The AI Credit Trap: Rethinking the Economics of the Modern SOC from Credits to Outcomes

90% of VMware users in a recent survey named higher licensing costs as a reason to migrate. The same report can't tell you how many have actually left. That gap is the story.

The Rimini Street 2026 IT Virtualization Survey, fielded by Unisphere Research, found that 40% of primary decision-makers named operational complexity as the barrier to changing course, and nearly as many flagged multi-vendor management. Its three-year migration figures describe intentions. Two more caveats: only 31% of respondents were primary virtualization decision-makers, and the study didn't disclose response rates.

Virtualization and security operations are different markets. Nobody should read a hypervisor survey as a forecast for SOC tooling. But the business lesson travels well: when a vendor changes how it charges, customers re-run the math on everything around the license, and the hardest costs to escape are the ones baked into daily operations.

Security AI is now going through its own commercial-model shift. CrowdStrike meters Charlotte AI in credits, and Microsoft meters Security Copilot in Security Compute Units, and last week's PHP2Shell disclosure showed what those meters face under load. The question for every MSSP and reseller is simple to ask and harder to answer: when every investigation has a meter on it, what does a resolved incident actually cost you?

The AI credit dilemma

CrowdStrike's Charlotte AI and Microsoft Security Copilot have both moved well past chat assistants. Charlotte now spans agentic detection triage, agentic response and AgentWorks, its no-code agent builder. Security Copilot ships agents inside Defender, Entra, Intune and Purview, including phishing and alert triage, plus an Agent Builder for custom agents.

The capability gap between them matters less to your P&L than the way each one is metered.

Pricing note for partners: every vendor dollar figure in this article is an estimated end-user price (est.), before partner or distributor discounts. Microsoft's figures come from its public list prices, CrowdStrike's per-credit figure is our estimate, and Sevii's reflects its current MSRP.

CrowdStrike Charlotte AI credits. Per CrowdStrike's licensing FAQ (updated September 28, 2026), one monthly credit pool covers chat, embedded features, Fusion SOAR agentic workflows, AgentWorks, agentic triage and agentic response. Since June 2026, customers pick a tier from 250 to 1,000,000 credits a month, and extra credits come in packs of 350.

The pool resets on the first of the month and unused credits don't carry over. Consumption is fractional, a chat costs less than an AgentWorks agent run, and the per-task rate shows in the Falcon UI rather than in a public price list. When the cap is hit, credit-powered features, agents and workflows stop running until the next refresh unless you license more.

There's an on-ramp: qualifying Falcon customers get 50 free credits a month, and Next-Gen SIEM paid ingest can add 30 to 300 more.

Microsoft Security Copilot pricing. Security Copilot runs on Security Compute Units (SCUs). Bought standalone, Microsoft's pricing page lists $4 per provisioned SCU per hour (est. end-user price) with a 1-SCU minimum, plus $6 per overage SCU (est.) above that, up to a limit you set. Its billing example puts one prompt at 3 SCUs and a Defender incident summary at 0.5.

Since November 2025, Microsoft has been folding Security Copilot into Microsoft 365 E5 and E7: 400 SCUs a month per 1,000 paid users, capped at 10,000, with no rollover. Per that FAQ (updated June 19, 2026), usage past the allocation will be throttled, with pay-as-you-go at an est. $6 per SCU offered later on 30 days' notice.

Run that at MSSP scale. A 250-seat E5 customer gets 100 SCUs a month, about 33 prompts or 200 incident summaries. A 250-seat customer on any non-E5 plan needs at least one provisioned SCU around the clock, roughly $2,920 a month (est. end-user price), or about $11.70 per endpoint before a single investigation runs.

Read the two side by side. CrowdStrike's hard cap protects the budget and puts capacity at risk; Microsoft publishes a unit price, but its included allocation throttles today and bills by the SCU once pay-as-you-go arrives. Both hand the forecasting job to whoever holds the contract.

For an MSSP, that's you, tenant by tenant. CrowdStrike pools credits at the customer ID level and Security Copilot capacity is tenant-wide, so 20 customers can mean 20 separate forecasts, each with its own month-end cliff.

From AI answers to security outcomes

A faster verdict and a fixed endpoint are two different deliverables. Your customer pays for the second one.

"AI in the SOC" now covers at least six layers of work, and they land on different lines of your P&L. The first three shrink the minutes an analyst spends per alert. The last three remove the remediation hours, the 2 a.m. callout and the customer bridge call, which is where most MSSP labor cost actually sits.

Capability claims are vendor-documented; scope varies by license tier, so verify per deal.

Two things follow for a service provider. Layers 1 to 3 make your analysts faster, so their value shows up as headcount you didn't add. Layers 4 to 6 change what you can promise in an SLA, so their value shows up as a price you can charge.

Both matter. They're priced and sold differently, and a credit model meters layers 3 to 5 by the task.

