Technology Sector
AI is entering a more competitive, cost-conscious, and politically complicated phase — and the companies that own their own stack are pulling ahead.
Meta has released Muse Image, its first proprietary AI image-generation model, aiming to pull ad revenue and subscription dollars away from rivals like OpenAI and Google (CNBC). The model — formerly codenamed Mango — powers both the consumer Meta AI app and Meta's Advantage Plus ad platform, which automates creative production for brands, a genuinely large prize given Meta's $160B-plus annual advertising business. It trails OpenAI's GPT Image 2 on internal benchmarks but beats Google's Nano Banana 2 on key editing tasks, which suggests Meta is competitive enough to reduce its costly reliance on third-party tools like Midjourney. The broader play here is diversification: Muse Image is the cornerstone of a new paid subscription tier, and a Muse Video model is already in the pipeline — Meta is clearly building a full-stack AI creative suite, not just chasing a trend.
Chinese AI models from DeepSeek and Z.ai are capturing a rapidly growing slice of U.S. enterprise usage, driven by price tags that are 60–90% cheaper than Anthropic and OpenAI equivalents — a cost gap that is now hard for budget-conscious engineering teams to ignore (CNBC). On OpenRouter, a developer gateway platform, the share of tokens processed via Chinese models has exceeded 30% every week since February 8, up from an average of just 11% over the prior 12 months. AI startup Lindy moved 100% of its traffic from Anthropic's Claude to DeepSeek in June, with its CEO telling CNBC the switch will save millions within months — a concrete data point that captures a broader pattern. Policy risk is real: the U.S. government is weighing export controls and restrictions on Chinese model adoption, which means this cost-driven shift is on a collision course with Washington.
Microsoft has begun routing a portion of user prompts in Word and Excel through its homemade MAI models rather than those from OpenAI and Anthropic, Bloomberg reported — a meaningful shift given that Office 365 (Microsoft's cloud productivity suite, used by hundreds of millions) was explicitly marketed as powered by those third-party models. The move is part of a broader cost-containment wave sweeping Silicon Valley: Amazon, Meta, Uber, and Accenture have all reportedly pulled back on AI spending after a period of aggressive token consumption sometimes called 'tokenmaxxing.' For OpenAI and Anthropic, whose valuations rest heavily on enterprise API revenue (fees charged per unit of AI output), losing even a slice of Microsoft's enormous internal demand is a material concern worth watching. The trend also signals that the AI infrastructure buildout is entering a more disciplined phase — where owning your own models is increasingly a competitive and financial imperative, not just a vanity project.