startupim
Ecosystem Intelligence
Artificial Intelligence · Report No. 02

The Israeli AI Stack 2026

How Israeli startups build with AI — the capability mix from agents to vision, foundation-model build-vs-buy, the infrastructure and security layers, and where it is all applied. Proprietary signals from the startupim corpus, read against external market data and the voices of the ecosystem.

Published June 2026 Coverage 13,152 companies · 5,802 AI-classified Source startupim corpus + radar, with cited external data
Fact-checked with a grounded Gemini 2.5 Flash adversarial review before publication
44%
of the startupim corpus is AI-classified (broad “applies-AI” definition) — 5,802 of 13,152
$7.9B
raised by Israeli AI startups in 2025, up ~61% YoY (SNC)
52%
of new Israeli GenAI startups now claim agentic capability (Remagine)
#1
Israel’s rank in AI-talent concentration — 1.98% of the workforce (Stanford / LinkedIn)

00Executive summary

AI is no longer just a sector of the Israeli startup ecosystem — it is a layer running through nearly half of it. This report measures that layer from the technologies companies declare and the code they ship (the startupim corpus and radar), and sets those proprietary signals against external market data and the people shaping the ecosystem.

Five findings frame the year. (1) Scale, two ways to count it. In our broad “applies-AI” classification, 5,802 companies — 44% of the corpus — carry an AI signal; external trackers using a stricter “AI-native” bar put it nearer a quarter to a third, but agree on the money: AI startups capture roughly 47% of all Israeli tech funding (Startup Nation Central, 2025). (2) The agentic turn is real and corroborated. Agents are the fastest-rising capability in our data, and Remagine Ventures independently finds 52% of new GenAI startups now claim agentic capability (CTech, Jul 2025). (3) Buy the model, build the system. Almost no Israeli company trains its own foundation model; the strategy is to orchestrate hosted intelligence — a posture even Israel's own national-strategy committee now accepts. (4) The edge is one layer up from the model: AI security and AI infrastructure, where Israel's cyber DNA compounds — AI-security funding alone reached ~$2.5B in 2025, 64% of all cyber investment (CTech/SNC). (5) Applied everywhere — AI concentrates in business software and health but is material across agriculture, industry, fintech and defense.

The through-line, in the words of Yonatan Mandelbaum of TLV Partners:

“The shift from training to inference to agents isn't just a change in bottlenecks. It's a change in the types of companies that will win… Everything about that change favors Israeli engineering talent.” — Yonatan Mandelbaum, Partner, TLV Partners, in CTech

01How much of the ecosystem is AI — and how to count it honestly

Of 13,152 active companies in our corpus, 5,802 carry an AI or machine-learning signal — 44%. That number is higher than external trackers report, and the gap is instructive: it is a question of definition, not of disagreement.

AI-classified companies by primary sector — startupim corpus (n = 5,802)
Business Software
1,745
Health Tech & Life Sci.
1,021
Media & Entertainment
486
Agriculture & Food
469
Industrial Tech
429
Fintech & Insurtech
379
Cyber Security
367
Automotive & Mobility
225
Aerospace & Defense
188
Education
164
Source: startupim corpus, June 2026. AI-classified = a match on classifications.technologies (Artificial Intelligence / Machine Learning), AI primary sector, or AI tags.

Reconciling the count. Our 44% is a deliberately broad “applies AI” union. External trackers measuring “AI-native” companies land lower: Startup Nation Central describes “more than 2,300 AI startups… roughly one quarter of tech companies,” while other industry counts put it above 30% (SNC). Both can be true — the difference is whether a company that uses AI in a vertical product counts as an “AI company.” Where the sources converge is on capital: AI startups draw ~47% of all funding and ~40% of rounds, and AI-startup funding rose from $4.9B in 2024 to ~$7.9B in 2025 (+61%) (CTech, citing SNC). However it is counted, AI is where the money is.

What this means. “Is this an AI company?” is the wrong question in 2026 — by the broad definition, nearly half the ecosystem qualifies; by the strict one, a quarter. The useful questions are which capability and applied to what — the next sections break both down.

02The capability mix: agents rise, vision endures

Reading each company's own description of what it builds, we counted the distinct AI capabilities named. Descriptions are terse, so these are conservative floors — but their relative order is the signal, and it matches what independent landscape research is reporting.

AI capability named in company description — startupim corpus (floor estimates)
Computer vision
472
Machine learning
455
AI agents / agentic
356
Generative AI
162
NLP
123
Deep learning
109
Chatbot / conversational
94
Copilot / AI assistant
54
LLM (explicit)
43
Fine-tuning
18
Vector / embeddings
16
Foundation model (own)
8
Source: startupim corpus — keyword incidence across about / one_liner / seo.about, June 2026. Counts are floors (terse descriptions undercount); read the ranking, not the absolute values.

