AI Circle Daily
https://aicircle.news/archive/2026-10-24?category=leaders&lang=en
Scan to read this page
OpenAI says it will stop training its AI models on the live internet, citing "significant risks" — just weeks after one of its own engineers said that approach wouldn't work.
Take: A public reversal of its own engineer's stance, and a signal that frontier labs are trading freshness for control. Offline training slows the data pipeline; OpenAI is buying safety optics at the cost of speed.
Sam Altman's OpenAI is reportedly closing a round of up to $30 billion at a valuation well above $1 trillion, while weighing when to go public.
Take: Raising private capital at a trillion-plus valuation means Altman is stockpiling compute and talent before an IPO. That resets the pricing anchor for every AI startup and squeezes second-tier labs.
On the SpaceX earnings call, Elon Musk said a merger with Tesla would be a good idea but warned it would be "overstating it" to assume the deal was certain, while flagging power and cooling as constraints.
Take: Dropping merger talk on an earnings call is Musk pitching vertical integration of compute, energy and manufacturing to public markets. Regulatory review is long, but the expectation alone levers up both valuations.
On his first visit to Kazakhstan, Jensen Huang pitched "sovereign AI," saying the country could transform its "energy into intelligence," and shrugged off bubble fears.
Take: Classic Huang world tour: convert each nation's energy and policy anxiety into GPU orders. The louder the bubble talk, the harder he pushes the demand story beyond the US to make the order book look diversified and structural.
Jensen Huang visited Kazakhstan for the first time, saying the country could become a major AI computing hub serving Europe, highlighting its AI ambitions and global business appeal.
Take: Central Asia just got put on the AI map on cheap power and geography. For Nvidia it's a new demand pool; for European hyperscalers it's a backup compute hinterland beyond their strained grids.
Turmoil hit the AI build-out this week as neoclouds came under pressure, credit risks mounted and Solidigm pulled its IPO, with stock slumps and wider CDS spreads setting the stage for hyperscaler earnings.
Take: The financing side of the compute arms race is cracking, and credit markets are judging second-tier clouds first. Hyperscalers may scoop up capacity cheaply — this earnings season will widen the divide.
A newly launched Meta stock class was partly engineered to let Mark Zuckerberg pay California's proposed billionaire tax, should it pass, without losing control of the company.
Take: Textbook founder-control engineering: turn a potential tax bill into a share-structure problem instead of selling stock. Other dual-class founders will copy it; regulators and shareholders will watch harder.
Jack Dorsey's Block announced Dessa, an AI model trained "from scratch on nothing but human conversation," and showed off human chats that are occasionally awkward and uncanny.
Take: Skipping the general-purpose model race to train purely on dialogue is a signal Block wants AI inside social products. Awkwardness is being sold as a feature — the bet is on "human-like," not "assistant-like."
With earnings in the rearview, the Tesla-SpaceX merger talk is dominating attention, though the companies face a long road of negotiations and regulatory reviews.
Take: One Musk line puts the merger narrative on front pages, but the real signal is markets pricing a space-plus-EV-plus-AI bundle. The longer the talks, the more the trade gets replayed.
A report casts Musk as "one of one," while Anthropic's Dario Amodei is called a visionary too — but pros say he isn't in the same class for market tolerance.
Take: The real question is how much premium investors will pay for a visionary CEO in an IPO window. Without Musk-style immunity, Amodei's Anthropic will have to sell revenue, not charisma.