Article · Cybersecurity

AI Funding Is Concentrated. The Next Large Rounds May Go to Security and Cryptography

Caleb Ngeno

By · November 18, 2026

Yellow AI funding and security cover

Abstract. Venture capital in 2026 has a clear shape. Crunchbase counted $159 billion in global venture funding in Q3, with $102 billion, or 64 percent, going to AI companies. PitchBook and the NVCA put AI at 82.7 percent of US deal value through September. KPMG counted $728.3 billion across the first nine months. The totals differ because the providers count different deals, but the direction is the same. Most new venture dollars are going to AI and the infrastructure around it. This article examines that concentration and makes a prediction: the next serious pool of capital will flow into AI-powered cybersecurity, agent control, crypto-agility, and post-quantum migration. AI will not replace the mathematics behind hashing and encryption. It will change how organizations find, configure, migrate, and prove those controls.

Keywords: AI funding, venture capital 2026, AI security funding, cybersecurity startups, crypto-agility, post-quantum cryptography, hashing, cyber safety, agent security

The funding story is easy to exaggerate. The useful version is more specific. AI now takes most of the money, the largest rounds are getting larger, and the infrastructure around models has become a venture category of its own. Security sits just behind that wave. It will not stay there for long.

The funding picture in November 2026

Crunchbase counted $159 billion in global venture funding in Q3 2026 across close to 6,000 funded startups. AI-driven companies took $102 billion of that, or 64 percent. The first nine months reached $679 billion, the highest first-three-quarter total in its records. A record 27 companies raised rounds of $1 billion or more, and those rounds took about a third of the quarter's capital. Late-stage funding was $105 billion, early-stage funding was $40.6 billion, and seed funding was $13 billion.

PitchBook and the NVCA reported a different US picture: $515.8 billion in deal value through September, with AI taking 82.7 percent. KPMG Venture Pulse put global funding at $728.3 billion through the first nine months and described AI megadeals as the dominant theme. Do not add these figures together. They use different databases, geographies, and cut-off dates. Read them as several views of the same shift.

Three facts survive those differences. AI takes most of the money. The biggest checks are getting bigger. Capital is moving from software into data centers, chips, energy, robotics, and defense.

What the money is buying

Frontier labs are the obvious destinations. Databricks and Safe Superintelligence each raised $5 billion in Q3, while Crusoe, Moonshot AI, Mistral AI, Nscale, and others raised more than $3 billion. These rounds buy compute, researchers, distribution, and time. For infrastructure companies, the data center and the power contract are part of the product.

Enterprise software is the second layer. Companies are funding copilots, workflow agents, search, customer service, coding, and analytics. The pitch is plain: if a task can be described in a ticket, an agent should draft the first pass.

Security is the third layer, and it is still small next to the attack surface. That is where the next wave of large checks is likely to land.

Why serious money will move into AI cybersecurity

Attackers already use AI to write convincing lures, find exposed systems, triage stolen data, and move faster after access. Defenders still buy tools that assume a human analyst will read a queue and click through a console. That gap is large enough to support a new funding cycle.

Dealroom's AI Security category recorded $2.15 billion across 119 rounds in January to August 2026, up from $932.6 million for all of 2025. Its ten-startup watchlist raised $1.68 billion in 2026 to mid-September. HiddenLayer raised $100 million, Zenity raised $125 million, and Onyx Security raised $113 million. Those are early signals of a larger budget line.

The work falls into a few jobs that buyers can understand:

  1. Continuous testing will replace the annual penetration test as the baseline. Startups such as Horizon3, Armadin, XBOW, and depthfirst run attacks against changing systems and turn the result into a fix, not a PDF.
  2. Agent security will become its own line item. Enterprises need to know which agents exist, what tools they can call, which data they can read, and what happens when one acts outside its remit.
  3. AI security operations will keep growing. Alert triage, investigation, and containment have measurable outputs: fewer false positives, shorter dwell time, and fewer repeat incidents.
  4. Model and supply-chain security will follow the models into production. Teams will pay for evaluation, runtime guards, artifact signing, and poisoned-data detection.
  5. Data security for AI will expand. A data-loss tool that ignores what an agent can retrieve is already behind the threat model.

The investors will follow the budgets. CISOs already buy exposure management, identity, detection, and response. AI gives those categories a reason to be re-bought. Vendors that show a measured reduction in exploitable paths or containment time will raise the large rounds.

The reinvention of cryptography and hashing

This prediction needs a boundary. AI models do not make AES, SHA-256, or Argon2id obsolete. Security proofs still come from mathematics, public review, and standards. What changes is the work around those primitives.