Two stories sit in this chart. The first is continuity: computer vision and machine learning remain the ecosystem's deepest AI competencies — a legacy of Israel's strength in sensing, autonomous systems, medical imaging and defense. The second is the agentic turn: AI agents have vaulted to third (356 companies) and clearly lead the older chatbot paradigm (94). This is not just our reading — Remagine Ventures' Israeli GenAI Landscape 4.0 counts 342 generative-AI startups and finds that 104 of the 198 founded since May 2024 (52%) claim agentic-AI capability (CTech, Jul 2025).

“Israeli entrepreneurs are seizing this moment with speed, creativity, and a deep technical edge, building agentic AI systems with real enterprise applications.” — Eze Vidra, Managing Partner, Remagine Ventures, in CTech

Note how few companies claim to train their own foundation model (8) or even mention fine-tuning (18). The named exemplars of the new cohort are agent companies — Alta (an “AI revenue workforce” of sales agents), Notch (agents automating insurance and finance workflows) and Wonderful (enterprise agents; $150M Series B at a $2B valuation within a year of founding). As Mandelbaum frames it, “the bottleneck is now agents.”

What this means. The center of gravity is expanding to include “AI that acts” alongside the ecosystem's enduring strength in “AI that perceives.” Vision and ML are the installed base; agents are the growth edge — and the independent landscape data confirms the slope.

03Under the hood: a Python build stack, AI-coded

For the 684 AI companies with a matched public GitHub organisation, the radar code signal confirms the build stack — and the wider data shows these teams are themselves now built around AI coding tools.

Build-stack signals among code-active AI companies — startupim radar (n = 684 with public org)
SignalCompaniesShareReading
Python in public repos26539%model / data lingua franca
Jupyter Notebook present517%research in the open
Repo themed “model / ML / inference”132core modelling work
Repo themed “vision / image / detect”130vision lineage, in code
Repo themed “agent”98agent frameworks & demos
Repo themed “chat / bot / conversational”86conversational interfaces
Repo themed “LLM / GPT / RAG”54LLM-native tooling
Source: startupim radar sources.github joined to AI-classified companies, June 2026. Repo-theme counts tally companies whose top public repos match the theme; a company can match several.

The repo themes mirror the capability mix almost exactly — model and vision repos lead, with a fast-growing agent cohort (98 companies) now out-publishing classic chat repos. And the engineers writing this code are unusually AI-augmented: the Israel Innovation Authority reports that 95% of Israeli high-tech employees use AI tools regularly (78% daily) and 74% of younger technical staff use code-oriented tools such as Copilot and Cursor (a joint Israel Innovation Authority / Startup Nation Central survey, 2025). By the same data, Israel ranks among the highest countries globally in Anthropic Claude usage per working-age person.

“Almost all workers in the sector already use these tools, and this is not a future projection but a present reality reshaping high-tech employment.” — Dror Bin, CEO, Israel Innovation Authority (2025)
What this means. The AI stack reads as Python + a TypeScript product layer, built by teams that have themselves adopted AI coding tools near-universally — agent-framework experience is now the distinguishing skill, not raw model training.

04The stack, layer by layer — and where Israel plays

Assembling the signals into the canonical AI stack shows a clear pattern: Israeli startups are thin at the foundation-model layer and thick everywhere a vertical product, an agent, an infrastructure pick-and-shovel, or a security control can be built. External funding data corroborates each layer.

The AI stack — Israeli startup presence by layer
Foundation models
Own-model / pre-training
Thin · ~8 build their own (e.g. AI21 Labs); the rest consume hosted OpenAI / Anthropic / Google
AI infrastructure & “picks and shovels”
Compute, inference, serving, GPU orchestration
Rising · ≈$490M / 11 rounds in 2025 (Tracxn est.); Run:AI, Nebius, Fundamental
Orchestration & agents
Agent frameworks, RAG, tool-use, eval
Fastest-growing · 52% of new GenAI startups (Remagine); Wonderful, Alta, Notch
AI security & integrity
Securing agents, models & data
Distinctive edge · ~$2.5B in 2025 = 64% of cyber funding (SNC); Cyera, Prompt Security, Aim
Vertical applications
Health, industrial, agri, fintech, media, defense
Deepest layer · the bulk of the 5,802 — applied AI by domain
Synthesis of startupim corpus + radar signals and external 2026 data (Startup Nation Central, Remagine Ventures, Tracxn, CTech). Layer thickness is qualitative; the named companies and funding figures are externally sourced (see Sources).

