Two forces defined the AI world this week, and they pulled in opposite directions. At the top, AI tightened its grip on the very definition of "the market": Alphabet walked into the Dow Jones Industrial Average as the index brushed record highs. At the coalface, a small startup did something that would have been unthinkable a year ago — it tore out a leading American model and replaced it, wholesale, with a Chinese one, and bragged about the savings. Between those poles sat a chipmaker betting billions on AI leaving the data center for the physical world, an incumbent quietly slipping its flagship's launch, and a US state quietly defanging the country's most ambitious AI law on the very day it was meant to bite. Here is the through-line.
Alphabet walks into the Dow
The clearest sign of AI's hold on the economy is no longer a funding round — it is an index reshuffle. Effective Monday, Alphabet replaced Verizon in the Dow Jones Industrial Average, with S&P Dow Jones Indices saying the change would "broaden and strengthen" the benchmark's exposure to artificial intelligence, cloud computing and digital advertising. Alphabet's shares rose roughly 4% on their first day as a component, and the Dow itself traded near record highs above 52,000. The mechanics matter: a Magnificent Seven name swapping in for a low-priced telecom means every fund that tracks the Dow must now own an AI leader, and the index that has stood in for "the market" for over a century is being rebuilt, quietly, around the AI economy.
A startup dumps Claude for DeepSeek
While the index economy embraced AI, one founder put a number on its costs. Lindy chief executive Flo Crivello announced he had switched 100% of the agent startup's traffic from Anthropic's Claude to China's DeepSeek V4, writing that the move "saves us millions of dollars and we're actually seeing an increase in performance on many core use cases." On the Artificial Analysis Intelligence Index, DeepSeek's V4-Pro reportedly ran the full benchmark for about $1,071 against roughly $4,811 for Claude Opus 4.7 — more than four times cheaper. It was not a quick swap: the migration took six to nine months of evaluation and prompt re-engineering, and Lindy runs the model through US-based Atlas Cloud to keep data on American soil. The episode captures the industry's turn from "tokenmaxxing" to efficiency: when an open Chinese model is multiples cheaper and, by one customer's account, no worse, the frontier labs' pricing power is suddenly the thing in question. Crivello says he would switch back the day Anthropic cuts prices — for a 25-person company where AI had outgrown payroll, he framed it as "a matter of survival."
A chipmaker bets $7 billion on "Physical AI"
The next chip-consolidation wave is forming away from the GPU. onsemi agreed to acquire Synaptics in an all-stock deal worth about $7 billion — a roughly 19% premium — bolting Synaptics' Edge AI compute, human-machine interface and wireless connectivity onto onsemi's power and sensing franchise. The pitch is "Physical AI": intelligence embedded in cars, factories, robots and devices, where sensing has to be fused with on-device compute rather than shipped to a cloud. onsemi says the combination could expand its addressable market by some $30 billion to around $243 billion by 2030. It is one of the first large semiconductor deals organized explicitly around AI in the physical world, not the training cluster — a tell about where the edge of the market is heading.
Google's flagship slips to July
For all the momentum elsewhere, the incumbent stumbled on timing. Google pushed Gemini 3.5 Pro — its long-context, "Deep Think" frontier model — from June to a July launch, with reporting saying the company is still refining coding, token efficiency and long-task performance while the model sits in limited preview for select enterprise customers. The delay lands amid a run of senior-researcher departures that has dented confidence in Google's roadmap. Frontier release schedules have themselves become competitive signals: a slip at the top hands rivals room, and turns July into a test of whether Google can ship on time and on benchmark.
America's first state AI law is hollowed out
Regulation, meanwhile, moved against the grain. Colorado's AI Act — the first comprehensive US state AI law — was supposed to take effect on June 30. Instead, a rewrite signed in May delayed it to January 1, 2027 and stripped out its risk-based core: the duty of care against algorithmic discrimination, mandatory risk-management programs, impact assessments and reporting to the attorney general are gone, leaving a narrower disclosure-and-transparency regime for automated employment decisions. The two-year fight ended with a watered-down law and little fanfare. As the EU begins handing out AI Act fines, the contrast is stark: the most ambitious US state framework was defanged on the very day it was meant to bite, signaling a domestic drift toward light-touch rules and federal preemption.
Put together, the picture is an industry being absorbed into the financial mainstream while its economics get ruthlessly re-examined from below. Watch three things from here: whether Gemini 3.5 Pro actually ships in July and how it benchmarks; whether more named US firms follow Lindy onto US-hosted Chinese models, and whether the frontier labs answer with price cuts; and whether onsemi-Synaptics opens a broader edge-silicon consolidation wave. Signal, not advice; no live prices.