The New START nuclear arms control treaty expires this month, and there is no indication that the agreement will be extended, after being extended for 5 years on February 5, 2021. The 10-year New START treaty was signed in April 2010 by Presidents Barack Obama and Dmitry Medvedev , replacing START I of 1991, which expired in December 2009, and the Moscow Treaty of 2002. Under the terms of the treaty, the U.S. and Russia must limit their strategic nuclear forces to a combined total of
- 700 missiles including land-based intercontinental ballistic missiles (ICBMs) and submarine-based intercontinental ballistic missiles (SLBMs) and heavy nuclear-armed bombers
- 1,550 nuclear warheads on ICBMs, SLBMs, and heavy bombers (each bomber is considered to have one warhead regardless of how many it can carry)
- 800 ICBMs, SLBMs and heavy bombers equipped, of which 700 are operational.
New START places no constraints on missile defenses, conventional long-range weapons (including on ICBMs, SLBMs, and bombers), and non-strategic (“tactical”) nuclear weapons. The treaty imposes no restrictions on the modernization of ICBMs, SLBMs, or heavy bombers as long as the aggregate numbers remain within the limits defined by the treaty. Finally, both countries have considerable autonomy in determining the actual mix of delivery systems. A key feature of New START is the comprehensive provisions to ensure that each party can verify that the other party is behaving in compliance with the treaty. The treaty requires the exchange of data on the numbers, types , and locations of treaty-limited items; notification of changes to this data; the application of unique identifiers on ICBMs, SLBMs, and heavy bombers; and invasive inspections to confirm the accuracy of all such information.
The state of the art of nuclear weapons
According to the SIPRI report of June 2025, we are currently witnessing a growing risk linked to a nuclear arms race. In summary, “ Almost all nine nuclear-weapon states—the United States, Russia, the United Kingdom, France, China, India, Pakistan, the Democratic People’s Republic of Korea (North Korea), and Israel—continued intensive nuclear modernization programs in 2024, upgrading existing weapons and adding newer versions.
Of the estimated total global inventory of approximately 12,241 warheads as of January 2025, approximately 9,614 were in military stockpiles for potential use (see table below). An estimated 3,912 of these warheads were deployed on missiles and aircraft, and the remainder were in central storage. Approximately 2,100 of the deployed warheads were maintained on high operational alert on ballistic missiles. Nearly all of these warheads belonged to Russia or the United States, but China may now maintain some warheads on missiles during peacetime. (…) Russia and the United States together possess approximately 90 percent of all nuclear weapons. The size of their respective military stockpiles (i.e., usable warheads) appears to have remained relatively stable in 2024, but both states are implementing extensive modernization programs that could increase the size and diversity of their arsenals in the future . If no new agreement is reached to limit their stockpiles, the number of warheads deployed on strategic missiles appears likely to increase after the 2010 bilateral Treaty on Measures for the Further Reduction and Limitation of Strategic Offensive Arms ( New START ) expires in February 2026.
The modernization of the arsenals obviously represents a significant effort from a technological and financial point of view. However, both Russian and US nuclear weapons deployments are likely to increase in the coming years. The Russian increase would primarily result from the modernization of existing strategic forces to carry more warheads on each missile and the reloading of some previously depleted silos. The US increase could occur through the deployment of more warheads to existing launchers, the reactivation of empty launchers, and the addition of new non-strategic nuclear weapons to the arsenal. Nuclear advocates in the United States are pushing for these steps in response to China’s new nuclear deployments. SIPRI estimates that China now has at least 600 nuclear warheads. China’s nuclear arsenal is growing faster than any other country, with around 100 new warheads per year from 2023. By January 2025, China had completed or was close to completing around 350 new ICBM silos in three large desert fields in the north of the country and in three mountainous areas to the east. Depending on how it chooses to structure its forces, China could potentially have at least as many ICBMs as Russia or the United States by the turn of the decade . Yet even if China reached its projected maximum number of 1,500 warheads by 2035, that would still only represent about a third of each of the current Russian and U.S. nuclear arsenals.”
We are witnessing a constant expansion and modernization of nuclear arsenals, as the last treaty for their control expires. The context in which this is occurring explains, on the one hand, the reluctance of the parties to renew the treaty, which was already laboriously renewed in 2021; since then, conflict in global relations has grown exponentially, and in this context, the danger of the lack of any means of limitation and mutual control—even between the two major nuclear powers—is clearly evident. Moreover, there is no shortage of conflicts involving nuclear powers, such as between India and Pakistan and between India and China. Iran’s development of nuclear weapons is a central, though not the only, issue in relations between Iran, the United States, and Israel, within the context of the strategic balance of power in the Middle East and the Persian Gulf.
