Publications

The National Committee for Accelerating the Field of Artificial Intelligence, headed by Prof. Yaakov Nagel, recently completed its work and published its findings. As its name suggests, the committee was established to examine how the development of the artificial intelligence (AI) field could be accelerated by creating a body within the Prime Minister’s Office, as well as to recommend a strategy, an initial work plan, and the mechanisms and resources required for this body to function.

AI may still be a relatively new field, but the Nagel Committee is already the third committee established by the Israeli government to examine ways to advance it, and the second committee whose recommendations have been adopted, in whole or in part. It was preceded by the National Initiative for Secured Intelligent Systems, chaired by Prof. Isaac Ben-Israel and Prof. Eviatar Matania, established by Israel’s 34th government under Prime Minister Benjamin Netanyahu, and by the Committee for Artificial Intelligence and Data Science, chaired by Dr. Orna Berry, who was appointed by Prof. Shimon Ullman, head of the TELEM Forum (the National Infrastructure Forum for Research and Development), and whose recommendations formed the basis for Decision 212 of Israel’s 36th government under Prime Minister Naftali Bennett. This decision established the National Artificial Intelligence Program with a budget of approximately one billion shekels to be allocated between 2021 and 2027, compared to the five-billion-shekel budget recommended by the Berry Committee.^[i] In the past two years, the suitability of the National Artificial Intelligence Program to address the challenges and its success have been the subject of intense public debate. On the one hand, the government (through TELEM, which has been managing the program) insists that the program is adequate and meets its goals. On the other hand, the State Comptroller and most practitioners in the field have argued quite the opposite—that the national program fails to provide sufficient solutions, and as a result, Israel’s position in the field has deteriorated.

Against this backdrop, the Nagel Committee made a significant contribution to the public debate. Most importantly, it “reached an unequivocal and severe conclusion: the State of Israel is not at the appropriate and desired point for acceleration in the field of artificial intelligence.”^[ii] This constitutes the first official acknowledgment of the failure of the current national program.

The Nagel Committee deviated from its mandate and recommended that the State of Israel aspire to achieve significant global leadership in the field of AI, as measured by various international indices, including Tortoise Media’s “Global AI Index” and Oxford Insights’ “Government AI Readiness Index.” The committee set a goal for Israel to be among the top five countries vis-à-vis implementation of AI, as tracked by the “Global AI Index.” In the committee members’ view, “those who lead the AI race will determine the geopolitical and economic structure of the near future.”^[iii] The committee also emphasized that “the world is engaged in an accelerated race for technological leadership in artificial intelligence, with the United States and China at the forefront, followed by the European Union and several other Western countries.”^[iv] Therefore, in the eyes of the committee members, the unequivocal and obvious conclusion—consistent with Israel’s long-standing security doctrine that emphasizes maintaining a qualitative edge—is that “leadership in artificial intelligence is not a strategic option but an existential necessity.”^[v]

Alongside its portrayal of AI as a race that must be “won” or at least “led,” the Nagel Committee also presents a national vision whereby “artificial intelligence will serve as a driving force for economic growth, improved efficiency in the public sector, and strengthened security and welfare of the citizens. Israel will leverage its assets—its high-quality human capital, entrepreneurial spirit, academic knowledge base, and technological infrastructure—to be among the nations shaping the future of AI in the world.”^[vi]

In addition, the Nagel Committee proposed seven strategic goals and seven measurable objectives for evaluating the success of the proposed strategy. Three of the strategic goals and two of the measurable indicators concern the achievement of status symbols in the global AI competition—specifically, ranking among the top five countries according to Tortoise Media’s “Global AI Index,” climbing approximately ten places on Oxford Insights’ “Government AI Readiness Index,” and establishing a state-owned supercomputer containing approximately 60,000 advanced GPU processors.

