JPMorgan Report: AI Shifts Software Profit Pools Toward Infrastructure

Written by: Rita

Over the past two weeks, a series of seemingly isolated events have occurred in the software industry: Starbucks built its own AI tool to replace Microsoft and IBM software, Meta is considering selling its cloud computing power for AI, and Microsoft is replacing OpenAI and Anthropic models with its own MAI model. JPMorgan Chase views these developments together as pointing in the same direction: the profit pool in the software industry is being reallocated. AI has fundamentally changed the scale and speed at which companies can build in-house solutions. To understand who will be most harmed and who will benefit most moving forward, it’s necessary to examine the chain reactions at each level.

Starbucks spends approximately $400 million annually on software, and its CTO is reviewing every contract and service, planning to replace some applications purchased from Microsoft (inventory tracking) and IBM (operations management) with internally developed AI-assisted tools, with some solutions potentially launching next year. JPMorgan views this as a concrete example of the "build vs. buy" and "software will be free" narratives. While building in-house is not a new trend, AI has significantly lowered the barrier to entry, giving companies greater motivation to develop their own solutions under cost pressures. The complexity and cost of maintaining internally developed software are often underestimated, but for a company the size of Starbucks, the overall accounting still makes it worthwhile.

Meta is developing a cloud infrastructure business called "Meta Compute," offering AI computing power and model access to external developers in a model similar to AWS Bedrock. The market's initial reaction was that Meta has excess computing capacity, but Morgan Stanley offered a contrary view: AI demand is so strong that Meta sees this business as worthwhile. Over the next 12 months, Meta plans to double its data center capacity from 7 GW to 14 GW. Morgan Stanley believes Meta’s sale of computing power will not reduce its existing contracts with NeoCloud providers such as CoreWeave and Nebius, as demand is simply too high.

DigitalOcean pre-announced its Q2 results, with remaining performance obligations (RPO) expected to exceed $800 million, more than ten times year-over-year growth, and the weighted average contract term extending from approximately 1.6 years to over three years. The company attributed this to multiple nine-figure annual customer commitments signed in Q2, with Q2 revenue growth projected at approximately 29%, up from 22% in Q1. J.P. Morgan highlighted two key points: AI inference now represents a significant portion of contracts, with AI workloads shifting toward inference; large, long-term contracts are transforming DigitalOcean’s customer base and business foundation, signaling maturity in the NeoCloud赛道.

Microsoft has begun replacing OpenAI and Anthropic models with its proprietary MAI model in products such as Excel and Outlook, with tens of thousands of AI prompts now running on MAI each week. Morgan Stanley believes this represents a profit margin lever for Microsoft, offering cost-reduction benefits both in the short and long term. More importantly, this signals a key industry shift: software companies are demonstrating that AI applications do not require the most advanced large models to function effectively, as the industry moves from a "model arms race" into a "cost optimization phase."

Cloudflare has launched a two-way bot paywall, enabling content owners to charge AI crawlers. The new default rules will take effect on September 15, automatically blocking training and proxy bots on pages containing ads. More than half of Cloudflare’s network requests now originate from AI agents. Morgan Stanley views this as a new revenue stream for Cloudflare; leveraging its extensive network that spans numerous websites, Cloudflare is positioning itself as a settlement layer between publishers and AI model providers.

Morgan Stanley’s true point is that these five factors collectively reveal an ongoing structural shift: AI profit pools are moving downstream from the model layer. Upstream model providers (OpenAI, Anthropic) face pressure to be replaced, as evidenced by Microsoft’s development of its own MAI. Midstream infrastructure providers (AWS, CoreWeave, DigitalOcean) continue to see strong demand, but growth rates and contract structures are evolving. Downstream enterprise customers (e.g., Starbucks) are gaining greater bargaining power, with a growing trend toward building in-house solutions instead of outsourcing. Do not equate AI beneficiaries solely with model providers. Models may become commoditized, while inference infrastructure and intermediary layers capable of collecting tolls from traffic (e.g., Cloudflare) may possess more durable pricing power. This is the signal Morgan Stanley has extracted from the noise: investment selection logic must follow the shifting profit pools.

Disclaimer

This article is a summary and interpretation by Chaoxiang Research of a third-party brokerage research report (J.P. Morgan, July 13, 2026). The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution, not those of Chaoxiang Research, nor do they constitute any investment advice.

The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.