Enterprise software costs used to be relatively easy to explain. A company bought a certain number of licenses, paid a predictable annual fee, and revisited the agreement at renewal. AI is changing the model somewhat the way cloud did.
Cloud computing shifted many IT costs from fixed infrastructure spending to a variable, consumption-based model. AI is doing something similar by moving businesses away from predictable per-seat licensing to a model based on tokens, API calls, model choice, compute, and agent activity. The effect is that spending grows continuously as employees and AI agents work, and a successful deployment can create a challenge very quickly: The more useful the AI becomes, the harder its cost may be to forecast or control.
The market’s response for cloud usage was FinOps and cost optimization solutions that have helped businesses manage their cloud usage and spend, like 1Password’s SaaS Manager, among others, providing visibility into software spending, licensing, application redundancy, and support for forecasting.
1Password is now extending its SaaS Manager with a new AI Spend and Consumption Management capability. It consolidates token consumption and spending data from Anthropic, Cursor, and OpenAI, with more vendors expected later. By connecting directly to vendor administrative APIs, 1Password says it can give IT and finance teams daily, token-level data rather than forcing them to piece together invoices, card transactions, spreadsheets, and separate provider dashboards. The system can break consumption down by vendor, model, team, and user, while supporting spending thresholds and alerts through email and Slack.
Though the announcement is broadly aimed at enterprise IT and finance teams, there’s a message here for MSPs, as well. AI cost governance is becoming another operational discipline that many customers will struggle to manage internally and, since MSPs already help clients control cloud costs, SaaS portfolios, licensing, security, and user access, AI optimization is the next logical extension. The opportunity is not simply to help customers spend less, but to help them understand whether their growing AI bills correspond to useful business activity.
It may not seem like a big deal at the moment, but there is some urgency if you consider Goldman Sachs Research forecast that AI agents could drive a 24-fold increase in token usage by 2030. Unlike a chatbot that handles one prompt at a time, an agent may perform numerous model calls, revisit context, use external tools, and coordinate several steps to complete a single assignment. That makes agentic workloads both powerful and potentially expensive.
Recent academic research into AI coding agents found that agentic tasks can consume significantly more tokens than ordinary coding chats, that repeated runs of the same task can vary by as much as 30 times, and that spending more tokens does not necessarily produce a better result. Given that, it’s not surprising the same study also found that frontier struggle to predict their own consumption accurately.
Many household name businesses are already responding to the issue. Uber spent its entire annual AI coding budget in just four months, prompting the company to place a monthly cap ion its AI coding tools. Meta and Walmart similarly capped its token usage, while Amazon is pushing its developers towards cost-effective models.
The point is not that companies should necessarily slow their AI adoption, but that adoption without visibility can create a hefty and unpredictable operating expense. Traditional procurement tools can tell finance what was paid last month, but they can’t explaining why AI usage is rising today, which model is responsible, or whether a prepaid token commitment will actually last until the end of the contract.
That’s precisely the visibility 1Password is trying to deliver by extending SaaS Manager’s existing view of application ownership, access, licensing, and spending to AI consumption.
That also means AI is becoming part of the managed software portfolio, which creates an opportunity for MSPs to add AI advisory and AI usage optimization to their service offerings.
Many SMB and mid-market clients will adopt several AI tools without creating a dedicated FinOps function. Individual departments may purchase subscriptions independently, developers may access models through APIs, and employees may use both sanctioned and unsanctioned tools. The resulting environment can combine shadow AI, duplicated capabilities, inconsistent security controls, and costs no single department fully understands.
An MSP could help bring those pieces together through a recurring AI governance and optimization service that might include consumption reporting, threshold configuration, budget forecasting, vendor and model comparisons, unused tool identification, and quarterly reviews linking usage to business outcomes. Instead of presenting a client with another technology dashboard, the MSP could answer useful questions, like: Which teams are driving the bill? Which workloads justify their cost? Are expensive models being used for tasks that cheaper ones could handle? Are agent deployments consuming more resources without producing better outcomes?
To be clear, 1Password has launched its AI Spend and Consumption Management as an extension of its enterprise SaaS Manager, not a multitenant solution for MSPs specifically. But, MSPs should look at is as an opportunity nonetheless because that certainly doesn’t prevent them from deploying it for their clients.
1Password does offer an MSP edition of its enterprise password manager with centralized client administration, usage visibility, and consumption-based billing, so it also wouldn’t be surprising to see a multitenant version of SaaS Manager in the future.
Regardless, the real takeaway from MSPs is they should treat AI consumption as a manageable operational category now. They should inventory the AI tools clients are using, determine which are seat-based and which are consumption-based, establish ownership for API and token budgets, and add cost alerts before agent deployments expand. Just as cloud adoption created demand for cloud cost optimization, widespread AI adoption is creating demand for token governance.
Edited by
Erik Linask
