Compute Wars Heat Up as Coding Tools Surge Past 90

The AI infrastructure market just revealed its true cost structure. Anthropic agreed to pay xAI $1.25 billion per month for compute capacity—a staggering $15 billion annualized commitment that exposes the capital intensity crushing even well-funded labs. Meanwhile, xAI itself burned $6.4 billion in 2025 while planning expansion, prompting a $2.8 billion natural gas turbine purchase over three years. These numbers explain why coding tools are surging. Developers need efficient ways to architect systems on platforms where compute margins matter more than innovation speed.

Higgsfield, GLM-4.7, and Roo Code each gained 45 points this week, all now sitting at 90. These tools serve a market suddenly obsessed with optimization. When monthly compute bills reach nine figures, engineering teams can't afford redundant code or inefficient model integration. The coding tools rising fastest are those promising tighter, more productive development cycles. Pollinations climbed 47 points to 87 despite being an image generation tool—suggesting broader infrastructure concern is bleeding across modalities as companies stress-test their tech stacks.

Anthropicís profitability claim adds crucial context. The company projects $10.9 billion revenue in Q2, more than doubling from prior quarters. That growth requires either improved unit economics or raw scaling. Smaller tools like Adzooma, which gained 43 points to 86, are benefiting from enterprises pruning their vendor lists. Marketing tools that integrate with AI coding pipelines become force multipliers when infrastructure costs dominate budgets.

OpenAI's claim about solving an 80-year geometry conjecture matters here too. If true, it signals that reasoning models produce valuable work beyond chat interfaces. That work requires careful engineering, better debugging, and integration frameworks—precisely what surging coding tools address. The math problem claim hasn't been independently verified, but the market is pricing it as credible, pushing technical builders toward platforms that can handle sophisticated inference chains.

SpaceX's IPO filing laid bare the xAI situation for first time. Revenue from compute sales to Anthropic creates a revenue floor but doesn't offset the burn. This structural problem—massive losses despite major customer contracts—ripples through the developer ecosystem. Teams building on top of these platforms are hedging with multiple tools, favoring those offering portability and efficiency. Momentum scores don't lie. Coding tools are winning because infrastructure costs are winning, and developers know it.