The ITA Today – A Model?
By almost any standard of measurement, the Information Trade Agreement has been a major success. The raw numbers bear this out. According to UNCTAD, global trade for 2024-2025 in ICT goods (including semiconductors/ electronic components, computers, peripheral equipment, and telecommunications hardware) exceeded $2.5 trillion. ICT services (software and infrastructure) total global trade hit $1.2 trillion. Also, digitally deliverable services (includes things like ICT, digital media, financial tech, and remote business management) trade reached $5.4 trillion in 2025. But growth in IT trade does not by itself prove that the ITA caused that growth.
How about qualitative success? What did the ITA deliver? First, it showed that government rules and regulations governing international trade (and often, domestic rules and regulations) can and should take into account revolutionary change in global supply chains as well as business and personal requirements. A second deliverable: When it comes to IT-related trade policy, certain traditional practices such as MFN continue to work well, but others like tariff rate tradeoffs within product categories do not. Sector-wide treatment is preferable. And third, and perhaps most importantly, the ITA has proven that global technology supply chains do not respect national boundaries.
The supply chain today is vastly more interconnected today than when the ITA was negotiated. It goes from semiconductor equipment to the chips to advanced packaging to servers to networking to data centers to electricity to cloud computing to foundation models to applications to users. These are spread across many countries. It seems evident that the economic logic that served as the basis for the ITA in 1996 hasn’t disappeared.
ITA Drawbacks
Where did the ITA not succeed? Remember that it was always a tariff cutting/elimination series of steps, but a number of issues were not addressed by the ITA that continue to hover over the system. These include, for example,
Products outside the original coverage
Disputes over classification
Non-Tariff Barriers
Limits on tariff liberalization when tariffs are already zero
Services treatment
Treatment of data transfers
These and other shortcomings merit our attention, because –
The issues have become much larger.
Supply chains are considerably more complex. Technology itself has not only sped up, but it has also become more pervasive – more frequently impacting people’s workplace and everyday lives. Can the lessons learned during the ITA process be applied in other situations that in many ways look uncanily similar to the 1990s environment? One such situation comes readily to mind – Artificial Intelligence (AI).
Proposal – An Artificial Intelligence Trade Agreement (AITA)
The ITA addressed a pressing need, because a rapidly developing technology had created an interconnected global production system that existing trade rules did not adequately accommodate. Now, Artificial Intelligence (AI) is at the heart of another interconnected technology ecosystem that now more pervasively includes services, data, computing, software, and regulatory barriers.
We need to ask a question: Can an international framework therefore be designed to address these problems via a more manageable international system more tailored to the revolutionary new world? Obviously, it needs to be more than a tariff-focused agreement. We need a trade agreement that takes into account not only physical product trade, but also the very nature of trade and information itself. We refer to that as an Artificial Intelligence Trade Agreement, or AITA.
A “Product Landscape” for AI Tariff Cuts/Reductions
OK, despite what is said above, we begin with tariff cuts. The first part of an AITA would act as a direct descendant of the ITA. We learned during the ITA negotiations that a “product landscape” is an absolute requirement as a building block of an agreement. We saw the importance of identifying what products should be covered leading up to the 11th hour at Singapore. At a minimum, an AITA must follow the ITA concept closely by getting substantially reduced tariffs on a list of AI-enabling hardware. The list presumably would identify products like –
Advanced processors and accelerators
GPUs and other AI chips
High-performance computing equipment
Servers
Networking equipment
Storage
Semiconductor manufacturing equipment
Advanced packaging equipment
Optical communications equipment
Specialized AI infrastructure
Not only would such a list reflect the current state of play in AI technology, but it should as much as possible anticipate where technology is going. The ITA itself serves as an example, because it has updated itself as technology itself changed. In 2015, the ITA product list had 201 products added, showing that the agreement was dynamic and could evolve naturally rather than remain static in the context of the 1996 world. The AITA should, if anything, be more flexible and adaptive.
