Enterprise Software Insourcing: The New Default, Not a Side Bet
Enterprise software insourcing is the shift in which large companies replace external vendor platforms with internal software development, using their own teams and AI-assisted tools to cut vendor cost, gain greater control over core systems, and reduce long-term dependency on third‑party licenses while tailoring technology more closely to their operational needs. That shift is no longer experimental; it is becoming a defining strategy for large enterprises. Microsoft is relying more heavily on its in-house MAI models rather than third-party AI from OpenAI and Anthropic, while Starbucks is trying to build in-house replacements for enterprise software it currently licenses. These are early signals that the build vs buy strategy has flipped: the default assumption is no longer "we must buy" but "we should build if the economics and risk pencil out."
Microsoft’s MAI Gambit: Vendor Cost Reduction Starts at the Core
Microsoft’s decision to route more user prompts in Excel and Word through its own MAI models instead of external systems from OpenAI and Anthropic is a textbook case of vendor cost reduction. The company has started using these MAI models for a portion of Office 365 workloads, marking a clear pivot away from paying outsiders to power everyday productivity features. The timing is not accidental. The move reflects a broader industry trend toward cost reduction following a period of heavy AI spending earlier in the year. Rising AI service costs have become contentious enough that some firms are reportedly turning to cheaper Chinese models for agentic tasks, despite security concerns. In that environment, building internal software development capacity is less a nice-to-have and more a defensive move: own the models, control the spend. Microsoft’s seven new MAI models, including an agentic coder and text-to-image generator, are less a product announcement than a statement of intent.
Starbucks: A USD 400 Million Question for Workflow Vendors
If Microsoft’s insourcing is about strategic control, Starbucks’ push is about the invoice. The coffee chain is developing its own alternatives to a Microsoft inventory system and an IBM platform that manages equipment maintenance. Internal materials put its software spend at roughly USD 400 million (approx. RM1,840,000,000) a year. According to internal presentations reviewed by reporters, Chief Technology Officer Anand Varadarajan told employees there are clear opportunities to reduce that software bill. That message rattled markets: IBM shares fell about 3% in premarket trading, ServiceNow dropped roughly 3.5%, and Salesforce slid close to 4% when word got out. Starbucks is not big enough on its own—with more than 40,000 stores worldwide and roughly USD 37 billion (approx. RM170,200,000,000) in fiscal 2025 revenue—to dent those vendors’ top lines. The risk is copycat behavior. If a retailer can use AI-assisted development to bring even some of that work inside, the vendor’s strongest argument—that complexity is too hard to reproduce—gets weaker.
Build vs Buy Has Changed Because Generative AI Changed the Math
For years, enterprise software vendors charged for the accumulated know-how baked into inventory, maintenance and workflow systems that sat inside the daily work of large companies. Prototype replacements were easy to sketch but hard to run across thousands of stores, markets and suppliers. Equipment maintenance is a sharp example: a down espresso machine is not a software abstraction to the store manager standing in front of a morning line. If internal tools route the wrong repair or delay service, savings disappear inside slower operations. Yet generative AI has changed the internal calculation enough that a coffee company’s CTO can openly tell employees the software bill is up for review. Enterprise software insourcing does not require replacing everything. Here is the thing: Starbucks does not have to replace every outside platform for this to matter. Just one or two expensive systems, rebuilt well enough in-house. Once that bar is met, procurement teams will ask a sharper question: what part of the invoice reflects current value and what part reflects old assumptions about how hard software was to build.
What Comes Next: Partial Independence, Customization and Pressure on Vendors
The next 18 months are a stress test for enterprise software insourcing. Starbucks’ internal inventory and maintenance tools could roll out by the end of 2027, if they pass testing. Starbucks itself has flagged that the rollout depends on testing, and that caveat matters; building software that works in a slide deck is much easier than running operational systems across thousands of locations. Whether other retailers follow depends on how reliably those tools work by the 2027 target. If they succeed, expect more large companies to ask their engineering teams a direct question: what exactly are we paying for. On the AI side, companies including Amazon, Uber, Meta and Accenture have reportedly taken similar steps to limit AI expenditure, reinforcing the trend. The likely outcome is not a world without vendors but a world where internal software development covers core, customized workflows and third-party tools must earn their keep with clear, differentiated value instead of locked-in licensing dependencies.
