On November 15, 2026, just hours before the bilateral talks between Trump and Xi Jinping at the APEC summit, the United States, together with Japan, Australia, Israel, the United Kingdom, South Korea, India, Qatar, and the UAE—nine countries in total—signed a multilateral agreement in San Francisco officially titled the San Francisco Declaration on the Secure Foundations of Enduring Prosperity. Germany, Sweden, the Netherlands, and the European Commission are expected to complete signing next week, bringing the total number of signatories to more than 13.
The declaration has only one core provision, but it carries enormous weight: the parties commit to jointly establishing a “Joint List of AI Critical Technology and Supply-Chain Vulnerabilities,” covering semiconductors, rare earths and critical minerals, energy infrastructure, and advanced manufacturing. According to Deputy Secretary of State Jacob Helberg, the list will help member states identify which sensitive technologies advanced AI development must depend on, and where supply chains need diversification—in other words, reducing dependence on China.
From Initiative to Institution: The Second Phase of Pax Silica
To understand the weight of this declaration, one must first understand its predecessor. On December 12, 2025, the U.S. State Department held an inaugural summit in Washington and launched the “Pax Silica” initiative. Australia, Israel, Japan, South Korea, Singapore, the United Kingdom, and the United States signed the framework document as the first members. That cooperation focused on six areas: connectivity and data infrastructure, computing power and semiconductors, advanced manufacturing, logistics, mineral smelting, and energy, but its actual binding mechanisms were quite vague. According to Wikipedia, over the following year, India, Qatar, the UAE, the Philippines, and several Nordic countries joined in succession, expanding total participation to 24.
The San Francisco Declaration is the institutional upgrade of Pax Silica. The difference is that it writes the “joint vulnerability list” into a formal multilateral document for the first time and establishes a working group mechanism to continuously identify supply-chain gaps. Moving from a loose political statement to an operational information-sharing framework—the significance of this step lies not in what it proclaims, but in the fact that it turns “coordinated exclusion of China” from a hidden issue on bilateral negotiating tables into a documented collective commitment.
The Choice of Timing Is the Clearest Signal
The declaration was signed on the eve of the Trump–Xi meeting, and that is no coincidence. Diplomatically, such timing often serves as a bargaining chip—making the other side aware before it enters the room that a multilateral framework is already in place, thereby narrowing the space for unilateral negotiations.
Since 2026, the AI supply chain has repeatedly become a direct arena of U.S.-China friction: chip export controls have continued to tighten, retaliatory export measures on critical minerals have escalated on both sides, and direct competition between Huawei and Nvidia in global markets has become increasingly intense. The San Francisco Declaration wraps this series of bilateral actions in a multilateral shell, shifting the logic from “the United States versus China” to “a majority of democracies versus opaque supply chains.” This shift in discursive framework is where the declaration’s true strategic value lies.
The Practical Limits of the List Mechanism
Beyond the declaration’s symbolic significance, several key details deserve scrutiny.
First, the “joint list” is an information-sharing and identification mechanism, not a sanctions or prohibition list. The parties agree to report vulnerable links in their respective supply chains and build a shared understanding, but specific responses remain at each country’s discretion. The declaration proposes establishing “economic hubs” to attract investment in protected industries, but the locations, financing mechanisms, and implementation timetable have not yet been determined.
Second, the lineup of signatories itself contains structural tensions. India, Qatar, and the UAE all have substantial trade with China. In particular, India’s deep reliance on Chinese equipment suppliers in chip manufacturing means that this commitment will face persistent conflicts of interest at the implementation level. There may be a considerable gap between the parties’ willingness to “identify vulnerabilities” and their capacity to “actually reduce dependence.”
Third, the most substantive technical part of the declaration—which specific AI technologies will be included in the vulnerability list, how screening criteria will be set, and how often it will be updated—has not yet been made public. This means the most critical bargaining is still to come in working group negotiations.
Supply-Chain Costs: The Underestimated Industrial Impact
While geopolitical discussion dominates public debate, one clearly underestimated dimension is what this mechanism will mean for global AI compute costs once it actually operates.
The current global high-end AI chip supply chain is highly concentrated: TSMC handles leading-edge foundry, Dutch ASML supplies EUV lithography equipment, Japanese firms supply key materials, and U.S. firms design chip architectures. This chain is already very fragile; disruption at any link would trigger systemic shock. The “diversification” path proposed in the San Francisco Declaration is essentially building another, more dispersed but more expensive parallel chain alongside the existing one—trading efficiency for resilience.
Rare earths and critical minerals are another direct cost node. China currently controls about 60% of global rare-earth refining capacity and is the major exporter of gallium, germanium, and other critical materials for AI chips. Shifting these links away from China to other regions will not only take time but also mean higher procurement costs and longer process qualification cycles. For AI labs in the midst of a compute arms race, this will directly show up in the hardware costs of training and inference.
Historically, the decentralization of semiconductor supply chains has been extremely slow. Since the U.S. CHIPS Act passed in 2022, new fabs still take years from planning to mass production, and subsidy costs are high. If the San Francisco Declaration’s “economic hub” mechanism is actually launched, in the short term it is more likely to offset cost disadvantages through subsidies than to lower costs through market efficiency.
The Risks of Bloc Formation: Isolation Remains a Double-Edged Sword
Over a longer time horizon, the greatest risk of AI supply-chain bloc formation may not come from Chinese retaliation, but from diverging technology paths within the bloc itself.
The current Western-led AI system is built on the high integration of Nvidia’s GPU architecture, the CUDA ecosystem, and specific cloud platforms. China, meanwhile, is rapidly building a parallel ecosystem centered on Huawei Ascend and proprietary frameworks—an ecosystem that currently lags in performance but is moving further along a path of incompatibility with the West. Once the two ecosystems fully diverge, global AI developers will face pressure to take sides in choosing toolchains, which would be a substantial loss for research and commercial development efficiency.
Most participants in the San Francisco Declaration are AI consumers rather than core producers. TSMC (Taiwan), which truly controls process technology; ASML (Netherlands, expected to sign next week), which controls lithography equipment; and key material suppliers are the technical fulcrums determining whether this mechanism can actually operate. The declaration itself does not—and cannot—resolve the geographic concentration risk at these key nodes. What it can do is establish a coordination framework at the political level and provide multilateral legitimacy for technology controls.
Independent Assessment
The San Francisco Declaration is an institutional starting point, but its actual effectiveness depends on the unseen parts that follow: the specific contents of the joint list, the results of working group negotiations, and whether domestic politics in the signatory countries allow genuine supply-chain relocation.
In the short term, the declaration’s most certain effect is that it attaches a multilateral legitimacy label to AI supply-chain issues, doubling the political pressure on individual companies or countries in technology procurement. This signal pressure is real, even if implementation mechanisms are not yet in place.
In the medium term, if the “economic hub” mechanism takes hold, investment attracted by it will flow first to manufacturing bases that meet alliance standards, further accelerating the geographic reconfiguration of supply chains. For the AI industry, this means compute costs will remain elevated for some time, especially in high-end training chips and special minerals.
The assumption to guard against most is an inertial one: that this bloc-based framework will naturally improve supply-chain resilience. Historically, every large-scale supply-chain restructuring has created new concentration risks, only in different locations. Institutional commitments to diversification will not automatically translate into actual supply resilience—unless there is sufficient capital and time invested, and signatories are truly willing to pay the cost premium. This is the core variable that truly needs to be tracked after the San Francisco Declaration.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接