
Hola JARGONESE
Welcome back, friend.
Lindy just moved 100% of its production traffic off Claude and onto DeepSeek V4, and CEO Flo Crivello says the cost curve "crashed to the ground." Three things worth knowing before the weekend.
Hot take
The model layer is getting commoditized in real time. Whoever owns the routing logic that decides which model handles which prompt, not whoever ships the smartest model, is about to own the margin in this market.
TODAY IN AI
AMD locks in Anthropic for 2 GW and $5B in equity

Anthropic committed to AMD's Helios racks (MI455X, EPYC Venice, ROCm) for H1 2027 capacity, the second major lab deal AMD landed in a week after Microsoft Azure. AMD now powers 5 hyperscalers and labs on Helios, and the vendor-customer relationship is turning into a joint venture.
Chinese open models now serve 30-46% of OpenRouter's tokens

That is up from 4.5% in H1 2025, per CNBC, with DeepSeek, GLM 5.2, and Qwen running 60-90% cheaper than Anthropic or OpenAI. Vercel saw GLM 5.2 adoption grow 27x in token volume and 80x in customers in its first week alone.
OpenAI ships ChatGPT Work, an agentic operating system for teams

Powered by GPT-5.6, it connects 1,400+ plugins including Slack, Salesforce, and Jira, plans multi-step work in "Plan mode," and runs recurring tasks on schedule. Zapier reported recovering a 7-figure pipeline per month using it
🛠️IN THE KNOW:
💰 VC funding hits a record $510B in H1 2026: AI ate 70%+ of Q2 capital, and OpenAI plus Anthropic alone pulled in $217B, 43% of all funding.
⚖️ EU AI Act enforcement goes live: The AI Office and AI Board are now operational, with fines up to €35M or 7% of turnover for prohibited AI.
🏗️ Global AI infra commitments now top $1T: Reuters tallied deals including OpenAI-Oracle at $300B and Stargate at $500B, all multi-year capacity reservations, not purchase orders.
🧠 Thinking Machines drops Inkling: A 975B MoE, 41B active-parameter open-weight model positioned as the first Western answer to the Chinese open ecosystem.
💵 GPT-5.6 launches in three tiers: Sol, Terra, and Luna price at $5, $2.50, and $1 per million input tokens, with Sol running at 750 tokens per second on Cerebras.
📊 Nvidia still holds 95%+ of the data center GPU market: AMD has 5 design wins now, but the crossover point is still years out.
Whenever you're ready:
Enjoy Jargonese?
→ Forward this to a friend who needs their AI game upgraded
→ Share on LinkedIn | Twitter
→ Leave feedback
P.S : Keep an eye on Lindy. A startup moving 100% off Claude to DeepSeek and calling it a cost curve "crashing to the ground" is the kind of quote that ages into a case study either way.
Until next time — Sid & the Jargonese team