“Critical Mass” for an AITA
The ITA, as adopted in 1996, addressed a rapidly changing technological ecosystem: computers, semiconductors, telecommunications equipment, components, peripherals, and software. What differentiated these products from others in global trade was that they were increasingly interdependent. We have addressed here the connections that had become vital in the global supply chain, especially where one product fed into another that was assembled or manufactured in other countries, etc.
Artificial Intelligence is following a similar path, and a remarkably similar ecosystem is developing. By looking at the ITA’s historical precedent, it’s fair to ask whether international trade rules can adapt to a more advanced technology ecosystem. What rules applied locally, nationally, regionally, or even internationally stand in the way of this ecosystem, and can a sufficient number of countries be persuaded to address those rules?
When we face this question, we can’t just look at the relatively simple questions that involve the classification categories of IT products and their consequent tariff rates. No, AI creates a more complex and, some might say, a more compelling version of the same problem. Can we say, for example, that an AI accelerator is a computer component or something else? Is an AI server simply a computer? Is cloud computing a good, a service, or both? Where does an AI model reside in the traditional distinction between goods and services? What happens when the value of a product is found increasingly in software rather than hardware? On a different level,how can we even measure things like AI services, cloud services, data flows, AI model deployment, and others?
Put another way: How do we construct trade rules for an economy in which the boundaries among goods, services, software, data, and computing power are increasingly blurred?
The Surrounding Geopolitical Environment
The world of the late 1980s and early 1990s was a unique period. At the same time, we were witnessing, for example, the end of the Cold War era and the rapid advance of information-relevant economies. It may not have seemed that way at the time, but the global geopolitical environment was much simpler. And in that unusual period, the major technology and trading nations were able to find enough common ground to construct the ITA. Thirty years later, things are considerably more complicated. Instead of information-relevant economies, we see now the development of increasingly information-pervasive economies.
The countries stimulating this change are simultaneously:
Trading partners;
Technological competitors;
National security competitors;
Supplier and customers of each other; and,
In some cases, strategic rivals.
Now, we must ask: Can countries cooperatively develop rules governing a technology (AI) that they simultaneously regard as an economic opportunity, a strategic asset, and a potential security concern?
Is it Possible to Get an AITA?
Perhaps. The ITA wasn’t just the result of a couple of industry people coming up with an idea and government authorities agreeing that it was a good idea. The process by which it was developed, negotiated, and implemented took place in a rather unwieldy environment which itself needed to be transformed, including –
Reorganization of the international rules under a new body,
Restructuring an outdated nomenclature,
Changing industry needs requiring adaptation by governments,
Meeting the needs of both developed and developing economies,
Governments were brought on board incrementally,
Individual obstacles were dealt with one at a time,
Political leaders provided opportunities to move the issue forward, and
Eventually a sufficiently broad group of countries concluded that the benefits of agreement exceeded the benefits of maintaining individual protection.
So, can the structure of the world’s trading system be transformed by which the ITA might serve as a model for building a consensus around many of the issues created by AI? Note two things that are key to the question: First, we are talking about a model, and second we emphasize the phrase “many of the issues”.
Consider, first, the similarities between the 1996 world and that of today.
1996 ITA Possible AI-era Counterpart
Computers AI computing systems
Semiconductors AI accelerators / advanced chips
Parts & Components AI hardware components
Telecommunication equipment AI networking infrastructure
Software AI models and software
Tariffs Tariffs and other border measures
Product classification AI products & services classifications
Supply chains Global AI technology supply chains
Critical mass Engagement of key hi-tech economies
We see no one-for-one equivalency in the above chart. But it does seem to demonstrate the need to see whether the historical analogy is worth investigating. And the fundamental premise challenges us: In 1996, the challenge was finding common ground about the nations that were building the information technology economy. In the AI era, the challenge could very well be to find common ground among the nations building the next one.
As it turns out, the WTO itself is already starting to examine this issue. There is, within the WTO, a standing Committee on the Information Technology Agreement. As recently as June 23, 2026, the committee held a thematic session titled “ITA and Artificial Intelligence.” Specifically, it looked at the ITA as a possible vehicle – can the ITA be used to improve access to the technologies that make AI possible? Some of the questions posed by the Committee include –
Does the ITA’s tariff-free treatment of AU-enabling hardware still matter?