However, in the context of looking to the future of the nuclear issue, the role of technological-digital innovation driven by the Artificial Intelligence ecosystem is central, which in recent years – thanks also to field experiments made possible by ongoing conflicts – has affected the dimension of warfare at all levels. The Japanese Toda Peace Institute 3 addresses this issue in briefing on the future of nuclear warfare. “A significant concern is that technologies such as AI, cyber capabilities, and space systems interact across domains, creating cumulative effects that cannot be assessed in isolation from one another. New capabilities could threaten both the security and reliability of nuclear forces. Advances in detection, tracking, and remote surveillance expose assets once thought to be hidden, while missile defenses, hypersonic weapons, and directed-energy weapons may, under certain conditions, erode confidence in second-strike stability. (…)
Disruptive emerging technologies also pose new challenges for arms control, as their dual-use convergence and intangible nature make it more difficult to distinguish civilian from military applications. Commercial companies now drive much of the relevant innovation, lowering barriers to entry and reshaping who should be involved in arms control discussions. The presentation suggested that the most practical approach might be to verify observable behaviors and effects rather than controlling broad R&D activities, and to rely more heavily on regional or issue-specific “clubs” and cartels that control access to key resources such as chips, computing power, or space launch services.
Governments alone will struggle to keep pace, which means the private sector will have to become an active participant in regulation and verification—not just an actor to be regulated. This raises the question of how to align incentives for private companies to contribute to arms control objectives.” In 2024, SIPRI released a report, Nuclear Weapons and Artificial Intelligence: Technological Promises and Practical Realities, which cannot be summarized but should be considered a reference text, from which it is worth extracting some information and assessments. The Introduction states, “Notably, all nine nuclear-weapon states—China, France, Israel, India, the Democratic People’s Republic of Korea (DPRK, or North Korea), Pakistan, the Russian Federation, the United Kingdom, and the United States—have shown interest in developing and integrating advanced AI capabilities into their militaries, with some explicitly making AI a strategic priority. While traditional AI systems have long been part of the nuclear weapons sector, the potential use of advanced AI in early warning and ISR, nuclear command, control and communications (NC3), delivery systems and conventional systems, with counterforce capabilities and the ability to strike military targets. (…)
Given the culture of secrecy surrounding nuclear programs, information on plans, strategies, or practices related to further AI-nuclear integration is severely limited. However, the discussion of AI in the broader defense context can provide some clues as to where states see the value of these systems, including for the nuclear domain. It also suggests implications for nuclear deterrence. In the nuclear context, machine learning is often discussed as potentially enhancing capabilities across all elements of the nuclear deterrence architecture: early warning and ISR; command and control; targeting, guidance, and navigation of nuclear weapons; and non-nuclear operations such as missile defense, cybersecurity, and anti-submarine warfare.” The document analyzes the paths of various countries, from India to the USA, Russia, and China, in the use of AI in the military field, specifically in the nuclear field. It is worth noting that “It would be inaccurate to suggest that any one of these elements constitutes a fully operational advanced AI capability, given their ongoing development and the rather narrow tasks assigned to the latest ML systems. However, the fact that some capabilities are already being tested suggests clear trends toward broader integration.” To remain with China, “For example, according to some assessments, there is a consensus in the PLA (People’s Liberation Army) media that generative AI has a role in warfare, human-machine interaction, decision-making, network warfare, logistics, the cognitive domain, the space domain, and training.”
Obviously, one of the challenges that must be addressed for full deployment of the Foundation Models and Transformer technologies is reliability, the risk of so-called “hallucinations,” which would be fatal in conflict management. “The expectations placed on AI are high for certain tasks—and perhaps unrealistic, at least in the near future. While AI, in some respects, offers capabilities that surpass those of humans, there are still barriers. Among these notable barriers are the limitations of AI itself and the amount of data available, both for training and in practice. AI is expected to be more precise in analyzing data than human-led efforts in this area or to be able to develop superior tactics. However, this is a question of access: to data, systems, and equipment.”
The starting point is that the development of technologies and equipment for military use is largely in the hands of US Big Tech companies, whose interaction and integration with the US military and security apparatus, as well as other countries, is increasingly tight. The one-and-a-half-fold increase in the defense budget requested by the Trump administration opens up immense opportunities for the integration of AI into military apparatus and strategies. Other competing countries, primarily China, cannot be outdone, although the resources allocated are not on the same scale. However, as the development of original AI models like Deep Seek demonstrates, the ability to innovate can at least partially compensate for inferiority in terms of financial and structural resources. China has far fewer data centers, but many more engineers graduate from universities each year. In essence, “Building a secure and effective system of AI data centers to process training data requires additional infrastructure for cloud computing and low-latency data transfer. Considering all of the above, advanced AI integration may simply not be economically viable for some nuclear-armed states or may not contain sufficient operational value to justify its career in the nuclear field.” Control over information flows about technologies and the sharing of knowledge and infrastructure becomes crucial, becoming a determining factor in establishing state autonomy and defining alliances. The role of Big Tech becomes fundamental, variously connected to the strategies of states and governments, however, as has been said, they are an integral part of any arms control strategy, both nuclear and conventional.