The MSSP margin equation

Cost per credit, cost per endpoint and cost per query are all inputs. The number that tells you whether a managed security service is healthy is cost per resolved incident, and it moves with volume.

To show how, we modeled a hypothetical MSSP for 12 months. Vendor prices are marked as estimated end-user prices (est.), and the MSSP's own costs and sell price are hypothetical. Partner pricing sits below end-user pricing, so real margins should run higher in every model; read the gaps between models rather than the absolute numbers.

The book of business. 20 customers, 250 endpoints each (5,000 total), sold at $15 per endpoint per month ($75,000 MRR). EDR/XDR licensing costs the MSSP $5 per endpoint in every scenario, so the platform choice stays neutral. Analysts cost $50 an hour fully loaded, and fixed SOC overhead runs $6,000 a month.

The workload. 0.15 investigations per endpoint per month (750 in total), 8% of which are true positives that need remediation (60 a month). Manual triage takes 25 minutes per investigation, our own planning assumption.

The four operating models.

• A. Human-led: EDR plus analysts. 25 minutes per investigation, 110 minutes of remediation and customer communication per true positive.
• B. AI-assisted on credits: 2 credits per AI verdict, 8 minutes of analyst review, 80 minutes per true positive.
• B2. Agentic on credits: heavier autonomy, 5 credits per investigation, 4 minutes of review, 40 minutes per true positive.
• C. Sevii fixed per-asset ADR plus human oversight: est. $25 per node per year (about $2.08 per month), paid upfront, then 3 minutes of spot-checks per investigation, 25 minutes of validation and communication per true positive, plus 10% QA time.
• What does one investigation cost on our telemetry, at our alert mix?
• What happens on the day we hit the cap: stop, throttle or bill?
• Can credits pool across our tenants, or is each customer its own forecast?
• What exactly counts as an asset, and which modules does the per-asset price cover?
• Which actions run without approval, and where's the audit record?

"Credits" below means any metered AI unit, Charlotte credits or Security Copilot SCUs. Because a credit and an SCU buy different amounts of work, we normalize both to a credit-equivalent at est. $1.50, bought per tenant at the cheapest committed tier (250, 500, 1,000 or 2,500) plus 350-unit packs, with no rollover.

In a perfectly steady month, the agentic credit model edges ahead. B2 leads Sevii by about two and a half points of margin, but only because it uses about 80% of the credits it committed to; in B, the MSSP uses only about a third. Breakage, the credits you paid for and never used, is the quiet cost of the credit trap.

Digital Logistix model, October 2026. Vendor prices are estimated end-user prices: Sevii est. $25 per node per year; metered AI est. $1.50 per credit-equivalent. MSSP costs hypothetical.

The picture changes when the queue grows. Investigation volume is the variable you control least: a new ransomware campaign, a noisy integration or one fast-growing customer can double it inside a quarter.

Credit spend steps up as tenants cross tiers, and every model with meaningful human time per incident loses margin with each extra alert. Sevii's fixed fee bends least because almost nothing in it scales with volume. Human-led delivery crosses break-even just past 2x volume.

At an est. $25 per node per year, Sevii becomes the lowest-cost model once investigation volume runs about a quarter above baseline. At double the baseline, it holds roughly 9 more points of margin than the best credit model, and if your real credit price runs about est. $1.90 or higher, Sevii leads even in a steady month. Put your own alert data and credit quote into the model to see where your book sits.

The PHP2Shell stress test

PHP2Shell is what a volume spike looks like outside a spreadsheet. On October 20, 2026, details of two pre-authentication remote code execution flaws in the PHP core leaked ahead of coordinated disclosure, and PHP shipped fixes the next day in versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35.

Both flaws, CVE-2026-109618 and CVE-2026-109619, score 9.2 on CVSS v4.0, and researchers from Wiz, OpenAI and Calif are credited with the discovery. Within days, watchTowr reported in-the-wild exploitation, Arctic Wolf Labs logged likely weaponized attempts, and PHP disclosed a third, configuration-dependent flaw, CVE-2026-115560.

For an MSSP, that week turned every tenant with a PHP web server into an exposure question, a patch job and a hunt for web-server processes spawning shells, all at once. Run it through the model as a single month at 3x investigation volume:

Over 12 months, one surge month leaves agentic credits and Sevii within about a point of each other, and a second one puts Sevii ahead. The surge also decides something the annual totals hide: whether you know your bill and your capacity before the incident starts. Here's how each model answers that during a PHP2Shell week.

That's the case for Sevii in one table. With Microsoft, a big incident turns into a cost you can't forecast; with CrowdStrike, it turns into a credit cap you hit, or a pack you buy, mid-response. With Sevii, you pay upfront, before the incident, and that one payment covers investigation and autonomous remediation together.

