Corma Raises $125M Series A at $1B Valuation to Scale Defensive AI
Corma raised a $125M Series A at a $1B valuation led by NEA to scale defensive-security AI, but says compute capacity — not model quality — is its real bottleneck.

Why it matters
- Corma raised a $125 million Series A at a $1 billion valuation on November 16, 2026, led by NEA, bringing total funding to $185 million.
- The company emerged from stealth on August 10, 2026 with $60 million in seed funding.
- Corma claims 99% of security tasks are completed autonomously, 100x faster end-to-end and at 95% lower cost per task.
- A Fortune 1000 CISO reported a 94% drop in end-to-end response time after replacing a former service provider with Corma.
- CEO Alon Pluda says compute capacity, not model quality, is the immediate constraint on serving demand and the waitlist.
Corma, a defensive-cybersecurity company building domain-specific foundation models, announced on November 16, 2026 that it raised a $125 million Series A at a $1 billion valuation. NEA led the round. Sequoia Capital, Khosla Ventures and Coatue continued their participation, bringing Corma's total funding to $185 million.
The round lands three months after the company emerged from stealth on August 10, 2026 with $60 million in seed funding. A billion-dollar valuation in a first priced round after a public launch is aggressive by any measure, and the investors are betting that defense is a distinct AI problem rather than a side effect of general-purpose model progress.
Corma said the new capital would accelerate scaling of its foundation models, expand compute capacity and support selective hiring. The company characterized demand from the world's largest organizations as overwhelming and identified securing enough compute for that demand and its waitlist as one of its central challenges. In plain words: the constraint right now is infrastructure, not model quality.
What Is Corma Actually Selling?
The company deploys its models through the Corma Workforce, a system designed to work autonomously alongside human security teams. It operates within customers' existing security and IT stacks. That includes identity, cloud, endpoint, security information and event management (SIEM, the log-aggregation layer most enterprise security operations run on), and firewall or intrusion-detection tools. Its work spans investigations, connecting evidence across those systems and taking action within them.
The design keeps humans in the loop. Employees can assign tasks through Slack or Teams, ask questions, review findings and approve actions. Corma positions this as AI-led execution layered into a customer's current operating environment, with human review built into the workflow rather than bolted on afterward.
Corma reports live deployments across healthcare, finance, energy, critical infrastructure and retail. The claimed performance numbers are steep:
- 99% of security tasks completed autonomously and successfully
- 100 times faster end-to-end task completion
- 95% lower cost per completed task
These figures come from the company itself, and the source material does not describe the methodology behind them. One anonymized Fortune 1000 services-company CISO, quoted on Corma's product page, said end-to-end response time fell 94% in a comparison with the customer's former service provider, which the customer subsequently replaced with Corma. A single named-by-role customer anecdote is thin evidence for system-wide claims; the anonymization limits how much weight it can carry.
Why Build Defense-Specific Models?
Corma's core technical argument is that recent gains in coding and software reasoning map more directly to offensive security than to defense. Vulnerability research and exploit development are code-reasoning problems aimed at bounded targets. Attackers need to find one path in. Defense, by contrast, requires continuous analysis of audit logs, events, configurations and network flows, with no bounded target and no finish line.
The company frames the defensive workload as connecting weak signals across systems and long time periods, while making thousands of accurate, consistent decisions each day under changing conditions. That is a different failure profile from writing an exploit. A model that hallucinates occasionally while coding can still be useful; a model that makes inconsistent triage decisions across thousands of daily security events is a liability.
That distinction drives Corma's focus on domain-specific foundation models rather than adapting general coding models. Its stated development program includes:
- Rethinking model architecture for the defensive workload
- Extending context windows to detect faint anomaly signals
- Adversarial self-play (models training against model-generated attacks)
- Continuous learning that can establish behavioral baselines over months and years
CEO Alon Pluda and the Corma team said the next generation of its foundation models would be shared soon, without giving a release date. "Soon" is doing heavy lifting in that statement; the financing supports the next phase of model development, but no benchmark results, external evaluations or release timeline accompany the claim.
What Are the Stakes?
The funding round sits at the intersection of two expensive bets. The first is that autonomous AI agents can reliably execute security operations work that currently consumes large human analyst teams. Security operations centers are chronically understaffed, and a system that completes tasks at 95% lower cost while keeping human approval gates would reshape enterprise security budgets.
The second bet is architectural. If Corma is right that defensive workloads need purpose-built models with months-long baselines and cross-system reasoning, the generalist labs' coding models will not simply absorb the security market. If it is wrong, a well-funded incumbent or foundation-model provider could close the gap. The company has not yet published evidence that settles this question either way, and the competitive landscape is not addressed in its announcement.
The near-term constraint is more mundane than the research agenda. Corma says broader deployments require enough additional compute capacity to serve current demand and the organizations on its waitlist. The Series A money, in part, buys GPUs and infrastructure before it buys breakthroughs. That detail is telling: it suggests the company believes its models are already good enough for the customers it cannot yet serve, and the market test will come when capacity opens up.
What Happens Next?
The sequence to watch is the next model release, then the waitlist conversion. The financing supports that next phase of model development and selective hiring, but Pluda's team has committed to no release date for the new foundation models. Until independent customers and external evaluations appear alongside the company's own deployment metrics, the 99% autonomy and 100x speed figures remain vendor-reported numbers. A billion-dollar valuation on $185 million raised means investors expect those numbers to hold up under scrutiny.
Source: unite.ai
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
722 articles
Related articles
- Modal Labs Nears $750M Round at $15.75 Billion Valuation
- Model ML raises $250M Series B at $1.5B valuation for finance AI agents
- Nebius Stock Down 31% in Three Days on $27 Billion Spending Plan
- Cohere Buys AISLE and Cubic, Extending Sovereign AI Into Cyber Defense
- Mercor Raises $750M at $20B Valuation to Expand AI Training