AI Startup Corma Raises $125M at $1B Valuation Three Months After Going Public
Corma, a startup that trains its own AI models to defend against cyberattacks, raised $125M in a Series A round at a $1B valuation. In other words, investors believe the company is worth a billion dollars, even though it emerged from stealth only three months ago. Demand from Fortune 500 companies has been so strong that Corma has increased its server capacity 24x, yet some customers are still waiting. The takeaway: for an AI product built on its own models, growth is limited by compute shortages, not by sales.
AI-processed from TNW; edited by Hamidun News
Corma, an Israeli startup that trains its own AI models to protect companies from cyberattacks, raised $125M in a Series A round. A "Series A" here means the first major round after launch, and a "$1B valuation" means investors believe the entire company is worth a billion dollars. For a company that emerged from stealth only three months ago, this is a very fast path. Meanwhile, since August Corma has increased its server capacity 24x, because customers are arriving faster than the company can buy compute, and some of them are waiting on a waitlist.
Who invested and what is known about the business
NEA led the round. Sequoia Capital, Khosla Ventures, and Coatue, which had already invested in Corma, invested again. In total, the company has now raised $185M. The funds will go toward scaling its models, buying compute, and, according to the announcement, selectively expanding the team. Corma currently has about 25 employees, according to Calcalist.
Founder and CEO Alon Ploda is a graduate of Unit 8200, the Israeli intelligence unit. According to the company, its team includes researchers from Google DeepMind and cybersecurity specialists from the same place. Corma does not disclose revenue. Ploda told Calcalist only that revenue is growing so fast that any figure named today may already be out of date. Its customers include companies in healthcare, finance, energy and critical infrastructure, retail, and telecom. Demand, according to Ploda, comes from organizations on the Fortune 100, Fortune 500, and Fortune Global 500 lists, meaning the largest companies in the US and worldwide.
A server shortage instead of a customer shortage
The most unusual detail of this story is the bottleneck. Startups usually struggle with sales, but Corma's problem is different. Ploda acknowledges that one of the hardest challenges is acquiring compute quickly enough. While the company adds capacity, customers wait. For founders, the picture is familiar: if a product is built on proprietary trained models, access to GPUs and the budget for them limit growth.
How it works and how the company sells itself
Corma trains models to sift through audit logs, security events, and network traffic, looking for weak signals over long time spans. A simple example: a single suspicious night-time login means nothing. But if it is followed over weeks by rare database queries and unusual network traffic, it could be the start of an attack. Humans find such chains hard to spot, while models trained on large volumes of data handle them more easily.
The product is sold as "AI workforce": agents work through security tools the customer has already purchased. Ploda tells Calcalist that the entire process is fully autonomous and runs without human involvement. That is a strong claim. For banks and infrastructure operators, a key question remains where the boundary of autonomy lies and who is accountable if an agent takes a wrong action. The article does not say how the company answers this.
Corma explains its position through the gap between attack and defense. In October, according to the company, a wave of attacks and breaches known as Black October took place. Ploda calls cybersecurity "a battle of intelligence against intelligence." Code-focused models are getting better at finding and exploiting vulnerabilities, while defense requires sifting through enormous volumes of logs. On its homepage, the company also criticizes the AI industry without naming names: some players, it says, are willing to profit from offensive model capabilities while talking about safety.
What to take with a grain of salt
The same website publishes a chart for DefenseBench V5. On it, Corma's model v0.4 outperforms Claude Fable 5.2, Gemini 4 Carbon, GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, Kimi K3.1 Pro, and Grok 4.7 in defensive task quality and execution speed. This material comes from the company itself, there is no independent verification, and the methodology is not disclosed in the text. Revenue is also not named, so the $1B valuation rests mainly on the account of demand and the team's composition.
Corma promises to show its next generation of models "soon." The practical takeaway for those responsible for security at a large company: autonomous defensive agents are worth evaluating, but you should demand transparent benchmarks, an audit log of the agent's actions, and the ability to limit its permissions. For founders, the signal is this: a narrow vertical with proprietary models and a strong team can reach a billion-dollar valuation quickly, as long as it has enough compute.
- Nov 21, 2026A viral GTA 6 Claude clip looks fake: half an hour of control burned through the 5-hour Pro limit
- Nov 21, 2026Claude Code Speeds Up Snowboard Kids 2 Decompilation: From 20% to 45% of Functions in Three Months
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026Anthropic cuts Claude Fable 5.2 price by 20%, and on the coding test it nearly matches Opus 5.5
Key Questions Answered
Who invested and what is known about the business?
NEA led the round. Sequoia Capital, Khosla Ventures, and Coatue, which had already invested in Corma, invested again. In total, the company has now raised $185M. The funds will go toward scaling its models, buying compute, and, according to the announcement, selectively expanding the team. Corma currently has about 25 employees, according to Calcalist. Founder and CEO Alon Ploda is a graduate of Unit 8200, the Israeli intelligence unit.
What are the key details on A server shortage instead of a customer shortage?
The most unusual detail of this story is the bottleneck. Startups usually struggle with sales, but Corma's problem is different. Ploda acknowledges that one of the hardest challenges is acquiring compute quickly enough. While the company adds capacity, customers wait. For founders, the picture is familiar: if a product is built on proprietary trained models, access to GPUs and the budget for them limit growth.
How it works and how the company sells itself?
Corma trains models to sift through audit logs, security events, and network traffic, looking for weak signals over long time spans. A simple example: a single suspicious night-time login means nothing. But if it is followed over weeks by rare database queries and unusual network traffic, it could be the start of an attack. Humans find such chains hard to spot, while models trained on large volumes of data handle them more easily.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.