Sheba Medical Center in Ramat Gan will deploy OpenAI’s ChatGPT for Healthcare across its entire hospital network, the medical center said on July 28. It is OpenAI’s first clinical AI deployment inside a hospital system outside the United States. Doctors, nurses and researchers get a clinical reasoning tool built to attach citations and publication dates to every answer.
Sheba has done this before. Two other AI partnerships landed at the hospital in the past year, with Hippocratic AI and with an NVIDIA-backed genome research effort. Global AI companies increasingly treat Israel as the place to test their highest-stakes tools first. Israel supplies the talent and the willing test subjects, hospitals, banks, security agencies and now factories. Its own government’s data shows a gap in the electricity and server capacity needed to keep up with the demand it has created.
A Hospital Network Becomes OpenAI’s First Test Case
The platform can synthesize peer-reviewed studies, clinical guidelines and public health sources, attaching citations and publication dates so clinicians can check the evidence themselves. Sheba is folding in its own clinical protocols and policy documents, so answers reflect the hospital’s own standards rather than general medical literature alone.
The deployment carries zero-data-retention and non-training guarantees, meaning patient interactions do not feed back into OpenAI’s broader models. Newsweek has ranked Sheba among the world’s best hospitals every year since 2019.
Israeli medtech company TriO also unveiled a device this month aimed at cutting the number of amputations linked to diabetic foot ulcers, part of a broader run of health tech announcements out of Israel.
Three AI Deals in a Single Year
Sheba’s ARC Innovation center has signed three AI partnerships in about a year.
- July 2025: Hippocratic AI signs on to bring generative AI into everyday patient care.
- 2025: ARC Innovation at Sheba and Mount Sinai launch a genome research collaboration built with NVIDIA.
- July 28, 2026: OpenAI names Sheba its first international ChatGPT for Healthcare deployment, covering the hospital’s full network.
Sheba also joined the Trustworthy and Responsible AI Network, a consortium of health systems that share AI safety standards.
The World’s Densest AI Talent Pool
Israel’s pull on AI companies rests on a talent base LinkedIn now ranks first in the world. Some 2.1% of LinkedIn members registered in Israel in 2025 carried AI skills, the highest concentration LinkedIn found anywhere, ahead of Singapore’s 1.8% and Luxembourg’s 1.6%.
| Country | AI Talent Concentration, 2025 | Global Rank |
|---|---|---|
| Israel | 2.1% of LinkedIn members | 1st |
| Singapore | 1.8% of LinkedIn members | 2nd |
| Luxembourg | 1.6% of LinkedIn members | 3rd |
The pattern repeats across small, dense economies. Israel, Singapore, Luxembourg and Estonia all rank near the top despite modest populations, out-producing far bigger countries on AI skills alone. Private investors have followed the talent. Around $19 billion has flowed into Israeli AI-related companies between 2013 and 2025, enough to rank the country roughly fifth worldwide for private-sector AI investment.
Israel Ranks 19th in the Infrastructure AI Needs
The same LinkedIn-linked index that puts Israel first on talent does not rank the country among the top 10 for government investment in AI. It places Israel 19th of 30 economies studied on energy readiness and 22nd of 30 on the cost of building data centers, according to Ynet News’ review of the underlying figures.
That gap matters because deployments like Sheba’s or Microsoft’s run on compute, not just code. Training and hosting the heaviest AI workloads takes server farms and steady, affordable power, infrastructure Israel has not built at the scale of its talent pool. Statista’s tracking shows Israel’s AI talent concentration climbing steadily since 2016, even as the infrastructure rankings lag behind it. For now, the hyperscalers bring their own compute and cloud capacity into Israel rather than relying on it already existing there.
Herzliya’s Cyber Team Takes on Anthropic
Microsoft unveiled Project Perception on July 27, an agentic cybersecurity platform built around AI agents that share intelligence and work continuously to find vulnerabilities before attackers do. Alongside it came MAI-Cyber-1-Flash, Microsoft’s first AI model built specifically for cybersecurity work.
Microsoft said a significant share of Perception’s underlying capabilities were developed at its R&D center in Herzliya, with the first capability built on the new architecture coming directly out of that Israeli team’s work. Microsoft said the new model outperforms Anthropic’s Claude-based cyber tool at spotting software vulnerabilities, a comparison drawn from its own testing rather than an independent benchmark.
