A few months ago, I wrote here on VC Cafe about The Global Race for AI Sovereignty: Where Does Israel Stand? The point was that AI sovereignty is no longer a theoretical debate for policy wonks. It is becoming a core pillar of national security, economic resilience and geopolitical leverage.
If a country cannot access compute, train or deploy models on its own terms, protect sensitive data, and build the applications that matter to its economy and defense, it will increasingly depend on the goodwill of foreign companies and foreign governments. The US government recently gave a good reminder of this when it decided to ban access to Anthropic’s Fable 5 and Mythos 5 models outside of the US.
Last week, the Israeli government took a major step toward answering that challenge, approving a new National Artificial Intelligence Program led by the National AI Directorate in the Prime Minister’s Office.
The plan is ambitious. It includes sovereign compute infrastructure, a target of 100,000 processing units, a national quantum computer, investment in AI education and workforce retraining, a National AI Institute, acceleration hubs, AI adoption across public services, and a focused push into Cyber AI, Physical AI and deepfake defense.
The number that grabbed my attention, though, was not in the government announcement. It came from the analysis that followed: a 100,000-GPU national infrastructure effort could carry a potential price tag of $20 billion to $30 billion or more, depending on the hardware, data center requirements, power, cooling, networking and refresh cycles.
That is a huge number for a country the size of Israel.
As a VC investing in Israeli startups, I read this less as a government press release and more as a market signal. AI is moving from “startup theme” to national infrastructure. The opportunity for founders is not simply to win grants or sell software to ministries. It is to build the next generation of companies that sit on top of this emerging infrastructure layer and translate national capability into global products.
Why this matters now
Israel has always punched above its weight in technology. Cyber is the obvious example. A small country created a globally dominant cyber ecosystem by combining military needs, elite technical talent, venture capital, multinational R&D centers and founders with a strong bias for action.
AI is different.
Cyber was born in a world where brilliant teams could create massive value with relatively modest infrastructure. AI, by contrast, is deeply dependent on compute, data, chips, power, networking, cooling, cloud availability and model access. Talent still matters enormously, but talent alone is not enough.
This is why the new national plan matters. It acknowledges that Israel cannot rely only on “Startup Nation” instincts in a world where intelligence is increasingly manufactured in data centers.
To compete in AI, Israel needs to become more of a Systems Nation: able to coordinate energy, compute, research, regulation, defence needs, procurement and entrepreneurship in one strategic direction.
That is a difficult shift. It is also necessary.
What is actually in the National AI Plan?
The government resolution touches several layers of the AI stack. Some are immediate and practical. Others are directional. All of them matter for founders and investors trying to understand where the next wave of opportunity may emerge.
1. Sovereign compute infrastructure
The headline ambition is the procurement or allocation of 100,000 processing units to serve Israel’s economy, government, academia and national security needs.
This is the most expensive and strategically important part of the plan.
Israel already has a first phase of a national AI supercomputer effort under the Telem program, giving companies and researchers access to Nvidia B200 accelerators at subsidized rates. But the gap between a first national compute cluster and a 100,000-unit sovereign infrastructure target is enormous.
This is not just a hardware procurement issue. It requires data centers, land, power, grid connections, cooling, networking, security, operating expertise and a regular upgrade cycle. GPUs become obsolete quickly. The cost is not one-off capex. It is a long-term national commitment.
For startups, the important point is this: if Israel can meaningfully expand local access to compute, it reduces one of the biggest constraints for AI companies building in regulated, sensitive or defense-adjacent sectors.
Not every startup needs to train a frontier model. But many startups need affordable, reliable and compliant access to advanced compute in order to fine-tune models, build domain-specific systems, run simulations, perform evaluations, and serve customers that cannot send sensitive workloads abroad.
2. A national quantum computer
The plan also restarts the push to establish a national quantum computer based, where possible, on Israeli-developed technology.
Quantum should not be confused with the immediate AI opportunity. AI is here now. Quantum remains earlier, more technical and more uncertain in its commercial timelines. But from a national strategy perspective, it makes sense to keep the two tracks adjacent: both are foundational technologies, both require deep scientific talent, and both can produce strategic advantage in defense, optimization, cryptography, chemistry, simulation and advanced computing.