The two layers where Israel's existing strengths compound are infrastructure and security. On infrastructure, capital is flowing into the “AI supply chain” — Nebius alone is deploying 4,000 Nvidia B200 GPUs and 80 MW of Israeli data-centre capacity (CTech). On security, Israel's cyber lineage has produced the defining companies of the AI-security category: Cyera ($600M at a $12B valuation in 2026), Prompt Security (acquired by SentinelOne) and Aim Security — and AI-security is now 64% of all Israeli cyber investment, up from 34% a year earlier. The model layer, by contrast, is a deliberate non-bet.

“The dependence on American or Chinese model providers is a strategic vulnerability… A more appropriate title for AI security in the agentic-infra era is: AI integrity.” — Yonatan Mandelbaum, Partner, TLV Partners, in CTech
What this means. The defensible Israeli AI bets are not at the model layer but around it — inference and serving infrastructure, agent orchestration, AI security/integrity, and vertical applications backed by proprietary data. The build-vs-buy decision has overwhelmingly favoured buy the model, build the system.

05The national bet — and the debate inside it

The “buy the model, build the system” posture is now national policy, but it arrived through genuine debate about whether Israel should build a sovereign model at all.

Israel approved a National AI Program (the directorate established September 2025, a broader package in June 2026), budgeted at roughly NIS 1 billion to date and moved into the Prime Minister's Office, with pillars in talent, sovereign compute, energy and regulatory acceleration (CTech; National AI Program, official PDF). The government's own Nagel Committee was blunt that Israel is “far behind” the hundreds of billions major powers are deploying — and questioned whether a sovereign Hebrew LLM is even feasible, given Hebrew lacks the corpus scale to train a frontier model from scratch, suggesting fine-tuning an existing model instead (INSS).

The debate, distilled. One camp argues a sovereign-model “build” strategy would catalyse the domestic stack and reduce strategic dependence; the other — now the mainstream — argues Israel should concede the model layer and win the application, agent, infrastructure and security layers “where excellence outweighs scale.” Our data shows the ecosystem has already voted with its code: it builds on models, not models.

06Outlook for 2026–27 — and the bull/bear case

The bull case. Agents keep compounding (already past chatbots in our data and the majority of new GenAI startups); the infrastructure layer thickens as GPU and data-centre capacity lands locally; and AI security graduates from feature to category, carried by Israel's cyber DNA. H1 2026 reinforced it: Israeli startups raised roughly $8.6B, up ~45% YoY, led by AI, cyber and enterprise software (CTech / Ynet).

The bear case, which the data also supports. 2025 set funding records on a decade-low 717 deals, with half of all capital in $100M+ rounds — the ecosystem is narrowing onto a few high-conviction security and AI fields, raising concentration and bubble concerns echoed globally (CTech / SNC annual report). Avi Hasson, CEO of Startup Nation Central, frames the year not as a return to normal but as “a pivot toward high-conviction maturity” — a description that contains both the optimism and the risk.

For builders, the strategic reading is consistent across our data and the external voices: own the system, not the model; compete on the agent, infrastructure, security and vertical layers; and treat proprietary data and Israel's engineering density — first in the world for AI-talent concentration — as the moat.

Sources

  1. Startup Nation Central — Israeli Tech Annual Report 2025. startupnationcentral.org
  2. CTech — Israeli AI funding 2025 (~$7.9B, +61%; AI-security 64% of cyber). calcalistech.com
  3. CTech — 2025 annual figures: $15.6B raised, 717 rounds, $74B exits. calcalistech.com
  4. Remagine Ventures — Israeli GenAI Landscape 4.0 (342 startups; 52% agentic), via CTech. calcalistech.com
  5. CTech — Mapping Israel's AI-infrastructure opportunity (TLV Partners; Nebius; Run:AI). calcalistech.com
  6. Israel Innovation Authority — AI-tools usage survey 2025 (95% use; 74% code tools). innovationisrael.org.il
  7. National AI Program — official PDF, Innovation Israel. innovationisrael.org.il
  8. CTech — National AI program moved to PM's Office (Nagel Committee). calcalistech.com
  9. INSS — Does Israel Need a National Language Model? inss.org.il
  10. Stanford AI Index / LinkedIn — AI-talent concentration (Israel #1, 1.98%), via CNBC/Times of Israel. timesofisrael.com
  11. Company rounds: Decart, Wonderful, Alta, Notch, Cyera, Prompt Security, Aim Security — CTech / Globes (linked inline).
  12. CTech / Ynet — H1 2026 funding (~$8.6B, +45% YoY). ynetnews.com
Related: Israeli Startup Tech Stack 2026 →