by the collaboration.
Levi said the relationship between Amazon and Anthropic has deepened since 2023 and has moved progressively closer to a joint undertaking. The agreements are multiyear arrangements that operate as a functional unit and create stable cooperation and growing interdependence between the companies, he said.
Those factors were not sufficient, however, to establish the shared organization and decision-making required for an associative agreement.
“The analysis reveals elements that bring the arrangement closer to a common enterprise, as well as functional unity among the multi-year agreements, and these factors demonstrate stable cooperation and growing interdependence. Conversely, no joint decision-making body, shared budget or accounting, mixed team, joint ownership of assets, joint pricing, common commercial targets, or joint responsibility toward customers was identified,”
Levi said.
Amazon and Anthropic have retained predominantly separate assets, personnel, products, responsibilities and commercial strategies throughout the different versions of their arrangements, according to Councilor Levi. Their growing cooperation hasn't produced a qualitative change in the allocation of decision-making powers, he said, and it “does not yet constitute a common enterprise for the pursuit of an economic activity, although the arrangement has progressively moved closer to such a structure."
He characterized the arrangements as a functionally integrated network of bilateral contracts involving qualified cooperation between companies that continue to maintain autonomous decision-making centers, assets, responsibilities and commercial strategies.
Levi said he had not identified a contractual clause, remuneration formula, practice or organizational mechanism showing that Amazon and Anthropic jointly bear and receive the risks and results of the same economic activity.
Levi’s warnings
Levi stressed that the decision is limited to the minority interests, contracts and provisions actually analyzed in the proceeding. Future changes in Amazon and Anthropic’s relationship may justify another CADE assessment, he said.
The authority said partnerships between computing-infrastructure providers and artificial intelligence developers may produce efficiencies, but depending on their structure may also raise concerns involving technological or commercial dependency, switching costs, access to competitively sensitive information and restrictions on rivals.
A future change in governance, decision-making powers, risk sharing, commercial integration or the allocation of assets and responsibilities could therefore alter the merger-control assessment.
CADE may also investigate how the agreements are implemented even if they don't constitute a reportable concentration. Conduct arising from the contracts could face scrutiny if it produces or is capable of producing harmful effects on competition.
The conclusion also applies specifically to the contracts and clauses submitted in the proceeding, he said. It doesn't establish that agreements between computing-infrastructure providers and artificial intelligence developers can never qualify as associative agreements.
Source: MLex