For startups, the near-term opportunity is less likely to be “build the Israeli quantum computer” and more likely to be in the surrounding layers: control systems, hybrid quantum-classical workflows, simulation tools, error mitigation, developer tooling, optimization use cases and eventually quantum-safe security.
3. Human capital and AI education
The plan calls for investment in AI talent across schools, academia, the workforce and professional retraining.
This may be the most important part of the plan, even if it is less eye-catching than GPUs.
Israel’s unfair advantage in AI will not come from outspending the U.S., China or the Gulf on infrastructure. It will come from the density of technical talent, research capability, applied engineering, security experience and entrepreneurial urgency.
AI is a deeply mathematical field. In cyber, elite military units have been the talent engine. In AI, academia plays an even more central role. PhDs, professors, research labs and mathematically trained engineers are critical to building systems that are not just demos, but robust, efficient and reliable products.
This is where initiatives like the National AI Institute, acceleration hubs, university programs and private ecosystem efforts can matter. The bridges between academia, industry, government and venture capital need to become wider and faster.
Israel does not have the luxury of letting its best AI research stay trapped in papers, or its best founders waste years without access to the right compute, customers and capital.
4. AI in public services
The plan also calls for AI adoption across public services: reducing waiting times, making government information more accessible, improving citizen services and streamlining ministries.
This could become a meaningful market, but only if procurement changes.
The Israeli government is often a difficult first customer: slow sales cycles, fragmented budgets, risk-averse procurement and unclear ownership. But AI creates pressure to modernize. Citizens already experience consumer-grade AI tools in their daily lives. They will increasingly expect government services to be faster, more personalized and easier to navigate.
For startups, the opportunity is not generic “AI for government.” It is specific workflows where the pain is visible and measurable: benefits eligibility, municipal services, healthcare administration, licensing, planning approvals, immigration, education support, emergency response, document processing, procurement and citizen-facing service agents.
The winners will not be the companies that pitch AI as magic. They will be the companies that can handle messy data, Hebrew and Arabic language nuance, privacy, security, auditability and integration into old systems.
5. Cyber AI, Physical AI and deepfake defense
The government has chosen three areas of national focus: Cyber AI, Physical AI and countering deepfake threats.
This is the right instinct. Small countries cannot win by trying to do everything. Israel should focus where it has real comparative advantage.
Cyber AI is a natural extension of Israel’s strongest technology sector. AI is changing both sides of the cyber equation. Attackers can generate exploits, phishing campaigns and malware faster. Defenders need autonomous triage, detection, investigation, remediation and simulation.
Physical AI is equally interesting. It is not just humanoid robots. It includes AI systems that perceive, decide and act in the physical world: drones, manufacturing systems, logistics, industrial inspection, agriculture, autonomous defense systems, sensors, robotics and edge devices. In the physical world, mistakes are costly. Reliability, safety, latency and robustness matter far more than in a chatbot demo.
Deepfake defense is becoming a national security issue, but also a commercial market. As synthetic media improves, governments, banks, media companies, insurers, marketplaces and enterprises will need tools for identity verification, content provenance, voice authentication, fraud detection and trusted communications.
These three areas fit Israel well because they combine software, security, operational urgency and real-world constraints.
The VC view: sovereignty is not autarky
There is a risk in any national AI plan: confusing sovereignty with total independence.
Israel is not going to build the entire AI stack alone. It will not fabricate all the chips, own all the frontier models, control every layer of the cloud, or outspend Microsoft, Google, Amazon, OpenAI, Nvidia, Meta, China or the Gulf.
That should not be the goal. Sovereignty does not mean autarky. It means the ability to make strategic choices without being completely dependent on external actors.
For Israel, a pragmatic AI sovereignty strategy should focus on four questions:
- Where must we have local control because the data, security needs or operational context are too sensitive?
- Where can we use global infrastructure but still retain Israeli IP, expertise and bargaining power?
- Where do we have a genuine comparative advantage that can produce export-scale companies?