Most enterprises cannot answer where RSA, ECDSA, or SHA-1 still live. Keys sit in images, certificates, vendor appliances, scripts, and old services. AI-assisted discovery can read code, containers, infrastructure as code, certificates, and tickets to build a cryptographic bill of materials. That inventory becomes a product category, because an organization cannot migrate what it cannot see.

Migration is the second category. NIST finalized ML-KEM, ML-DSA, and SLH-DSA in 2024 and expects quantum-vulnerable algorithms to be removed from its standards by 2035. AI tools will generate migration patches, test hybrid handshakes, find broken certificate chains, replay traffic against new libraries, and explain which service still depends on an old algorithm. The standards stay human-led. The migration work becomes automated.

Implementation review is the third category. Models can scan for nonce reuse, weak randomness, hardcoded keys, wrong key lengths, and bad password-hashing parameters. They can propose fixes and write regression tests. Formal verification and human review still decide whether the fix ships.

Hashing takes on a new job in AI systems. A hash is a cheap way to pin a dataset version, model checkpoint, retrieved document, prompt template, or tool output. Signed manifests and content-addressed storage will matter more as agents consume content from email, tickets, websites, and other agents. The question changes from whether a file exists to whether it is the file the system expected.

Identity is the fifth category. An agent that can call tools is an identity. It needs short-lived certificates, hardware-backed keys, scoped permissions, signed tool calls, and a revocation path. Passkeys and workload identity become more useful when the workload is a model.

What cyber safety becomes

Safety moves from a training problem into an operating control. A model that can send email, move money, edit code, or query a production database needs an authorization boundary, an audit trail, and a kill switch. Accuracy scores alone will not satisfy a regulator or an insurer.

Expect safety cases to become normal for high-impact agents. A safety case should name the intended use, forbidden actions, evaluation results, red-team findings, residual risks, and the person accountable for the system. Security teams will run evaluations before release and continue them after release, because tools and prompts change.

This is where AI security and traditional security meet. The same control plane has to handle human users, service accounts, agents, models, keys, and certificates. The tools that win will give one answer across all of them.

Predictions for 2027 to 2029

  1. AI security rounds will grow faster than the wider cybersecurity market, with the sharpest growth in agent control and autonomous testing.
  2. A large security vendor will buy an agent-security company for more than $1 billion.
  3. Cryptographic inventory and a cryptographic bill of materials will appear in enterprise CI/CD pipelines and vendor questionnaires.
  4. Post-quantum migration will be funded as infrastructure work, with budgets for discovery, testing, and certificate lifecycle changes.
  5. Detection engineering will shift toward generating and validating the telemetry an attack would leave, so teams can prove coverage without running the attack.
  6. Buyers will demand outcome metrics such as exploited paths closed, mean time to contain, false positives removed, and migrations completed.

Practical checklist

For founders, pick one job and measure it. A wrapper around a model is easy to copy. A system that proves a reduction in exploitable weaknesses, containment time, or migration effort is harder to replace.

For CISOs, start with inventory. List the models, agents, tools, data stores, and cryptographic algorithms in use. Add policy for agent actions and key ownership. Run two pilots with measurable outcomes before buying a platform.

For investors, ask what the tool changes in the customer's environment. Round size and model access are entry signals. Retention, deployment depth, and evidence of risk reduction are the real tests.

What to take away

AI funding is huge and concentrated. The next large checks are likely to target the gaps that concentration creates: machine-speed attacks, agent actions, model supply chains, and crypto migration. The core cryptography will stay mathematical. The systems around it will be rebuilt with AI doing the inventory, testing, migration, and evidence work. That is where serious money is likely to go.

Sources

Caleb Ngeno

About the author

Caleb Ngeno

Lead Developer

Founder and lead developer with expertise in full-stack development and cloud solutions.

Share this article

Work with TechGaetano

Hardening a production environment?

TechGaetano helps teams design zero-trust controls, review identity and network boundaries, and turn security policy into engineering practice.

Talk to a security engineer

Frequently asked questions

What is AI Funding Is Concentrated. The Next Large Rounds May Go to Security and Cryptography?

Abstract. Venture capital in 2026 has a clear shape. Crunchbase counted $159 billion in global venture funding in Q3, with $102 billion, or 64 percent, going to AI companies. PitchBook and the NVCA put AI at 82.7 percent of US deal value through September. KPM.

How should a team get started with AI Funding Is Concentrated. The Next Large Rounds May Go to Security and Cryptography?

Start with asset and identity inventory, define the trust boundaries, and roll out controls in phases: identity, device posture, least-privilege access, segmentation, then continuous verification.

What should teams avoid when working on AI Funding Is Concentrated. The Next Large Rounds May Go to Security and Cryptography?

The main risks are treating zero trust as a product purchase, leaving standing privileges in place, and assuming the model or the network perimeter is a security boundary.

Further reading

Related posts