The Israel Innovation Authority in collaboration with the National AI Directorate has asked the tech community for feedback on its National AI Strategy for Israeli high tech. Perhaps most importantly, Israel is not trying to outspend the United States, outscale China, or train the world’s largest foundation models. It’s not delusional on a call to build a national version of OpenAI. The strategy is more mature understanding of the global AI race.
Instead, the document argues that Israel should focus on the parts of the AI stack where speed, technical depth, domain expertise and entrepreneurial density matter more than sheer scale. The phrase that runs through the strategy is simple: Israel should lead where “excellence outweighs scale.”
That is a meaningful shift. In the early phase of generative AI, the race was dominated by foundation models, massive GPU clusters, proprietary datasets and unprecedented capital expenditure. This phase naturally favoured the US hyperscalers, frontier labs and China’s state-backed technology ecosystem. Israel participated as a sophisticated user and builder, but it was never structurally advantaged in a contest defined by the size of compute budgets. While AI21 Labs built its own foundational models, it struggled to gain market share and compete on the global AI race.
The new strategy effectively accepts that reality and moves the conversation to a more promising question: where can Israel build global companies in an AI world shaped by foundation models, rather than trying to become a foundation model superpower itself?
What changed? AI is now treated as a full stack
The previous phase of Israel’s national AI effort was largely about building the basic infrastructure for participation in the AI economy. The TELEM national AI programme, launched in 2021 with a budget of roughly NIS 1 billion, focused on compute infrastructure, research, human capital, access to data repositories and enabling regulation.
The new document describes that earlier effort as the foundation for the next stage, including the establishment of a national supercomputer operated by Nebius, a voucher model for compute access, scholarships for advanced researchers, and direct Innovation Authority investment in deep-tech companies.
The change is also intellectual. The strategy no longer treats AI as a single technology category. It treats it as a stack, a supply chain and a geopolitical system. That is the right framing. AI leadership will not belong only to the companies that train the largest models. It will also accrue to those that build the applications, infrastructure, security, chips, edge systems, robotics, data layers and regulatory pathways that allow AI to be deployed in the real economy.
The four pillars of Israel National AI Strategy
The strategy is built around four pillars: AI-native applications, AI enablers, Physical AI, and Israel’s role in global AI alliances. Together, they amount to a thesis for how a small, technically advanced country can remain relevant in a technology race increasingly shaped by superpower competition.
- AI-native applications. This is the most immediate opportunity for founders. The strategy argues that Israel should lead in sectors where it already has competitive advantages, including cybersecurity, fintech, digital health, enterprise software and defence. It also recognises that traditional SaaS is under pressure. AI agents and large language models are beginning to challenge the assumptions behind seat-based software, fixed workflows and user-operated tools. The next generation of companies will not simply add AI features to existing products. They will be built around AI as the core engine of value creation.
From an early-stage VC perspective, this is the most investable part of the strategy today. Israel has always been good at taking general-purpose technology shifts and commercialising them in specific markets. Israeli founders did not invent the internet, cloud computing or mobile, but they built globally important companies in cybersecurity, DevOps, data infrastructure, fintech and enterprise software. The same pattern can repeat in AI, particularly in verticals where domain knowledge, customer pain and trust matter.
- AI infrastructure and enablers. This is not about Israel building the largest data centres in the world. It is about building the technologies that make AI infrastructure work: chips, networking, storage, cooling, edge compute, inference optimisation, cybersecurity for data centres, MLOps and operational software for AI systems. The strategy points to Israel’s existing strengths in semiconductors, networking, cybersecurity and systems engineering, citing examples such as Mellanox, VAST Data, Tower Semiconductor, ZutaCore, NextSilicon and Hailo.
This is a strategically important point. The AI boom is often described through the lens of model companies, but much of the value may accrue to the infrastructure layer. The companies that make AI cheaper, faster, safer and more reliable will become essential suppliers to the entire industry. For Israel, this could be one of the most durable opportunities: not to own the largest AI factories, but to provide the components, software and security systems that every AI factory needs.
- Physical AI. This is the most ambitious part of the strategy and potentially the most differentiated. Physical AI refers to systems that perceive, understand and act in the physical world, including robotics, autonomous systems, computer vision, sensing, world models and edge AI. The strategy argues that Israel has a real shot at leadership here because the field depends less on massive model-training scale and more on the integration of sensors, hardware, software, algorithms and real-world deployment.
That plays to Israel’s strengths. Defence, autonomous systems, computer vision, edge computing, sensing and robotics are not abstract capabilities in Israel; they are embedded in the country’s military, academic and startup ecosystems. This matters because Physical AI companies are harder to copy than software-only wrappers. A chatbot interface can be replicated quickly. A trusted autonomous system that works in the physical world, under difficult conditions and within regulatory constraints, is a much deeper technical and operational moat.
