The artificial-intelligence boom has also caused a surge in new buzzwords for investors to learn, from "inference" to "agents" to "edge." Next up on the list: "Orchestration."

At its core, orchestration refers to coordinating tasks, data, and outputs across different AI models. For example, a company could use Anthropic's Claude to code a software tool and OpenAI's models to analyze the data needed to make that tool run. Orchestration streamlines the whole process into one workflow.

The word is taking on new importance as enterprises grow increasingly cautious with their AI budgets. And Microsoft has positioned itself to win the so-called orchestration layer.

Rewind a year and executives were pushing employees to maximize their use of AI, including frontier models from Anthropic, OpenAI, and Alphabet's Google. But each coding problem or chat prompt uses tokens -- another entry in the AI lexicon -- and therefore money. Some companies seemed to forget this.

Palantir Techologies CEO Alex Karp told CNBC last week that "every single enterprise" Palantir deals with is unhappy about the return on their tokens from the frontier AI labs.

Karp can be prone to hyperbole, but he has a point. Some enterprises are turning to cheaper alternatives from Chinese labs, which have open-source or open-weight models that allow companies to host the AI compute on their own servers or private cloud.

Data last month from AI infrastructure platform Vercel showed that lower-cost models from Chinese developer DeepSeek handled a surge in token volume starting in late April. Anthropic and OpenAI still claimed the bulk of dollar spending. Using multiple AI products is where orchestration comes in.

"The solution is becoming obvious -- companies will want to build a way to make sure they can switch AI models without disrupting their business, " wrote D.A. Davidson analyst Gil Luria in a weekend research note. "They also want to manage costs by routing each individual query and task to the least expensive model."

That equilibrium would be good for Microsoft. The company's AI helper, Copilot, can funnel tasks to different AI models depending on how much horsepower users need, Luria pointed out.

Beyond Copilot, Microsoft has staked a position as a "secure, model-agnostic toll for enterprises to access tokens," said Melius Research analyst Ben Reitzes in a research note Monday.

If frontier models dominate AI consumption, Microsoft's Azure cloud business can host them. If the open models gain popularity, Microsoft's Foundry AI platform can orchestrate them and Windows can run them locally. The Foundry catalog already has more than 11,000 models available for use, Reitzes noted.

Sure, an enterprise can triage which tasks go to which models by itself. And many large internet companies will build their own internal orchestration layers, Barclays predicted in a research note last month.

But down-market, companies are living in a world where costs are opaque and complex AI processes can chew threw tokens without proper oversight. Microsoft's pitch is that it can more effectively conduct these customers' AI orchestra.

If successful, the perhaps more pressing challenge for Microsoft is to show investors that consumption-based contracts focused on orchestration can offset any slowdown in its traditional per-user software model, Reitzes said.

Microsoft stock has tumbled 20% this year alongside other software stocks, with those losses concentrated in the first quarter. The company was a Barron's stock pick in March.

Amy Coleman, Chief People Officer at Microsoft, announced Tuesday that the company would lay off 4,800 people, or around 2.1% of its workforce.