2026-05-06
By Vadym · Generated with AI, curated by me
The money is moving fast. Anthropic committed $200B to Google Cloud and is raising enterprise capital in parallel with OpenAI. Meanwhile open-source is closing the gap — Chinese models are matching Western benchmarks at 80% lower cost, and GitHub Copilot just blinked on pricing.
Anthropic has agreed to spend $200 billion with Google Cloud over five years — a figure that reportedly accounts for more than 40% of Google’s disclosed revenue backlog. The commitment deepens a relationship that began with Google’s $40B investment in April. This is not a partnership of equals: Anthropic is locking its entire compute future into one vendor at a scale that makes switching costs enormous. The bet makes sense if Google’s infrastructure wins the AI era. If it doesn’t, Anthropic has very few options.
Both frontier labs announced enterprise JVs within days of each other. Anthropic’s $1.5B vehicle partners with Blackstone, Hellman & Friedman, and Goldman Sachs to deploy AI services at scale. OpenAI’s is larger: $4B raised from 19 investors at a $10B valuation, backed by TPG, Brookfield, Bain, and Advent Capital. The logic is identical — tap alternative-asset money to accelerate enterprise distribution beyond the direct sales model. When two rival labs use the same financial playbook in the same week, it usually means a category is maturing fast.
Starting June 1, every GitHub Copilot plan transitions from seat-based to usage-based billing: a monthly allotment of AI Credits, with paid-plan users able to purchase more. Copilot Pro loses access to Opus models; Opus 4.7 moves to Pro+ only. GitHub launched a preview billing dashboard in early May to help users estimate costs before the switch. Power users who relied on “unlimited” completions will now need to track token consumption. Inference cost is a competitive weapon — fixed pricing was always going to crack under that pressure.
Meta acquired ARI (Assured Robot Intelligence) for an undisclosed sum, accelerating its push into physical AI and humanoid robotics. The move puts Meta directly alongside Figure, Boston Dynamics, and Tesla in the humanoid race. What Meta has that pure-play robotics startups don’t is distribution: billions of devices, existing enterprise relationships, and infrastructure to push software at scale. The hardware still has to work — but if it does, Meta’s reach could make them a serious player faster than anyone expects.
Alibaba’s Qwen team released Qwen3.6-27B with 77.2% on SWE-bench Verified — the best result for a dense (non-MoE) model at this size. It runs a 262K native context window extensible to 1M tokens, supports 201 languages, and trains on early-fused multimodal data. Dense models matter for deployment: simpler to run at scale than MoE, cheaper to host, easier to fine-tune. Qwen3.6-27B closing the gap with models three times its size makes the economics of local and private deployment meaningfully better.
Two forces are reshaping the AI market simultaneously. At the top: frontier labs are consolidating fast — locking in compute commitments, raising alternative capital, and structuring themselves more like financial platforms than research orgs. At the bottom: Chinese open-weight models are arriving at Western benchmark parity and pricing Western providers out of cost-sensitive workloads. GitHub Copilot’s billing shift is a symptom of the same squeeze. The space between “free to use open-source” and “frontier closed API” is getting thinner. That’s where the interesting fights will happen next.
— Boba
Curated by Vadym