It should be noted that most of the Nagel Committee’s recommendations are not new, including the aspiration to reach the top five leading countries and the establishment of a supercomputer that would rank among the world’s 20 most powerful supercomputers. However, the committee’s updated and expanded national vision requires a reexamination of whether these goals truly align with that vision or whether they create a negative incentive that could hinder the realization of the vision by misallocating resources.

When examining the alignment between the goal of “leadership in the artificial intelligence race” and the proposed vision, the first question to ask is whether placement in international rankings is indeed the only—or even the most appropriate—measure of success. Specifically, one must assess the significance of ranking fourth or fifth globally and whether this declared target genuinely serves the vision of AI as a driver of economic growth, improves efficiency in the public sector, strengthens national security, or enhances the welfare of Israel’s citizens. In particular, it is necessary to determine whether there exists a contingency in which the expected value of the investment exceeds its cost.

At the heart of the recommendation to aim for a global ranking among the top five countries according to various indices lies the observation that Israel’s standing has declined in recent years (see Figure 1). Let us first examine this claim more closely.

Figure 1. Israel’s Rankings in Leading International AI Indices, 2019–2024

When reviewing Tortoise Media’s “Global AI Index,” one can identify a consistent downward trend in Israel’s ranking. However, an examination of Oxford Insights’ “Government AI Readiness Index” reveals a more complex picture: Israel’s ranking remained stable between 2019 and 2022, declined in 2023, and then improved significantly in 2024 (it appears that this index was not presented to the committee). Similarly, Stanford University’s Global AI Vibrancy Tool—which, alongside the reports of Tortoise and Oxford Insights, is considered one of the three most significant and comprehensive in the field—shows a similar pattern: Israel’s ranking was relatively stable from 2020 to 2023, and then improved markedly in 2024. This nuanced trend calls for additional analysis, particularly since, according to the most recent Oxford Insights ranking, Israel has already reached and even surpassed the strategic target proposed by the Nagel Committee, all before the committee completed its work and published its conclusions.

The different indices are composed of a wide range of indicators and sub-indicators based on the original research by the index authors, quantitative data from third-party sources, and occasionally, other indices. For example, all three studies rely on World Bank data to assess internet penetration and electricity infrastructure in different countries. Similarly, they all draw on the Top500 list to evaluate the availability of computing capacity, such as supercomputers. However, all three indices rely on independent academic judgment when assessing policy-related issues. In each ranking, the score for every indicator is the average of its sub-indicators, and the final ranking is based on the average of all indicator scores. The differences among the indices stem from variations in sub-indicator selection, source data, normalization methods (if any), and the method of averaging (some use a weighted average calculation).

A close examination of the sub-indicators that make up each index reveals a plausible explanation for Israel’s ranking trends. For example, the Tortoise Media index is the only one to use a weighted average without normalizing sub-indicator scores but only normalizes the final composite score between a 0 and 100 range (rather than applying a simple average and normalizing the scores at the sub-indicator level). This index also uses its weighting method to strongly emphasize purely technical factors, such as the number of AI models developed, the number of LinkedIn profiles from the country declaring AI expertise, and even the popularity of AI-related projects on code-sharing platforms such as GitHub and Hugging Face. Moreover, the Tortoise Media ranking does not account for the population size or economic scale of the ranked countries. Consequently, given Israel’s small size, its position will continue to decline unless resources are invested on a scale that is disproportionate to its size.^[vii]

Furthermore, when examining Israel’s position across the three indices solely based on sub-indicators taken from external sources, a pattern emerges showing stability in the Stanford and Oxford Insights rankings (without improvement) and a gradual decline in Tortoise. However, when focusing on the original research-derived sub-indicators, a different picture emerges. In the 2024 rankings, Israel received significantly higher scores on the Oxford Insights and Stanford indices compared to the Tortoise Media index. This is because the first two mistakenly interpreted a publication by Israel’s Ministry of Innovation, Science, and Technology, and the Ministry of Justice—titled “Principles of Policy, Regulation, and Ethics in the Field of Artificial Intelligence”^[viii]—as a formal national AI policy officially adopted by the State of Israel. This mistake substantially boosted Israel’s score in those indices, whereas Tortoise Media’s “Global AI Index” did not err in the same way and did not count it as a national policy (and, in any case, it assigned less weight to policy documents). Our analysis suggests that this error fully explains the apparent improvement in Israel’s ranking in these indices, revealing that the rise was, in fact, superficial.