Perception opens in preview to all Microsoft customers on August 3.
Physical AI Moves from Software to Steel
Physical AI pairs AI software with sensors and actuators so a system can perceive and act in the real world, not just answer questions on a screen. Israel now counts 125 startups building in the category.
- 125 startups now build physical AI products in Israel, spanning robotics, sensors and autonomous hardware.
- $10 billion-plus raised by the sector to date.
- Eight unicorns have emerged from physical AI specifically, alongside 24 exits.
That is a shift for a start-up scene long dominated by cybersecurity. Wiz, valued around $12 billion by one tracker, alone accounts for close to a fifth of Israel’s total unicorn valuation, a reminder of how concentrated the older wave of Israeli tech success has been. Physical AI’s eight unicorns show that concentration starting to spread into hardware.
A Second Address in Shanghai
Physical AI collaboration between Israel and China runs through a dedicated hub. The China-Israel Innovation Hub, based at Taopu Smart City in Shanghai’s Putuo district, focuses on AI and robotics, healthcare and life science, and internet and IT. It now counts more than 500 registered enterprises, alongside a joint research program offering subsidies of up to 2 million yuan, about $280,000, covering as much as half of a project’s R&D costs. The hub keeps an overseas office in Tel Aviv.
That relationship now runs into a tightening US policy environment. Washington’s Commerce Department issued sweeping export controls on AI chips bound for China in December 2024, and in May 2026 extended licensing requirements to advanced computing exports headed to Chinese-headquartered entities anywhere in the world. Congress approved the Chip Security Act in March 2026, directing tracking technology to be built into chips themselves. A CSIS analysis of allied export control authority found Washington increasingly leans on partners such as Japan, South Korea and the Netherlands to align their own rules, since chip and AI supply chains run through those countries’ companies as much as through America’s. Israel, with its own deep security and technology ties to Washington, sits in a similar position.
The Sandbox as Israel’s Real Export
Israel’s Health Ministry and Innovation Authority opened a first-of-its-kind regulatory sandbox for AI in July, letting companies test high-autonomy AI systems directly in clinical settings under government supervision rather than waiting through a standard approval track. The sandbox already has three pilot companies running supervised trials: Plessanmore, Cordio Medical and Simhawk.
Egypt is taking a different route toward a similar goal, leaning on a World Bank pledge to fund its AI and digital economy plans rather than standing up its own sandbox, though the World Bank has not yet attached a price tag to that support. Israel’s approach costs less in headline dollars and produces pilot data faster.
The sandbox model may be the piece other health systems copy first, a working example, three companies deep, months before most competitors have drafted a rulebook.
Frequently Asked Questions
What Is a Regulatory Sandbox in Healthcare AI?
A regulatory sandbox lets a company test a high-autonomy AI tool directly on real patients or in live clinical workflows, under a regulator’s direct supervision, before the tool goes through standard approval. Israel’s Health Ministry and Innovation Authority run theirs jointly, a structure few national health systems have formalized so far, with results from the pilot trials expected to feed into future approval rules.
Does OpenAI Train on Sheba’s Patient Data?
No. The Sheba deployment carries zero-data-retention and non-training guarantees, so patient interactions and clinical outputs stay isolated within Sheba’s own systems rather than feeding into OpenAI’s broader models. Sheba is also layering in its own clinical protocols, so the tool’s answers reflect hospital-specific standards rather than generic guidance alone.
What Does a Physical AI Startup Build?
Examples range widely. Haifa-based Lynxight applies computer vision to standard pool security cameras to spot swimmers in distress in real time, one of 125 Israeli companies building physical AI products spanning healthcare devices, warehouse robotics and autonomous vehicles.
Why Do Global AI Companies Keep Choosing Israel First?
Companies point to the density of AI talent, the government-run sandbox that speeds up real-world testing, and Israel’s long-standing security and technology ties with the United States. Microsoft’s Herzliya team, OpenAI’s Sheba deployment and the country’s 125 physical AI startups all point to the same underlying draw: a small market willing to let outside companies test first, at scale, before those tools reach anywhere else.
