According to CNBC reporting published Monday morning, Microsoft will use its Build conference keynote on June 2 to launch a new, lower priced coding model inside GitHub Copilot. The move is the clearest signal yet that the hyperscalers have decided coding is the single most strategic AI workload, and that they are willing to subsidise it to recapture share from Anthropic's Claude Code and OpenAI's Codex.
The market they are fighting over is substantial and growing fast. Mordor Intelligence pegs the AI code tools segment at $9.3 billion in 2026 and projects it will reach roughly $30 billion by 2031, a 26 percent annual growth rate. Tomasz Tunguz of Theory Ventures argues AI could eventually represent 30 to 60 percent of total R&D spending inside large software organisations. Those numbers explain why the cap ex commitments are so aggressive and why none of the hyperscalers can afford to cede the segment.
Google moved first at its recent developer event. The company unveiled Antigravity 2.0, a multi agent orchestration layer that runs parallel agents on a single task, plus Gemini 3.5 Flash positioned as "frontier performance for agents and coding." A new developer subscription tier at $100 per month is the visible price weapon. Sundar Pichai admitted publicly that Google had fallen behind on agentic coding, instruction following, and long horizon tasks. Google has now raised Antigravity's rate limits twice in response to developer complaints and absorbed the cost rather than throttle adoption.
Anthropic's counter is on capability rather than price. The just shipped Claude Opus 4.8 now defaults to a one million token context window, which is roughly 750,000 words of code and documentation in a single prompt. That depth matters for the kind of multi file refactors and migration work that drive enterprise spend. Claude Code remains, by most measures we hear from buyers, the strongest pure coding agent on the market. The question is whether that lead is worth a two or three times price premium over a Microsoft or Google alternative that is good enough for most tickets.
Microsoft's positioning at Build will be telling. The company has already shifted Copilot to usage based pricing to manage rising costs, and it has the unique advantage of owning both GitHub and the underlying Azure infrastructure. Michael Turrin at Wells Fargo warned that Microsoft "is going to have to give some specific use cases beyond just cost for that audience." The keynote needs to show concrete agentic workflows, not just a cheaper sticker.
The strategic logic behind the price war goes beyond seat revenue. Ken Parmelee at Forrester captured it cleanly: "Anytime you ask any of these tools, build this thing for me, they are burning tokens. This is the new gateway drug. It is a way to hook people into their other products." Every coding session generates demand for inference compute, storage, observability, and eventually production deployment. Whoever owns the IDE owns the funnel into the cloud. That is why Google paid $2.4 billion to license Windsurf's technology last year and hired its CEO Varun Mohan.
For CTOs and VPs of Engineering, the practical implications are immediate. First, do not sign multi year coding tool contracts this quarter. The price floor is moving daily and the buyer side has the negotiating advantage. Second, invest in a thin abstraction layer so your developers can switch backends without rewriting their workflow. MongoDB's CJ Desai told CNBC his team buys these tools one year at a time precisely to preserve that optionality. Snowflake reports that its most productive engineers spend up to $50,000 per year each on AI tokens, so the unit economics of the choice are no longer trivial.
The operator angle for digital commerce teams at MediaMarktSaturn or REWE digital scale is to push procurement to run a real bake off in June. Take a representative migration task, your worst legacy checkout module, and run it through Claude Opus 4.8, Codex, Antigravity 2.0, and whatever Microsoft announces tomorrow. Measure passing tests, time to merge, and total token cost. The results will likely surprise the team that has been defaulting to one vendor on muscle memory.
Risks to watch include vendor lock in through proprietary agent frameworks that make migration painful even with a price advantage, security and IP issues around training data that vary widely between vendors, and the simple chance that the cheapest model produces subtly worse code that surfaces as bugs months later. Microsoft's Build keynote at 9:30 a.m. Pacific is the next data point.
A final note on market structure. Cursor remains the wild card. The company grew from $4 million to $2 billion in annual recurring revenue in eighteen months with just 300 employees, and SpaceX holds an option to acquire it for $60 billion later this year. If that deal closes, the developer tooling layer suddenly belongs to a company with no interest in selling cloud capacity to anyone, which would force Microsoft, Google, and AWS to bid even harder for the remaining independent surface area. Watch for any signal at Build that Microsoft has anticipated this scenario, either by deepening its GitHub moat or by quietly preparing its own acquisition response.