The most important conclusion emerging from this examination of the significance of international rankings is that it becomes clear these rankings can be easily influenced, and thus a country’s position can be artificially improved. Another example of the potential bias in these rankings is their reliance on social networks and code-sharing platforms based on popularity metrics (which are known to be particularly easy to manipulate). Therefore, one cannot escape the conclusion that a country’s ranking can, in effect, be engineered artificially, such as by the publication of government policy documents that may never actually be implemented or through influence campaigns on social media—all without any tangible benefit to Israeli citizens. The opposite is also true—improvements in the well-being of Israeli citizens will not necessarily enhance Israel’s ranking in these indices. For example, in our estimation, even the full achievement of targets, such as increasing the number of government data repositories or completing flagship government projects, would have no effect on Israel’s position in the Tortoise Media or Oxford Insights indices—despite their clear potential to improve the welfare of Israel’s citizens. As a result, focusing primarily on improving Israel’s standing in these indices could have a destructive effect by creating incentives to invest resources in activities that might raise the ranking (even by artificial means) but do not improve citizens’ lives.

The Outline of a National Computing Infrastructure

Another central goal recommended by the Nagel Committee, and the one carrying the majority of the financial burden of its recommendations, is the establishment of a national supercomputer: a system owned by the State of Israel, based on approximately 60,000 AI-optimized processing units by the end of 2029. The supercomputer is intended for the use of at least 50 research groups and 100 companies, and its estimated cost is about 18 billion shekels—comprising the vast majority of the 25 billion shekels the committee recommends allocating.^[ix] The astronomical cost, together with the bitter experience and difficulties encountered during the tender for constructing the current national computing system, has led many to question the state’s economic and managerial capacity to implement this seemingly groundbreaking recommendation. Above all, the sheer cost forces us to ask whether this is truly an essential requirement for realizing the national vision or merely a status symbol in the race for global prestige.

Let us first examine whether this recommendation is truly unprecedented. The prevailing view holds that the field of AI requires substantial computing infrastructure; accordingly, all previous committees that examined the field have recommended establishing dedicated infrastructure. The Ben-Israel–Matania Committee recommended purchasing a supercomputer that could also serve for training and using AI models. Its performance would rank around the 20th place on the Top500 list of supercomputers, and 20% of its components could be upgraded each year.^[x] According to that committee’s report, the cost of establishing and maintaining the system was expected to be around a total of one billion shekels for the period 2021–2025.^[xi] The Berry Committee presented a similar recommendation to the TELEM Forum but estimated the establishment and maintenance costs for 2021–2025 at approximately 2.2 billion shekels.^[xii] A careful examination of the Nagel Committee’s recommendation in light of current trends in the supercomputing market shows that a computer with approximately 60,000 processing units would rank roughly 20th on the leading lists of AI-dedicated supercomputers in 2027.^[xiii]

Given that prevailing trends indicate a doubling of hardware costs for model training every nine months,^[xiv] the estimated 18-billion-shekel cost for building and maintaining such a system over five years is broadly consistent with the conclusions and recommendations of both the Ben-Israel–Matania and Berry Committees. It should be emphasized, however, that if current growth trends in computational demand and hardware costs continue, the proposed 18-billion-shekel budget will not be sufficient to maintain a 20th-place ranking beyond 2028. It is also worth recalling that the Institute for National Security Studies (INSS) previously recommended establishing a supercomputer with at least 40,000 processing units to meet a similar target by 2027.^[xv] Therefore, although the figures (60,000 processors and 18 billion shekels) may appear striking or even intimidating, the Nagel Committee’s recommendation is neither unprecedented nor new.