Is the existing ITA product coverage adequate for the AI era?
Can the ITA help narrow the global AI infrastructure gap?
How should the ITA interact with other WTO agreements and institutions dealing with AI?
Notice that the WTO seems to be focused principally on using the existing ITA framework to facilitate access to AI-enabling technologies. In fact, as recently as September 14, 2026, WTO Director-General Ngozi Okonjo-Iweala drew a direct connection between the ITA and AI development since it removes tariffs on which AI development depends. As she said about the supply chain: “No trade, no compute.” She is thereby making a link between tariffs and AI development. The Director-General especially identified the ITA, the General Agreement on Trade in Services (GATS), the Agreement on Trade-Related Aspects of Intellectual Property RIghts (TRIPS) and the Agreement on Technical Barriers to Trade (TBT) as existing areas that may address AI issues in part.
The truth is, however, that the ITA is insufficient to address the needs of an AI economy. Given that the current economic, security, and technology environments are more complex than ever, we are suggesting here negotiation of an Artificial Intelligence Trade Agreement (AITA)
Are There Areas that an AITA Should Probably Avoid?
Going beyond tariff to non-tariff barriers should avoid potential political minefields. We are NOT suggesting that an AITA should consider issues like whether AI is safe, ethical, responsible, or properly regulated. These are enormous questions, and we should recognize that governments have legitimate differences about how to answer them. For example, the following are areas that an AITA negotiation should probably avoid –
AI safety and existential-risk questions. These might ask -
How dangerous could advanced AI become?
What safeguards should developers employ?
How should governments deal with hypothetical superintelligence?
What level of testing should be required before deploying an advanced model?
These are huge issues and important ones, but they should be the topic of AI-safety and international-governance discussions, not a trade agreement.
Military applications of AI – clearly, such issues should be avoided. AI use involving autonomous weapons, military targeting, intelligence systems, battlefield AI, and other national security applications is far beyond the scope of a trade agreement
Privacy and protection of personal data – again, comprehensive international privacy and data protection. However, some related discussions might be held regarding cross-border data flows necessary for AI-related commerce.
Labor and employment policy – these include questions like whether AI will eliminate certain jobs, whether and how workers should be compensated, whether governments should provide AI-related unemployment benefits, and how companies should reorganize workforces.
Ethics – some have suggested that algorithm bias, fairness, discrimination, and acceptable use of AI should be on the table, but for the purposes of an AITA, they fall completely outside the scope of such an agreement.
AI development subsidies and industrial policy – some governments are already pursuing policies designed to encourage domestic semiconductor, cloud-computing, and data center industries. This area gets into something of a gray area, since questions can and should be asked about whether these policies represent trade-distorting barriers. If an AITA were to step into this area, its parameters should probably be strictly limited.
National security controls on sensitive AI technology – this is an area where trade policy and national security export controls have generally maintained an arms length. An AITA should not try to eliminate legitimate national security export controls. It is worth noting that national security exceptions will be required, just as they are under the WTO system.
What Would the Framework of an AITA Look Like?
OK, yes, tariffs should be a part of an AITA, but to stop there would be to stop far short of whether we can find trade-related rules that operate efficiently and productively. What we are suggesting here is this: The objective of an AITA would not be to harmonize national AI regulation. Rather, it should help to prevent unnecessary trade barriers from fragmenting the emerging global AI economy. An explicit goal should be to ensure that different national regulatory systems do not unnecessarily prevent AI technology, equipment, and services from crossing borders.
So, the AITA can and should address issues such as these:
Discriminatory treatment of foreign AI products and services
Data-localization requirements
Restrictions on cross-border data flows
Discriminatory licensing requirements
Unnecessary duplication of technical requirements
Interoperability
Transparency of AI-related regulations
Customs treatment of AI hardware and equipment
We should be careful, however – reaching too far would be an almost certain recipe for failure. Consider the above list, therefore, to be a kind of shopping list. Maybe the best is to go for the low hanging fruit first. Then, like the ITA itself, we should allow the list over time to become more expansive as countries see the demands of the global AI economy adjust and grow.
