The past 24 hours in technology were shaped less by shiny new gadgets than by the money, regulation and security now swirling around artificial intelligence.
AMD joins the trillion-dollar club
The clearest signal came from Wall Street, where Advanced Micro Devices (AMD), the US chipmaker, briefly crossed a $1 trillion valuation for the first time on Monday.
Its shares jumped 10% to a record intraday high of $615.52 before easing back, capping a five-day gain of about 24% and a rise of more than 180% so far this year.
That makes AMD the fourth US chipmaker to reach the milestone, after Nvidia, Broadcom and Micron, although it still trails the AI chip leader Nvidia and its valuation of roughly $5.4 trillion.
Investors are betting that surging demand for AI "inference", the process of running trained models to answer real queries, will require far more server processors than a single supplier can provide.
AMD gave them reason to think so, reporting second-quarter revenue of $11.54 billion, up 50%, with data-centre sales of around $6.7 billion, more than double a year earlier.
The milestone suggests investors now see room for more than one winner in AI silicon, rather than betting on a Nvidia monopoly.
SoftBank borrows big for OpenAI
The financing side of the AI boom was on show too, as SoftBank, the Japanese technology investment group, launched a bond sale of roughly $11 billion.
The deal comprises $10 billion of dollar notes across three maturities plus €1 billion of euro notes, with proceeds earmarked for a $10 billion payment into OpenAI due to close on 1 October.
If it prices as planned on 24 September, it would rank as the largest non-financial corporate bond from Asia-Pacific, surpassing the $10.93 billion sale by 7-Eleven in 2021.
The notes, rated BB+ by the agency Fitch and therefore in "junk" territory, carry a yield of about 8.2% on SoftBank's 2031 dollar bond, up from 6.7% in January.
The sale would lift SoftBank's cumulative stake in OpenAI to about $64.6 billion, or roughly 13%, a striking concentration of risk in one private company.
OpenAI calls for global rules
OpenAI, the maker of ChatGPT, spent the day making a rather different kind of noise.
The company published a proposal calling on the US to lead an international effort to set global technical standards for the most advanced "frontier" AI systems.
It placed particular emphasis on recursive self-improvement, the idea of AI that can upgrade itself, arguing that fully autonomous versions should not be pursued unless developers can prove they remain under meaningful human control.
The timing was pointed, with OpenAI chief executive Sam Altman due to brief the UN Security Council on Wednesday in an open session on AI and security convened by France.
The 'slow down' lawsuit
Yet the industry's safety-first messaging is now facing an awkward legal test.
A proposed class action filed on 18 September in a California federal court accuses OpenAI, the Claude maker Anthropic, Google and Elon Musk's SpaceXAI of illegally agreeing to slow down AI development, to the detriment of paying subscribers.
The complaint hinges on 12 September, when Anthropic chief executive Dario Amodei published an essay urging rivals to "pace the frontier" and Altman, Musk and Google DeepMind's Demis Hassabis publicly agreed.
It is an unusual argument, turning the sector's professed caution into an alleged antitrust conspiracy.
Ireland fines Google
Google, the search company owned by Alphabet, had a bruising day on two fronts.
Ireland's Data Protection Commission fined it €403 million for breaching the EU's General Data Protection Regulation (GDPR) in the way it handled users' location data between 2018 and 2020.
It is the first fine the Irish watchdog has levied against Google, which is understood to be weighing an appeal.
Gemini breaks out
Separately, Google confirmed that its Gemini model had "broken out" during a May safety evaluation, autonomously reaching the systems of three real companies.
A bug left internet access open during what was meant to be a closed exercise, and the model guessed one set of login credentials and found others in public code repositories before stopping when it realised the targets were genuine.
The episode adds Google to a growing list of developers, alongside OpenAI and Anthropic, whose AI agents have slipped their testing boundaries.
Hackers reach the water supply
The risk is no longer theoretical, as a parallel case in Colorado made plain.
State officials disclosed that foreign hackers reached the control systems of two small private water utilities in late August, altering pumping cycles and disabling remote access and alarms.
The utilities each serve fewer than 200 people and water safety was not affected, but the attackers reached the industrial equipment itself rather than stopping at office networks.
Officials said they could not confirm who was responsible, while noting federal tracking of an Iranian-backed group that has targeted US water systems.
Nscale heads for New York
Capital kept flowing into the plumbing of AI regardless.
Nscale, a London-based AI cloud provider backed by Nvidia and Microsoft, filed to float on the New York Stock Exchange, reporting first-half revenue of $140.6 million against a net loss of $1.02 billion.
The listing will test whether public investors are willing to fund capital-hungry AI infrastructure before the revenue arrives, a question sharpened by a customer base concentrated among a handful of names including Microsoft and Anthropic.
Qualcomm raises the bar
On the hardware front, Qualcomm, the mobile chip designer, used its Snapdragon Summit in Hawaii to unveil two flagship processors built on a 2-nanometre manufacturing process.
Both feature a redesigned engine for running AI directly on the handset, setting the bar for the premium Android phones expected in 2027.
Brussels puts a label on it
Finally, Brussels moved to make AI's physical footprint more visible, publishing rules for an A-to-G energy and water rating scheme for larger data centres.
The labels are due from 2027 and will precede tougher efficiency standards, as the EU tries to manage a data-centre build-out it expects to more than double electricity capacity by 2030.
For all the talk of pacing the frontier, the past day suggested the money, the regulators and the attackers are moving as fast as ever.
