# Starbucks is quietly building the software it used to buy from IBM and Microsoft > Source: > Published: 2026-07-12 13:28:38+00:00 *Starbucks is using AI-assisted coding to replace software it buys from IBM and Microsoft. Wall Street noticed because the risk is bigger than one coffee chain.* Word got out this week that Starbucks is trying to build in-house replacements for enterprise software it currently licenses, and the market reaction was quick. According to Bloomberg, which reviewed an internal presentation, the company is developing its own alternatives to a Microsoft inventory system and an IBM platform used to manage equipment maintenance. Some of that software could roll out by the end of 2027, if it passes testing. That is real money. Chief Technology Officer Anand Varadarajan told employees there are clear opportunities to reduce Starbucks' software spend, according to Bloomberg. The internal materials put that spend at roughly $400 million a year. For a company still working through Brian Niccol's turnaround, cutting even a piece of that bill is not a side project. Wall Street noticed. IBM shares fell about 3% in premarket trading, ServiceNow dropped roughly 3.5%, and Salesforce slid close to 4%, according to reporting from Investing.com and TipRanks. Microsoft barely moved. That split matters. Investors were not punishing every vendor with a Starbucks contract. They were punishing companies whose pitch rests on selling big enterprises expensive workflow software that those customers may now believe they can build for themselves. ## The vendor bill is now negotiable One retailer swapping out an inventory tool is not a crisis for IBM or Salesforce by itself. Even with more than 40,000 stores worldwide and roughly $37 billion in fiscal 2025 revenue, based on public company figures, Starbucks alone can't dent either company. The risk is copycat behavior. If AI coding tools let Starbucks rebuild a maintenance platform that used to require IBM's expertise, you can expect other large companies to ask why their own contracts are still priced as if the old development costs never changed. That is the moat concern. Enterprise software vendors have long charged for complexity that customers could not cheaply reproduce. Inventory, maintenance, service routing and customer records are not glamorous systems, but they sit inside the daily work of large companies. If a retailer can use AI-assisted development to bring even some of that work inside, the vendor's strongest argument gets weaker. ServiceNow and Salesforce are especially exposed to this story because so much of their business sits around workflow and customer-management software that large clients already bend into company-specific shapes. IBM's maintenance platform is a cleaner example. It is narrow. It is operational. It has a clear cost attached to it. If you run Starbucks, you do not have to believe every vendor system is replaceable to see why this one is worth testing. ## The prototype is not the hard part None of this means Starbucks will pull it off cleanly. Building software that works in a slide deck is easy compared with running inventory and maintenance systems across thousands of stores, different markets, different suppliers and different failure points. The tools still have to work. Starbucks itself has flagged that the rollout depends on testing, and that caveat deserves to stay in the story. Equipment maintenance is a good example because it sounds boring until it breaks. A down espresso machine is not a software abstraction to the store manager standing in front of a morning line. If the system routes the wrong repair, misses a recurring fault or delays a service call, the savings on vendor fees can disappear inside slower operations. That sounds obvious. It is also exactly why IBM and other enterprise vendors have been able to charge for years of accumulated know-how. Still, Starbucks is trying. Frankly, that is the part software vendors should take seriously. A few years ago, a non-tech company openly planning to replace vendor tools with internal systems would have sounded expensive, risky and mostly theatrical. Not any more. Generative AI has changed the internal calculation enough that a coffee company's CTO can tell employees the software bill is up for review. 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. That's the bar. Once that happens, procurement teams get a sharper question to ask. Not whether IBM, Salesforce, ServiceNow or Microsoft are useful. They are. The question is what part of the invoice still belongs to the vendor, and what part now belongs to old assumptions about how hard software was to build. Whether other retailers follow depends on the next 18 months. If Starbucks' inventory and maintenance tools work reliably by the 2027 target, expect other large companies to put the same question to their own engineering teams: what exactly are we paying for. **Also read:** [The UN's AI Summit Showed Robots Racing While Governments Argue Over Rules](https://startupfortune.com/the-uns-ai-summit-showed-robots-racing-while-governments-argue-over-rules/) • [China's Net-and-Cable Rocket Catch Reignites the Race Against SpaceX](https://startupfortune.com/chinas-net-and-cable-rocket-catch-reignites-the-race-against-spacex/) • [Researchers hacked a quantum neural network on real trapped-ion hardware and gutted its accuracy](https://startupfortune.com/researchers-hacked-a-quantum-neural-network-on-real-trapped-ion-hardware-and-gutted-its-accuracy/)