As noted, the Nagel Committee’s recommendations are consistent with those of earlier committees and with research published by INSS. Yet, it is necessary to assess whether this represents an essential infrastructure for advancing the proposed national vision or a status symbol in the global prestige race. It is well known that infrastructure investment is one of the conventional methods for generating economic growth. However, beyond the infrastructure itself, the other components of the proposed vision primarily require computing resources for the inference stage of AI models, which is characterized by a much lower computational demand than the model training phase. An examination of the AI field over the past decade reveals that large-scale models, particularly frontier and foundation models, have been developed almost exclusively by the private sector or, at least, in collaboration with it. Governments worldwide generally do not attempt to run such massive projects independently, and certainly not under direct state management. Even in China, which competes with the United States for global leadership in the field and operates under a highly centralized system, the government conducts its national AI efforts through commercial partnerships, albeit they are ultimately subject to state control.

A review of the models published during the current decade shows that academic institutions have produced two frontier models, two large models, and about five foundation models, while during the same period, dozens of such models have been developed in collaboration with industry.^[xvi] Since academia’s strength, particularly in recent years, does not lie in developing large frontier and foundation models, in contrast to industry, which excels in this area,^[xvii] allocating massive resources to enable academia to develop such models at the expense of the private sector would be misguided. By contrast, there is no doubt that Israel’s universities currently face a significant computational capacity gap. Therefore, the establishment of a National Academic Center for Artificial Intelligence, as proposed by the Committee of University Heads, represents a significant and necessary step toward strengthening Israel’s position and fulfilling the Nagel Committee’s vision.^[xviii]

In contrast, the Nagel Committee’s recommendation to establish a national supercomputer under direct government ownership runs counter to global trends and raises concerns that the committee assessed the need primarily through the lens of a status symbol rather than practical feasibility. It is worth asking whether the government is indeed the right body to develop an infrastructure that demands rapid adaptability and constant alignment with market needs. Most experts would argue that it is not—governments generally struggle to allocate resources efficiently, especially when dealing with highly competitive domains where there are strong market incentives for innovation and continual optimization. Accordingly, it can be assumed that constructing a 60,000-GPU supercomputer for AI applications, managed directly by the Israeli government, would not be an efficient solution and would not optimally advance the committee’s vision. The Nagel Committee justifies the need for state ownership by citing operational continuity and protection of sensitive data.^[xix] However, even this rationale does not fully support the demand for a government-managed AI supercomputer—particularly since the committee itself does not envision an Israeli autarky.^[xx]

There is no doubt that the Nagel Committee is correct in asserting that a national high-performance computing (HPC) infrastructure is necessary to support AI development. Yet, in our assessment, building a single, centralized supercomputer would not best serve the committee’s broader vision. Rather, it would primarily provide Israel with a status symbol and a modest boost in global rankings. Still, the need for robust computing infrastructure, operational continuity, and the safeguarding of sensitive information cannot be ignored. The solution to these needs lies in recognizing that the relevant measurable indicator for evaluating Israel’s position should be the total computing power available within its market. Unfortunately, the indices on which the Nagel Committee focuses do not currently take this factor into account. Israel’s infrastructure strategy should thus aim to achieve this cumulative computing capacity by the most efficient means possible. This situation may be compared to the energy sector: Israel’s economy requires a certain level of electricity supply, but it is self-evident that the government would not invest all its resources in a single massive power plant. Instead, production capacity is distributed among multiple power stations, employing diverse energy technologies. Similarly, the key to achieving sufficient national computing capacity requires abandoning the idea of a single, centrally managed infrastructure in favor of a diversified, multi-layered approach.

For academic research needs, it would be appropriate to continue advancing the establishment of a National Academic Center for Artificial Intelligence, to be operated by the universities themselves, as is already being discussed. Any remaining capacity gaps could be bridged through vouchers—state-funded computing credits allocated to universities from the national budget. For public-sector needs, particularly those involving sensitive citizen data, it would indeed be sensible to establish a dedicated government computing center; however, since such a center would mainly be used for running pre-existing models rather than training new ones, its computational requirements could be far more modest.

