Opinion
Israel’s next AI giants could come from infrastructure
"Developers and their agents are driving a new infrastructure wave," write Glenn Solomon and Dan Cahana of Notable Capital. "Companies that run the workloads can capture far more of its value, and Israeli founders are well placed to build them."
Every application an AI agent creates needs somewhere to run. As software production accelerates, that is opening a vast infrastructure market. We expect the largest companies of the next decade to emerge from this layer.
In all of last year, developers pushed nearly 1 billion commits to GitHub. In August this year, GitHub saw 2.9 billion in a single month. That 35x jump in the monthly pace comes from developers directing agents that write, test, and ship code around the clock.
For most of software history, developer time limited how much software got built. The cloud gave each developer far more reach and forced a rewrite of the entire infrastructure stack, from compute and data to networking and security. But in the cloud era, one developer was still fundamentally one unit of labor. Agents remove that limit, forcing a second, much bigger rewrite. Infrastructure categories built for human-scale consumption will become orders of magnitude larger as agents become their main consumers.
In the cloud era, most infrastructure startups sold software that helped customers manage a workload, such as running a database, serving a web application, or processing video. AWS, Azure, or Google Cloud supplied the storage and compute underneath. The startup captured roughly 5–10% of that cloud spend, and the hyperscaler kept the rest. The three largest cloud providers now generate about $386 billion in combined annual run-rate revenue, based on their second-quarter earnings this year. The software built around them earns a fraction of that.
Agents make code cheap to produce, which makes workload-management software easier to replicate. We expect that 5–10% attach rate to compress toward 1–2%. Many of Israel's strongest cloud-era companies sold software that sits on top of hyperscaler infrastructure. That model built enormous businesses, but for a company starting today, it's a harder place to begin.
When a company takes responsibility for running one kind of workload, the infrastructure spend flows through its own business. At Notable, we call these companies workload clouds. Agents create far more workloads, and the workload cloud captures a much larger share of the spend on each one.
Snowflake set the early template in data. It became the place data warehouse workloads ran, where an earlier generation of vendors had sold software for customers to operate themselves. We now see the model spreading to new workloads: Vercel in web and AI applications, fal in generative media, and Browserbase in browser sessions, all Notable Capital portfolio companies.
Building these businesses means running infrastructure at scale, managing compute margins, and winning self-serve adoption from developers and their agents before any enterprise sales motion begins.
At Notable, we've followed developers for 20 years because their choices predict where infrastructure goes. Developers embraced AWS while enterprises were still skeptical of cloud computing. When cloud-native infrastructure companies arrived to sell to large enterprises, developers inside those companies already used and trusted them. That reinforced a simple investing framework: Follow the developer
That signal led us to Anthropic. We invested early last year, when consumer attention favored OpenAI. But developer behavior tilted toward Anthropic. Claude 3.5 Sonnet led the coding benchmarks developers watched closely. Cursor moved to Sonnet after running its own internal evaluations, and co-founder Michael Truell later described the model as a major jump. Anthropic's API business surged on that demand. By the end of last year, OpenRouter's analysis of 100 trillion tokens showed Anthropic with more than 60% of programming-related spend.
Israel already has engineering teams that run complex systems under pressure. Security is the clearest opening for Israel. Companies built here developed much of the cloud security stack. Agents spin up their own environments, call services on a user's behalf, and act without a human reviewing each step. Securing them is a new problem with no clear AI native winners yet, and it draws directly on expertise developed here.
Which infrastructure gains the most value when agents become its main consumers? Which categories that looked small in the cloud era become large in an agentic one? Developers still shape the market, and agents are expanding the infrastructure they consume. There’s room for billion- and trillion-dollar companies that run and secure those workloads.
Glenn Solomon is Managing Partner and Dan Cahana is a Principal on the Investment Team at Notable Capital.
