What does TSMC stand to gain by putting money into the AI labs? Quick catch-up before we dig in: OpenAI’s private funding round valued it at $1.25 trillion and had brought in about $11 billion at its first close. Anthropic then launched its Nasdaq roadshow with shares offered at $70 to $78, alongside a $26 billion private placement involving major chip and cloud companies. TSMC was already among those investors, tying the fundraising story directly to the companies supplying AI infrastructure. This is Lantern AI Daily Briefing. New York opens safety registration, and BotPsy puts an assistant on-device—how much can it handle? First, let’s follow TSMC’s money.
Here's Jeremy Hsu at Ars Technica:
The chipmaker’s board approved plans to invest up to $500 million in OpenAI preferred stock reserved for large investors and another $500 million to purchase Anthropic common shares at the latter’s expected IPO price, according to Reuters. Such an investment may give markets “a shot of confidence as the artificial-intelligence trade shows signs of strain,” The Wall Street Journal reported.
Anthropic’s IPO financing already had TSMC in the mix; now the board has approved up to $500 million there, plus another $500 million for OpenAI, according to Reuters. The company making the chips is buying a seat on the demand side, too. And those seats come with different paperwork: OpenAI preferred stock versus Anthropic common shares at its expected IPO price. Same headline dollar ceiling, different securities—don’t read that as an identical bet on both labs. The Journal reports analysts warning that these supplier-investor-customer loops make actual demand harder to judge. That’s where I land: TSMC’s investment appetite tells me very little about whether a software customer can afford another year of inference. I’d still give TSMC’s willingness to commit capital some weight. But the Journal’s potential confidence boost needs that asterisk: a chipmaker buying AI equity gives you a much murkier signal than customers paying their own bills.
From Aiden Reed at AI Tech Daily:
New York on Tuesday opened an online registration portal for large frontier AI developers covered by the state’s Responsible AI Safety and Education Act, known as the RAISE Act, Governor Kathy Hochul’s office said in a Nov. 10 release. Large frontier developers must register through the RAISE Portal by Jan. 1, 2027, the release said. The statute takes effect then.
New York's RAISE registration portal is open, with a January first deadline. But the coverage test needs both numbers: more than five hundred million dollars in annual revenue, counting affiliates, and a frontier model above ten to the twenty-sixth training operations. Revenue alone won't tell your engineering team whether they're covered. And the obligations split. Large frontier developers have to publish safety frameworks starting in January and submit quarterly catastrophic-risk assessments to DIGIT, inside the Department of Financial Services. I'd watch those public frameworks: OpenAI and Anthropic will have to put commitments somewhere people can actually compare them. Compare the incident reports, too. The state's release says critical-incident reporting and model transparency reports apply to all frontier developers, not just the large ones. And that compute definition includes later fine-tuning and material modifications—compliance needs somebody who understands the training pipeline, not just somebody filling out the portal.
Hugging Face, with Milan Gritta and colleagues:
BotPsy is a "personal assistant" that uses tools to perform a variety of simple tasks for the user, in domains such as web search & research, travel & lifestyle, commerce & finance, business & customer operations, device & vehicle control, files & documents, calendar, and email & messaging. Due to its small parameter count, it is best suited to tasks of low to medium difficulty.
Tether AI Research’s BotPsy squeezes tool calling into 800 million parameters, built on Qwen3.5-0.8B. For calendar edits and file chores, I’d happily test something that doesn’t need a cloud model on every request. And Tether says it beats every model up to 1.2 billion parameters that they tested. Keep those last three words attached. The useful claim here is a small, self-hosted assistant you can actually control. But their task list also includes finance and vehicle control. I’d start with calendar drafts, require confirmation before changes, and measure how often it picks the wrong tool on messy requests. Tether explicitly warns that BotPsy makes mistakes and isn’t intended for high-stakes use. Keep that warning beside the finance pitch—and remember, a local model can still send data out through web or messaging tools.
Have feedback, a story idea, or a correction? Email aidailybriefing at lantern podcasts dot com. We read every note and may use your ideas in a future briefing.
We’ll be watching the Jan. 1, 2027 deadline for large frontier AI developers to register through New York’s RAISE Portal. Links to every story are in the show notes, so take a look at the ones you’d like to explore further. That’s Lantern AI Daily Briefing for today—thanks for listening, and we’ll be back tomorrow. This is a Lantern Podcast.