The platforms responded too. CrowdStrike published PHP2Shell protection guidance for Falcon customers. Sevii released a response paper describing its Autonomous Preemptive Security module ingesting the intelligence, mapping each customer's exposure, patching affected assets and running follow-up hunts within minutes, with no token charges.

A different operational model: remediation priced per asset

Sevii takes the opposite commercial bet. Its Autonomous Defense & Remediation (ADR) platform is sold at what the company describes as a fixed price per asset protected, and its sales materials define assets as endpoints, identities and cloud assets, with no surcharges on AI tokens, cases or events.

What it does. Sevii launched ADR in July 2025 around agentic "Cyber Warrior" agents that process detections, run escalated and retrospective hunts, isolate assets and remediate across endpoint, identity and cloud. It's agentless, deployed through APIs and MCP rather than software on customer assets, and each case produces a full case file: timeline, evidence, remediation steps and the code used to remediate.

How it's governed. Autonomy runs inside customer-defined policy that Sevii calls the Guards Matrix, backed by a real-time policy engine and a kill switch. Sevii says every action is logged with its trigger, the rule that allowed it, the exact commands run and the before-and-after state. Humans in the loop are optional and set per action, so an MSSP decides exactly where approval stays mandatory.

Where it sits in the stack. Sevii runs on top of the EDR your customer already owns, with native integrations for CrowdStrike Falcon and Microsoft Defender, plus Google Chronicle and Splunk. That covers both platforms in this article, so the Falcon or Defender investment stays in place and Sevii works as the remediation layer above it.

The managed version. In July 2026, DigitalERA and Sevii launched Managed Autonomous Defense & Remediation (M-ADR), running Sevii inside a 24x7 SOC. Remediation runs autonomously by default, analyst triage is an option the provider sets per policy, and the release says the service is in production for multiple customers.

Disclosure: DigitalERA and Digital Logistix are both DigitalEra Group companies, and Digital Logistix distributes Sevii.

What Sevii brings to the table. Sevii put its TCO typically 40 to 60% below traditional vendor MDR, with real-world examples of 30-day Fortune 500 reporting coverage for 2.1 million assets, an 8-minute-32-second average time to shed a threat and 81% of cases within SLO. Sevii doesn't post pricing publicly; our model uses an est. end-user price of $25 per node per year, and partners can get current pricing through DLX deal desk.

None of this makes Charlotte AI or Security Copilot weaker products. Both ship real agentic capability, and both vendors run their own managed services in Falcon Complete and Defender Experts for XDR. The difference is who carries the volume risk: under credits and SCUs, you do; under a fixed per-asset fee, the vendor does.

Strategic implications for resellers and MSSPs

The winning move for most partners is a hybrid, built on the EDR the customer already runs. Keep the platform AI that's bundled or nearly free for analyst assistance, and put the high-volume remediation work on whichever model your own numbers favor.

Five moves to make before your next renewal cycle:

1. Measure cost per resolved incident, per tenant, every month. Pull 90 days of alert and investigation counts first. Without that baseline, every pricing model looks good in the demo.
2. Decide who owns overage, and write it into the MSA. Pass credit overage through, absorb it, or cap it. Credit and SCU planning can become its own advisory line item, which turns a cost risk into revenue.
3. Sell the outcome you can guarantee. A verdict SLA and a remediation SLA are different products at different prices. Tier your managed packages on mean time to remediate, and price the top tier where governed automation makes the promise safe to keep.
4. Plan for multi-tenant reality. CrowdStrike pools credits per customer ID, and each customer's Security Copilot allocation lives in its own tenant. Map which customers are on E5 and which would need provisioned SCUs before you quote.
5. Put the freed hours where customers notice them. Analyst time freed from triage is worth the most in tuning, hunting and quarterly customer reviews, which is where retention gets won.

And ask every vendor the same five questions:

Predictable price, ready to remediate

The VMware story is still being written, and it already shows how far a licensing change travels: into staffing, tooling and the vendor relationships around the license. Security AI pricing travels the same way, and PHP2Shell showed where it hurts most, in the middle of an incident.

Here's the choice in plain terms. With Microsoft Security Copilot, a big incident makes your cost unpredictable. With CrowdStrike Charlotte AI, it puts you up against a credit cap, so you either pause AI work or pay for more credits mid-response.

With Sevii, you pay a fixed per-asset price upfront, so the incident is already paid for, and the same fee covers investigation and governed autonomous remediation, on top of the Falcon or Defender stack your customer already owns. In a perfectly steady month, agentic credits can still edge it out; once volume moves, and in a PHP2Shell week it moves fast, the fixed fee pulls ahead. Bring your alert data, and we'll show you where your book sits.

Run the numbers with us. Digital Logistix works with resellers and MSSPs from Santo Domingo to Bogotá to Bermuda on exactly this math. Our team is ready to walk you through how Sevii's autonomous remediation could sit on top of the Falcon and Defender deployments you already manage.