Alongside these measures, to ensure business continuity for Israel’s private sector, it would be wise to adopt the emerging concept of “sovereign compute infrastructure,” which has recently gained traction across Europe and parts of Asia. Under this model, governments encourage major American tech companies to form partnerships with local firms to establish AI-focused data centers within national borders. The advantage of this model is twofold: It provides a high degree of domestic resilience through locally based infrastructure while retaining the flexibility and efficiency of market-driven solutions. This approach also facilitates the transfer of missing know-how, ensures continuity of service through physical location within the state’s sovereign territory, and maintains the presence of local technical professionals who can continue operating the centers if the foreign partner withdraws for commercial or geopolitical reasons.^[xxi] Adopting this approach would not only enable Israel to achieve the minimum necessary level of independent computing capacity for national resilience but could also improve its standing in international rankings. This is because distributing investments among multiple private entities would allow the aggregate number of processors located in Israel to surpass what the government alone could finance and manage directly.

Conclusion

The Nagel Committee Report is significant because, for the first time, a major body within the Israeli government explicitly recognizes that the country’s current standing in the field of AI is inadequate and engages in a serious discussion of how to improve it. The committee’s worldview is based on the notion that there is a global race in the field of AI and that achieving a higher ranking in this race would have substantial implications for Israel’s national security. At the same time, the committee identifies what it perceives as a downward trend in Israel’s international rankings and views this as evidence supporting its conclusions about Israel’s position. Accordingly, the committee proposes an expansive vision for Israel’s AI sector and a set of strategic goals; foremost among them is climbing into the global Top 5 in the Tortoise Media “Global AI Index” and establishing a national supercomputer with approximately 60,000 processors within the next five years.

In this article, we conducted an in-depth examination of the indices that formed the basis of the committee’s report, showing that the situation is considerably more complex. The main findings presented here indicate that there is not necessarily a positive correlation between improvements in international rankings and actual economic growth, citizen welfare, or national security. Worse still, actions in the field of AI that could genuinely contribute to economic growth or improve the welfare of Israeli citizens are often those that have no direct impact on Israel’s rankings. As a result, relying on the indices ranking for measuring success may create negative incentives, leading to the allocation of resources toward activities that improve rankings but not the lives of Israeli citizens.

Our examination of the recommendation to establish a national supercomputer yields a similar conclusion. The proposal is impractical for several reasons. First and foremost, it is doubtful that the state currently possesses the financial capacity to fund such a project. More troubling is that, even in cases where much smaller budgets were allocated, it took years for the computing capacity to be realized. In that period, the amount of processing power that could have been purchased for the same sum had already become insufficient, given the exponential growth in computational demand. In the authors’ view, the Nagel Committee’s recommendation is inefficient: constructing a large, centralized, government-run computing system is not the optimal solution. As argued above, there are more cost-effective approaches that could deliver greater benefits, such as promoting economic growth, improving citizens’ welfare, and even modestly enhancing Israel’s position in global rankings.

As Shimon Peres famously said, “Polls are like perfume—you can smell them, but you shouldn’t drink them.” Paraphrasing his words, AI indices are like polls; they are a useful tool, but they should never become an end in themselves. AI indeed holds tremendous potential to contribute to the qualitative edge that is vital to Israel’s national security; but to succeed, the approach must be grounded in a broad, forward-looking vision, free from fixation on competition and rankings. Such metrics can help identify actions that have succeeded elsewhere, but the ranking itself must never be the goal.

 _________________________

[i] “A Program for Promoting Innovation, Encouraging Growth of the High-Tech Sector, and Strengthening Technological and Scientific Leadership,” [Hebrew] Decision 212, Government of Israel 36 (2021), https://www.gov.il/he/pages/most_policy20210801.