Field experiences, for now in conventional terms, are fundamental in the testing of new weapons systems, such as drones which play such a large part in the conduct of current conflicts, and in the collection of information and data necessary to feed AI-based forecasting and governance models.
As we have already had occasion to point out, “The problem is that in the world there are probably only two databases of millions of videos on which it is possible to train artificial intelligence for new weapons systems: the Ukrainian one and the one that, in all likelihood, the Russians have accumulated. The latter is undoubtedly at the disposal of China, which is thus refining and improving its warfare technologies by studying Ukrainian drones that have fallen in territories controlled by Moscow”. See also the Swift Beat project linked to the acquisition of that data for the production of drones in Ukraine by Eric Schmidt, former CEO of Google. The Swift project is based on interceptor drones and sources close to the project have revealed that the effectiveness demonstrated in the initial tests is at excellent levels: approximately nine out of ten Russian Shaheds have been shot down by Ukrainian forces using Swift Beat drones . If deployed in massive numbers, they could therefore build a sort of real anti-drone shield for Ukrainian cities. Schmidt was one of the first major international investors to seriously invest in the Ukrainian military-tech sector, having previously financed the D3 Fund , an accelerator for Ukrainian startups specializing in the defense technology sector, with $10 million. However, the most malicious rumors claim that the Swift Beat operation is a sort of field test of the most advanced military technologies for the former Google CEO, a fine-tuning to then market these devices on the global market at higher prices.
For further insight into the role of technological innovation in the conduct of war, we can read contribution, which analyzes in a lengthy article the ways in which different countries can achieve technological innovation that also operates in the military field. “The growing strategic rivalries and competition for future supremacy between the United States, Russia, and China influence different national responses to the same technological breakthroughs, conditioned by different defense innovation trajectories, priorities, and resources. To project the different trajectories, it is necessary to conceptualize a comparative framework for defense innovation that integrates the different phases, paths, and patterns. To begin with, conceptualizing emerging technologies into military capabilities involves internal processes of military innovation as well as external processes of adaptation or emulation of benchmarking. Disruptive military innovation may not require simultaneous technological, doctrinal, and organizational breakthroughs, but can span the spectrum between incremental modernization and discontinuous transformation. Based on these assumptions, defense innovation trajectories can be triangulated along three axes:
- Conceptual paths—emulation, adaptation, and innovation;
- Technological models—speculation, experimentation and implementation;
- Organizational change—exploration, modernization, and transformation.”
(…) In this context, one of the key sets of variables in the matrix is the level and sophistication of a country’s defense-innovation ecosystem, which can be defined by a range of “hard” and “soft” innovation capabilities: from research and technological development (R&D) structures and innovation clusters to non-technological factors such as political, institutional, relational, social, and ideation factors. Together, these factors determine the relative level of states’ indigenous capabilities for independent defense-related research and development, science and technology (S&T) programs, manufacturing, and communities that support innovation. We are witnessing the exaltation of the ‘dual use’ of technologies, of the integration, and of the total reversibility of the ‘civil’ and ‘military’ uses of technologies. As we well know, this integration is the opposite of the action that directs the technological-digital transition towards the energy-climate transition and, at the very least, towards the reduction of social inequalities.
It goes without saying that the use of digital technologies and AI, particularly in the military, is parallel and mutually reinforcing, with the use of these technologies in the field of social control, surveillance, and the deep and widespread control of individual and collective behavior. The role of the Palantir and Anduril companies in managing information flows on the battlefield and in society is well-known and exemplifies current trends.
Israel’s use of AI systems has been widely discussed and pervades its entire military apparatus, the conduct of warfare at various levels, on various terrains, in the most diverse ways, and its widespread social control in the territories of Gaza and the West Bank.
In conclusion, the end of the New START treaty occurs amidst an extraordinary technological transformation of the military establishment as a whole, inextricably linked to the oligopolies and research centers that spearheaded this innovation. It is the new version of the military-industrial apparatus once denounced by President Eisenhower, representing a quantum leap in power, pervasiveness, and complexity compared to what was essentially the product of the Fordist industrial model, which had seen the quantum leap of Big Science with the Manhattan Project.
The fusion of digital technologies, life, materials, and everything else unites all areas of political and social conflict, as we have already said, in that global process defined as polycrisis, in which movements and demands against war and against the authoritarian drift of all political systems, starting with those still defined as democratic, constitute an unavoidable step in any liberation movement against the current state of affairs.
