Wiz went from a standing start in March 2020 to a $32 billion acquisition agreement in under five years, reportedly the fastest any enterprise software company has scaled to that level of revenue. A lot of that momentum traces back to one thing: AI. Search for Wiz AI and you land on two overlapping stories. One is a product line covering AI security posture management. The other is how the platform itself uses machine learning to connect millions of cloud resources into something a human can actually reason about. Both matter if your company runs anything meaningful in AWS, Azure, or Google Cloud.
Wiz’s pitch, minus the marketing
Cloud security tools generally work one of two ways. Agent-based products install software on every VM and container, which gives deep runtime visibility but takes months to roll out across a large estate. Agentless scanners query cloud provider APIs instead, which is fast but often shallow.
Wiz is agentless, and the interesting part is what it does with the data it pulls. It walks every account, snapshots resources, including VMs, Kubernetes clusters, IAM roles, storage buckets, serverless functions and database snapshots, then builds a graph of how they relate. Wiz calls it the Security Graph. Instead of firing 4,000 disconnected alerts at an analyst, it answers a narrower question: is there a realistic path from the public internet, through a misconfigured load balancer, into a role that can read customer records?
That framing sounds like a minor difference. In practice it removes most of the noise. A publicly exposed storage bucket with nothing sensitive inside is a finding worth noting. The same bucket reachable from an identity that can assume a production admin role is an emergency.
Where the Wiz AI label actually comes from
Wiz does not sell a single SKU called Wiz AI. The name gets used as shorthand for AI-related capabilities spread across the platform: Wiz Cloud, Wiz Code, Wiz Defend, and a dedicated module called Wiz AI-SPM.
AI-SPM: mapping the AI nobody told you about
Shadow IT was already a headache. Shadow AI is worse, because AI workloads behave nothing like a normal web app. A data scientist can sign up for an API key on a Friday, export a customer dataset to fine-tune a model, and leave the results sitting in a bucket with public read access. Nobody filed a ticket.
Wiz AI-SPM hunts for that. In a typical enterprise scan it surfaces:
- Managed AI services in active use, including Amazon Bedrock, Vertex AI, Azure OpenAI and SageMaker, plus third-party endpoints from OpenAI and Anthropic.
- Self-hosted inference on GPU instances or Kubernetes, which frequently sits outside standard build pipelines.
- Long-lived API keys and tokens for AI providers hiding in environment variables, CI logs, or code repositories.
- Training datasets and model artefacts stored in buckets that are world-readable or missing encryption.
- Over-permissive roles that let a notebook job reach production data it has no business touching.
- Known vulnerabilities in model-serving frameworks such as unpatched Jupyter, Ollama, or vLLM deployments.
The real value sits in the graph rather than any single finding. Wiz can show that a particular notebook is reachable from the internet and can assume a role with write access to a regulated data store, a chain no standalone scanner catches on its own.
The assistant layer
There is a natural-language side too. Wiz Assistant lets an analyst type a question in plain English, something like which internet-facing workloads can reach production data, and returns an answer built from the same graph. It is the least glamorous use of AI in the product and probably the one teams touch most often.
AI is turning up in tools well outside security as well. Consumer hardware is filling up with it, from OpenAI’s rumoured smart speaker to assistants baked into laptops. Every new endpoint is another thing to inventory.
Why AI workloads broke the old playbook
Traditional cloud security posture tools assumed a fairly static estate. Things get provisioned, they live for years, they get patched. AI teams do the opposite. A GPU cluster spins up for a two-week training run and vanishes. A model gets pulled from Hugging Face and dropped into a container with no lockfile. Prompts and completions move data in ways no network firewall inspects.
None of that fits a change-approval process built for quarterly releases. Which is why plenty of security leaders spent 2024 in the same awkward position: a compliance dashboard showing green, and a genuine inability to say where the company’s training data actually lives.
The $32 billion question
Wiz was founded in March 2020 by four veterans of Microsoft’s cloud security group. It crossed $100 million in annual recurring revenue in roughly 18 months, raised a $1 billion round at a $12 billion valuation in May 2024, and by early 2025 said ARR had passed $500 million with a path to $1 billion.
Google first came knocking in mid-2024 with a reported $23 billion offer. Wiz declined and started preparing for an IPO instead. In March 2025 Google returned with $32 billion in cash, the largest acquisition in its history and one now subject to a lengthy regulatory review.
That number has drawn scepticism, and not only from outside the industry. Some investors have been openly warning about AI euphoria in venture capital, arguing that valuations have run well ahead of the revenue underneath them. Whether that criticism lands on cloud security specifically is debatable, since security budgets tend to hold up in downturns and Wiz was selling into an urgent, funded problem. But $32 billion is a statement about where buyers think AI infrastructure spend is heading, not just about one company’s current book of business.
How Wiz AI compares to the rest of the field
The competitive set is crowded. Palo Alto’s Prisma Cloud and Microsoft Defender for Cloud come bundled into platforms many enterprises already own. Orca Security takes a similar agentless, graph-first approach. CrowdStrike and Sysdig compete hard on runtime. A wave of startups is chasing AI-SPM specifically, and most of them are two years old.
Wiz’s differentiators are speed and coverage. A first full scan typically completes in well under an hour, versus weeks or months for an agent rollout across a large environment. That matters more than any feature checklist, because the honest reason most cloud security programmes quietly fail is that nobody finished deploying them.
If you are evaluating categories rather than vendors, it pays to stay skeptical of the whole shelf. A wider look at which AI tools genuinely earn their keep is a useful corrective before you sign a multi-year contract.
What to check before you commit
Run the pilot against a single business unit and see whether the tool can, inside a week, tell you where your AI workloads are, who can reach them, and what data they touch. If it cannot, the graph is not as good as the demo suggested.
Three things are worth pressing vendors on. First, data residency, since AI-SPM tools ingest metadata about your models, datasets, and identities. Ask exactly where that telemetry lands and whether it leaves your region. Second, licensing, because Wiz prices per workload and AI workloads multiply fast. A team running 20 instances last quarter might be running 300 today, and per-workload pricing has a way of surprising finance six months in. Third, an off-ramp.
Security teams are not the only group getting nervous about AI quietly becoming load-bearing infrastructure. Linux users have been asking for a kill switch over AI features shipping inside their operating system. The same instinct applies to your security stack. If you cannot turn the AI features off without losing the underlying detection, that is worth knowing during procurement rather than after.
Get clear answers to all three, and the rest of the decision is straightforward.