- Where should we avoid wasting national resources by trying to replicate what hyperscalers and frontier labs are already doing better?
This is where founders need to be disciplined. A government plan can create tailwinds, but it can also create noise. The existence of a national priority does not automatically create a venture-scale market. The best startups will use the plan as leverage, not as a business model.
Where are the startup opportunities?
Here are the areas I would be most excited to see Israeli founders attack. I’ve also taken the liberty to plug in Israeli startups that we (Remagine Ventures) invested in.
1. GPU efficiency, orchestration and cost reduction
If compute becomes the new oil, efficiency becomes a huge business.
Every company is waking up to the cost of AI. Training is expensive, but inference at scale can be even more painful. Enterprises that were enthusiastic about encouraging employees to use AI as much as possible are now starting to ask basic CFO questions: which models are we using, for what tasks, at what cost, and with what measurable ROI?
Israel has an opportunity to build the tooling layer around compute efficiency: model routing, inference optimization, caching, distillation, workload scheduling, GPU utilization, observability, cost controls, evaluation and procurement intelligence.
A national compute layer will create demand for exactly these capabilities. Someone will need to help companies and researchers use scarce GPUs intelligently. Our investment in TroupAI is a clear example of this: enabling enterprises to move to local small language models or model swarms, increasing the accuracy while reducing costs.
2. Secure and sovereign AI infrastructure
As more sensitive AI workloads move into local or controlled environments, startups can build the software layer that makes those environments usable.
This includes model deployment, access controls, data governance, secrets management, audit logs, compliance, air-gapped AI operations, private RAG, secure fine-tuning, evaluation, red-teaming and monitoring of model behavior.
The opportunity is especially strong in defense, healthcare, financial services, critical infrastructure and government. These customers want the benefits of frontier AI, but they cannot treat public APIs as the default architecture.
Israeli founders understand these constraints better than most. The next wave of enterprise AI infrastructure will be built around trust, not just performance. We invested in Pandorian AI, a governance layer for machine generated code, enabling organisations to control what code gets merged intro production.
3. Domain-specific models for regulated sectors
Competing head-on with U.S. frontier labs to build a general-purpose foundation model is not a sensible strategy for most startups.
But domain-specific models are a different story.
There is room for models and systems trained around specific workflows, languages, datasets and regulatory environments: cyber, healthcare, legal, financial crime, defense, insurance, industrial operations, education and public services.
The opportunity is not just to fine-tune an open-source model and call it a company. The opportunity is to own the workflow, the proprietary data loop, the evaluation layer, the user experience and the go-to-market.
In many verticals, the model is only one ingredient. The product is the system around it. We invested in vertical AI agent startups still in stealth, enabling exactly this.
4. Physical AI and edge intelligence
Physical AI is one of the most interesting areas for Israel because it maps to the country’s strengths: defense systems, sensors, robotics, computer vision, embedded systems, drones, autonomous operations and mission-critical software.
The world does not need more AI demos. It needs systems that can operate reliably in factories, fields, warehouses, hospitals, roads, borders and disaster zones.
Startups can build in simulation environments, robotic control, edge inference, sensor fusion, safety layers, industrial copilots, autonomous inspection, fleet coordination, predictive maintenance and human-machine interfaces.
The key is reliability. In the physical world, “mostly correct” is not good enough. Our investment in Trulux AI, which developed visual AI models to verify the authenticity of physical goods, using nothing more than a mobile phone.
5. Cyber AI and autonomous defense
AI is changing cyber faster than most enterprises can adapt.
Attackers can now scale reconnaissance, phishing, vulnerability discovery and social engineering. Defenders need systems that can reason across logs, code, identities, endpoints, cloud infrastructure and business context.
Israel’s cyber ecosystem is already world-class. The next generation of cyber companies will use AI not as a feature, but as the core operating layer: autonomous SOC analysts, AI red teams, agentic remediation, identity security for AI agents, AI supply chain security, model abuse detection and continuous security testing.
The government’s focus on Cyber AI should reinforce what the market is already demanding. We made an investment in stealth to protect data centres and critical infrastructure. It combines the strengths of AI to simulate attacks with the physical defence aspects.