- Geopolitics. The strategy positions Israel as a critical node in global AI alliances, especially Pax Silica, with the goal of securing access to chips, advanced models, compute infrastructure, R&D collaborations and international standards. The strategy also calls for Israel to strengthen cooperation with the US, pursue Tier 1 positioning under the evolving Bureau of Industry and Security framework, deepen ties with at least three Pax Silica countries, and actively participate in AI governance standards through bodies such as ISO, the OECD and the UN.
This is where the strategy moves beyond classic innovation policy. In AI, access to compute, chips, energy, models and data is increasingly geopolitical. For Israel, alliances are not just diplomatic achievements. They are economic infrastructure.
Sovereign AI, Israeli style
In March this year I’ve written on VC Cafe on AI sovereignty and how governments worldwide are competing and positioning themselves on AI supply chain and independence. The global conversation around sovereign AI is often framed as a push for countries to build their own national models and domestic AI infrastructure. NVIDIA defines sovereign AI as a nation’s ability to produce AI using its own infrastructure, data, workforce and business networks. McKinsey argues that sovereign AI is not a single asset but an ecosystem that connects energy, compute, data, models, cloud platforms and applications, while making deliberate decisions about what needs to be sovereign and what can be supplied through partnerships.
That nuance is crucial for Israel. Sovereign AI should not mean trying to build everything locally. That would be expensive, slow and unrealistic. For a small country, sovereignty should mean control over the critical points in the stack: sensitive data, local inference for national-security and regulated workloads, secure access to compute, trusted model deployment, privacy-preserving use of public-sector datasets, and the ability to participate in allied supply chains without being dependent on any single external actor.
This is where Israel’s strategy fits well with the sovereign AI debate. Israel does not need a full-stack autarkic AI ecosystem. It needs minimum sufficient sovereignty: enough domestic capability to protect strategic interests, enough allied access to remain competitive, and enough technological contribution to be indispensable to partners.
The report also includes the criteria for success in five years.
A cultural advantage: usage
One striking data point sits outside the strategy but reinforces its logic. According to Anthropic’s Economic Index, Israel leads the world in per-capita Claude usage, with an Anthropic AI Usage Index of 7, meaning its working-age population uses Claude seven times more than expected based on its population share.
This matters because AI adoption is not only an economic metric. It is a cultural signal. Countries that use AI intensively will discover its limitations earlier, develop better taste in products, and generate more founders who understand the difference between a demo and a workflow. Israel’s high usage of Claude suggests that the ecosystem is not waiting for policy to arrive. Developers, founders, investors and operators are already experimenting with frontier AI tools in daily work.
For early-stage investing, that is important. The best AI-native companies are likely to come from founders who have internalised the new workflows themselves. They will not build from abstract market maps. They will build from lived experience.
What this means for founders
The strategy is directionally strong, but execution will determine whether it becomes a real advantage or another government document. The practical implications for founders are clear.
First, Israeli founders should not define their ambition by model size. The relevant question is not whether a startup has trained a frontier model, but whether it owns a painful workflow, has access to proprietary or hard-to-replicate data, and can deliver a measurable business outcome.
Second, AI-native does not mean AI-branded. The companies that matter will redesign the product, interface, pricing model and operating model around AI. In many cases, the shift will be from software used by humans to systems supervised by humans.
Third, infrastructure is back. For years, many investors preferred capital-light software businesses. AI changes that equation. Chips, networking, storage, security, edge compute, robotics and data-centre software may require deeper technical risk, but they may also produce more defensible companies.
Fourth, Physical AI deserves more attention from early-stage investors (see my post about what AI is doing to push investors from Bits to Atoms). It is slower and harder than pure software, but it is also less likely to be commoditised by the next foundation model release. Israel’s defence and deep-tech talent base gives it an unusual advantage here.
Finally, policy can matter. Access to compute, public-sector data, regulatory sandboxes, procurement pathways and international partnerships could all materially affect startup formation. The government cannot pick winners, but it can remove bottlenecks that prevent winners from emerging.
The VC view
From a venture perspective, the strategy’s most important contribution is focus. Israel should not measure success by whether it builds a national LLM that competes with OpenAI or Anthropic. It should measure success by whether the next generation of Israeli companies becomes indispensable in the AI economy.
For investors this means that the opportunity lies in:
- AI-native application companies that replace labour and own workflows.
- Infrastructure companies that reduce the cost, latency and risk of AI deployment.
- Physical AI companies that bring intelligence into the real world.
- Israeli technology becoming embedded in the supply chains, standards and platforms of allied countries.
The risk is that Israel spreads itself too thin or confuses activity with advantage. The opportunity is that it concentrates national effort around the layers where the ecosystem is already strong.
Israel has never won by being the largest market or the deepest capital pool. It has won through talent density, urgency, technical courage and global ambition. The AI era does not change that. It makes those traits more valuable, provided they are aimed at the right part of the stack.
That is the central message of the strategy, and it is the right one: Israel should not try to become a smaller version of the US or China. It should become the country that wins where excellence outweighs scale.
