Anthropic has expanded its partnership with Google and Broadcom to secure multiple gigawatts of next-generation compute, deepening ties built on Google Cloud's TPU capacity and Broadcom's custom silicon work. The vast majority of the new capacity will be sited in the United States, extending the company's November 2025 pledge to invest $50 billion in domestic computing infrastructure.
The scale of the buildout tracks a demand curve Anthropic describes as accelerating: run-rate revenue has surpassed $30 billion, up from roughly $9 billion at the end of 2025, while the count of business customers spending over $1M annually has more than doubled — from 500 to over 1,000 — in less than sixty days.
Claude already runs across a deliberately diversified hardware stack — AWS Trainium, Google TPUs, and NVIDIA GPUs — letting Anthropic route workloads to whichever chip suits them best, a hedge against single-vendor supply risk as the frontier-model compute race intensifies.
Ode with Anthropic has gone public with its name: a $1.5 billion AI implementation company built as a joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, formed around the acquired startup Fractional AI. It currently employs 100 engineers working directly with Anthropic's applied AI team.
"It's pretty easy to imagine this as a trillion-dollar company someday if we execute well," CEO Chris Taylor told TechCrunch — framing enterprise AI adoption, not raw model capability, as the real bottleneck labs now need to solve for.
While Washington and the labs fought over who gets access to frontier models, developers quietly moved on. Chinese open-weight releases from Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai now fill OpenRouter's top six spots — with Claude Opus 4.7 trailing in seventh place — while open models absorbed nearly a third of AI requests on Vercel in June.
Hugging Face CEO Clem Delangue argues the shift reflects a deeper reluctance to "outsource core capabilities to a black box API" — a trend that could push frontier models toward a narrower, high-value niche while commodity workloads run on cheaper, ownable alternatives.
In an unexpected Sunday blog post, Microsoft CEO Satya Nadella joined VCs and Palantir's Alex Karp in warning that frontier labs risk becoming competitors to their own customers by learning from enterprise usage data. His fix: companies should build their own "proprietary learning environments" and "orchestration layers" to stay model-agnostic.
Nadella's argument cuts both ways — Anthropic itself accused Chinese open models in February of scraping millions of Claude prompts to train rivals, underscoring how contested the boundaries around model distillation have become industry-wide.
Emergent, the Bengaluru-and-San Francisco AI coding platform aimed at entrepreneurs and SMBs, closed a $130 million Series C led by Creaegis at a $1.5 billion post-money valuation — a five-fold jump in just six months, with Khosla Ventures, SoftBank Vision Fund 2, Lightspeed, and Y Combinator returning.
The company reports a $120M annualized revenue run rate and 200,000+ paying customers building shipping software, factory workflows, ERP tools, and property-management systems — competing directly with Lovable, Replit, and Cursor in an increasingly crowded field.