6. Deepfake defense, provenance and trust
Deepfake defense is often discussed through a national security lens, but the commercial implications are broad.
Banks will need to know whether the person on a video call is real. Media companies will need to verify content provenance. Enterprises will need to protect executives from voice cloning and impersonation. Marketplaces will need to detect synthetic fraud. Governments will need trusted public communication channels.
This market will likely include detection, authentication, watermarking, provenance standards, secure identity, reputation networks and real-time risk scoring.
Israel’s strengths in cyber, fraud prevention and intelligence analysis are highly relevant here. We were early investors in HourOne, a pioneer in AI avatars acquired by Wix in 2025. We are aware on how uncanny the technology has gotten and are looking now at defence against uncanny video and voice AI.
7. AI for government and public-sector productivity
The public sector is one of the largest possible beneficiaries of AI, but also one of the hardest markets to serve.
The opportunity is not to replace government workers with chatbots. It is to make complex systems navigable: forms, claims, permits, medical records, municipal requests, benefits, procurement, regulation and citizen support.
Startups that can build secure, multilingual, auditable systems for the public sector may find opportunities not only in Israel, but globally. Every government faces the same pressure: deliver better services with limited budgets and outdated infrastructure.
8. Hebrew and Arabic AI
One under-discussed component of AI sovereignty is language.
If the dominant AI systems are trained primarily on English, or optimized for U.S. enterprise use cases, Israel will need better models, datasets and evaluation tools for Hebrew, Arabic and mixed-language environments.
This matters for government services, education, legal workflows, media, healthcare, defense and customer support. It also matters for inclusion. A country cannot claim sovereign AI capability if large parts of its population cannot use AI in their native language with high accuracy and cultural nuance.
There is a startup opportunity in localized models, synthetic data, evaluation benchmarks, speech recognition, translation, search, agentic workflows and vertical applications built for Hebrew and Arabic from the ground up.
There are many more opportunities of course. This list is not exhaustive. Founders should not build a company whose only customer is a government grant program.
It’s also a bit too early to assume that “sovereign AI” is a category buyers understand. Smart founders will avoid building generic AI infrastructure unless they have a specific wedge, a clear buyer and a reason to win.
Also, at the speed startups move, they should not wait for the national plan to be fully implemented. The best companies will move MUCH faster than government.
The real test: execution
Israel has had AI programs before. The issue has not been a lack of smart people or good intentions. The issue has been execution: budgets, ownership, timelines, procurement, infrastructure and coordination.
That is why this new plan should be viewed as a reset, not a victory lap.
The ambition is correct. The timing is urgent. The direction is encouraging. But the outcome will depend on whether Israel can actually build the infrastructure, connect it to entrepreneurs, empower academia, remove bottlenecks, and create demand from real customers.
A 100,000-GPU ambition makes headlines. But the real measure of success will be the companies that get built because of it.
- Can Israeli startups train models locally, protect sensitive data, reduce compute costs, build trusted AI systems, serve regulated industries and export products globally?
- Can academia become a faster engine for company creation?
- Can government become a better design partner and first customer?
- Can Israel turn national necessity into global category leaders, the way it did in cyber?
That is the opportunity.
Israel does not need to own every layer of the AI stack. But it does need enough control, talent and infrastructure to build from a position of strength.
The move from Startup Nation to Systems Nation will not be cheap, simple or immediate. But if the government can provide the rails, and founders can build the products, Israel’s $30B AI sovereignty bet could become more than a policy headline.
It could become the foundation for the next great wave of Israeli tech.
He is a former General Partner at Google Ventures (GV) in Europe, former head of Google for Entrepreneurs in Europe, and founding head of Campus London, Google's first startup hub. Eze writes on Israeli tech, venture capital, artificial intelligence, and founder strategy.
He is also the founder of Techbikers, a nonprofit that brings together the startup ecosystem on cycling challenges in support of Room to Read.
- Israel’s $30B AI Sovereignty Bet - June 28, 2026
- Weekly Firgun Newsletter – June 26 2026 - June 26, 2026
- Cheap Prototype, Expensive Maintenance - June 25, 2026