The ITA experience suggests a possible starting point: identify the areas in which governments have a shared interest in allowing the new technology to develop and move across borders, determine which trade barriers and policy differences stand in the way, and then see whether a sufficiently broad group of countries can agree on a common set of principles.
Such an effort should not require agreement on every aspect of AI policy. Nor should it eliminate legitimate differences over privacy, security, competition, intellectual property, employment, or regulation. For example, recent U.S. and Chinese suggestions about cooperation on AI controls are not the question addressed here.
Rather, trade negotiators should look for something narrower and more achievable — an international framework that keeps markets open for the technologies, products, components, services, and knowledge on which the AI economy depends. In that sense, an IATA would draw upon the most important lesson of the ITA – major international agreements do not necessarily begin with agreement on everything. They begin by finding enough common ground to make cooperation possible.
Lessons from the Past - Possibilities for the Future
ITA - AITA
The story of the Information Technology Agreement is, at one level, a story about tariffs. It began with a relatively straightforward question: why should products at the heart of a rapidly developing technology industry be burdened by tariffs that increase their cost and impede their movement across borders? But the history developed into something much larger. The ITA emerged from the changing international economy of the 1980s and early 1990s, the experience of the Uruguay Round, disagreements over semiconductors and other information-technology products, and years of work by governments and industry to identify a basis for agreement.
The agreement that concluded in Singapore in 1996 wasn’t the product of a single negotiation or a single government initiative. It was the accumulation of many seemingly unrelated pieces that, over time, became connected – a puzzle, or matrix, that came together. The Coalition that helped bring those pieces together demonstrated that an industry can play a constructive role in international trade policy without attempting to substitute itself for governments. Its contribution was to identify a problem, help build consensus among countries with different interests, and keep working until the political circumstances made agreement possible.
The circumstances surrounding the ITA also matter. It was reached at a particular moment in international economic history—after the Cold War, following the creation of the WTO, and before trade policy became deeply entangled with the political divisions and strategic rivalries of later years. That does not mean the conditions of 1996 can simply be recreated. They cannot. Nor does it mean that every subsequent development in information technology has validated every assumption underlying the original agreement. What history does demonstrate is that international economic cooperation remains possible when the participants can identify a sufficiently clear mutual interest.
The ITA further illustrates the importance of persistence and continuity. The Singapore Ministerial wasn’t the end of a process. Agreement on the basic framework was followed by the difficult work of securing the participation and domestic implementation necessary to make the agreement effective. This lesson is important: an international agreement is not finished when the negotiating language is completed. It is really never done.
The subsequent history of the ITA gives the story an additional dimension. Information technology changed dramatically after 1996, while the agreement continued to provide a framework for eliminating tariffs on an expanding range of products and for bringing additional economies into that framework. At the same time, the use of tariffs as an instrument of trade and economic policy in more recent years has raised questions about how durable the assumptions underlying the ITA remain.
Those developments do not erase what the ITA accomplished, but they do make its underlying idea worth reconsidering. The central question is no longer simply whether tariffs on computers, semiconductors, and related products should be eliminated. It is whether the international trading system can develop a comparable forum of cooperation for technologies that are now becoming fundamental to virtually every sector of the economy. Artificial intelligence presents that challenge in an especially consequential form because its development depends upon a global network of hardware, software, data, computing capacity, services, research, and investment.
So we conclude here with a much larger proposition. An Artificial Intelligence Trade Agreement (AITA) should explore the possibility of building a structure to address the scope, participants, obligations, and relationships required to address the needs of AI users, developers, and host countries. The ITA suggests an historical example of how such a process might begin. Identify the barriers that unnecessarily impede the development and international movement of AI technologies and the areas in which countries have interests sufficiently aligned to make cooperation worthwhile. Then bring together governments, industry, technical experts, and other affected participants to determine whether those interests can be translated into practical commitments. Finally, attempt to get widespread agreement on trade-related artificial intelligence issues. The history of the ITA cannot tell us whether an AITA is possible. It can, however, start us down a path where we ask –
Can the ITA provide a remarkably relevant example of how the search for answers in an AI-based world might begin?