[ii] Yaakov Nagel et al., Report of the National Committee for Accelerating the Field of Artificial Intelligence – August 2025, [Hebrew] National Committee for Accelerating the Field of Artificial Intelligence, 2025, 4, https://www.gov.il/he/pages/event-ai050825.

[iii] Nagel et al., Report of the National Committee, 2.

[iv] Nagel et al., Report of the National Committee, 18.

[v] Nagel et al., Report of the National Committee, 11.

[vi] Nagel et al., Report of the National Committee, 65.

[vii] “The Global Artificial Intelligence Index Methodology Report,” Tortoise, September 2024, tortoisemedia.com/_app/immutable/assets/AI-Methodology-2409.BGTLUPC-.pdf.

[viii] Ministry of Innovation, Science, and Technology and the Ministry of Justice, Policy, Regulation, and Ethics Principles in the Field of Artificial Intelligence, [Hebrew] December 14, 2023,

https://www.gov.il/BlobFolder/policy/ai_23/he/2023%20Artificial%20Intelligence%20Regulation%20and%20Ethics%20Policy%20Principles%20Document.pdf.

[ix] Nagel et al., Report of the National Committee, 65–69.

[x] Isaac Ben-Israel et al., eds. The National Initiative for Smart and Secure Systems for Strengthening National Security and Scientific-Technological Resilience: A National Strategy for Israel. Special Report to the Prime Minister—Part B: Subcommittee Reports, [Hebrew] vol. 2 (Tel Aviv University, Yuval Ne’eman Workshop for Science, Technology, and Security, 2020), 104–105.

[xi] Ben-Israel et al., The National Initiative for Smart and Secure Systems, vol. 2, 90.

[xii] Artificial Intelligence and Data Science Committee, [Hebrew], December 2020, 83–84, 86–87, https://tinyurl.com/bdfub2f7.

[xiii] Lennart Heim et al., “Data on GPU Clusters,” Epoch.ai, April 2025, https://epoch.ai/data/gpu-clusters.

[xiv] Epoch AI, “Key Trends and Figures in Machine Learning,” 2023, https://epoch.ai/trends.

[xv] Ariel Sobelman and Michael Genkin, “The Artificial Intelligence Challenge: Israel’s Decline and the Urgent Need for a New National Strategy,” [Hebrew] INSS Policy Paper, 2025, 8, https://www.inss.org.il/he/publication/ai-challenge/.

[xvi] Epoch AI, “Data on AI Models,” October 8, 2025, https://epoch.ai/data/ai-models.

[xvii] Nestor Maslej et al., “The AI Index 2025 Annual Report,” AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, 2025, 457, https://hai.stanford.edu/assets/files/hai_ai_index_report_2025.pdf; Epoch AI, “Data on AI Models.”

[xviii] Sagi Cohen, “The Universities Promote the Establishment of an Artificial Intelligence Computing Center at a Cost of Half a Billion Dollars,” [Hebrew] TheMarker, August 27, 2025, https://www.themarker.com/technation/2025-08-27/ty-article/.highlight/00000198-e67f-d91d-af9b-e77f64ed0000.

[xix] Nagel et al., Report of the National Committee, 46–49.

[xx] Nagel et al., Report of the National Committee, 3.

[xxi] Georgia Butler, “AWS Establishes European Sovereign Cloud as Separate Company,” Data Center Dynamics, June 5, 2025, https://www.datacenterdynamics.com/en/news/aws-establishes-european-sovereign-cloud-as-separate-company/; Sophie Rice, “Microsoft Expands European Data Centre Footprint,” Data Center Magazine, May 1, 2025, https://datacentremagazine.com/technology-and-ai/microsoft-expands-european-data-centre-footprint; Dan Swinhoe, “Microsoft and G42 to Set up Sovereign Cloud in UAE,” Data Center Dynamics, September 6, 2023, https://www.datacenterdynamics.com/en/news/microsoft-and-g42-to-set-up-sovereign-cloud-in-uae/.