This is the AI Affairs briefing for Friday 6 November. Read by a synthetic voice.
Nvidia has completed a one-billion-dollar investment in Naver, the South Korean internet company, as the two pursue an AI factory venture. Yonhap reports that Nvidia paid Naver today for a four-point-one-three percent stake. Naver issued more than six-and-a-half million new shares through a private placement and received one-point-three trillion won, according to a regulatory filing described by the news agency.
The payment completes a transaction Naver had scheduled for today. The stake acquisition is expected to accelerate the AI factory venture the companies announced in June. Nvidia chief executive Jensen Huang and Naver founder and chairman Lee Hae-jin attended a strategic agreement ceremony at Nvidia’s California office in July, Yonhap reported. Naver has also signed a nine-billion-dollar deal with Brookfield for the factory, while KB Kookmin Bank is set to lead a ten-point-seven-trillion-won loan for Naver’s AI project.
Naver plans to start building graphics-processor-based infrastructure at its Gak Sejong data centre and expand its GPU-as-a-service business. The completed share purchase gives Nvidia a direct stake in Naver as those infrastructure plans move forward.
Microsoft is winding down an arrangement under which it rented AI computing capacity from Oracle. Bloomberg reports that Microsoft let the contract lapse and has been switching off the rented capacity over recent months, citing people familiar with the matter. The decision comes as Microsoft reassesses cloud partnerships involving billions of dollars.
There is an important distinction between this contract and earlier reports about a possible Microsoft-Oracle lease worth more than three billion dollars. That figure concerned talks about a separate potential deal, not the value of the rental agreement Microsoft is now ending. Oracle said details in reports about those failed talks were inaccurate. In the current case, Bloomberg’s sources say Microsoft had already been using the capacity and has been turning it off after choosing not to renew the contract.
Microsoft has also been examining other AI infrastructure arrangements. AI Affairs previously reported tighter Microsoft oversight of neocloud projects and its decision to decline an Nscale expansion in Texas. Those involved a different provider and project. Oracle, meanwhile, has filed a plan to raise twenty billion dollars through bonds and shares for its AI expansion. That financing plan is separate from Microsoft’s decision on the existing rental contract; Bloomberg’s report does not describe an outcome for every partnership Microsoft is reviewing.
OpenAI says it will keep investing in AI data centres despite voter opposition to their construction in this week’s US midterm elections. Chief financial officer Sarah Friar told CNBC that the company needs to continue investing to meet customer demand, and that its approach relies on relationships it has already built with hyperscalers — the large cloud providers that supply computing capacity.
Friar said the infrastructure approach “will continue to mean that we raise capital”. The election results have raised concerns elsewhere in the industry, too: Meta warned that US data centre projects could become harder after the vote. OpenAI has separately pledged to pay data centre power costs and seek local input. Friar’s message today was that the company still intends to fund and build capacity to serve its customers, even as opposition to new facilities becomes part of the discussion.
Enterprise spending on AI tokens slowed sharply in late summer, according to data from the spending platform Ramp reported by TechCrunch. Spending grew just one-point-five percent across August and September, compared with forty-two percent month-on-month growth in July. Token volumes still rose. In other words, the spending figures and the amount of model use did not move together.
Ramp lead economist Ara Kharazian told TechCrunch that spending by the top one percent of firms slipped below the summer peak of ten thousand dollars per employee per month. The gap between buyers remains wide: Ramp’s data puts the median firm at twelve to fifteen dollars per employee per month, and the top ten percent at six hundred to one thousand dollars.
Prices are part of the picture. OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna in September. Anthropic said in a filing that price cuts for Fable 5.1 and Opus 5.5 reduced its revenue growth by about six hundred million dollars, TechCrunch reported. The spending data arrives as Anthropic prepares for an initial public offering. Bloomberg reported that its potential valuation had fallen from an estimated two trillion dollars in early October to between one-point-two and one-point-three trillion dollars. Ramp also recorded an eight-point-nine percent rise in token spending from late September to late October, so its figures do not describe an uninterrupted decline.
TSMC is considering making advanced memory chips, including three-dimensional DRAM. Finimize, citing The Information, reports that the contract chipmaker has begun mapping out production plans for the newer type of memory, and that Nvidia has requested it.
TSMC is best known for making logic chips designed by companies including Apple and Nvidia. Memory would take it into a part of the semiconductor business that Finimize describes as more cyclical than its established logic-chip work. The prospective move follows a shortage of memory that has pushed prices higher. Finimize reports that insufficient DRAM can prevent complete AI systems from shipping even when processors are available.
AI Affairs has previously reported two other developments involving memory for AI: Samsung’s plan to use TSMC’s three-nanometre process for Nvidia memory chips, and SK hynix’s work on memory designs for AI inference beyond high-bandwidth memory. Today’s report concerns a possible additional manufacturing role for TSMC, not a confirmed decision to produce three-dimensional DRAM.
The Trump administration has postponed today’s scheduled meeting of the President’s Council of Advisers on Science and Technology. An Energy Department spokesperson confirmed the postponement, according to CNA. The council is an outside panel advising the White House on US science and technology policy; its members include Nvidia chief executive Jensen Huang, Meta chief executive Mark Zuckerberg, AMD’s Lisa Su and Andreessen Horowitz co-founder Marc Andreessen.
The meeting was due to take place after technology shares sold off yesterday amid concerns that a new Congress could slow AI development and that heavy spending on AI infrastructure might not produce sufficient returns. President Donald Trump has promised not to limit AI development, while executives have urged caution. CNA reported no immediate White House response to its request for comment on why the meeting was postponed. No reason for the change was given in the account.
Two United Nations bodies have issued a governance toolbox for cities considering AI in public services. The International Telecommunication Union, the UN agency for digital technologies, and UN-Habitat announced the United for Smart Sustainable Cities toolbox at the Smart City Expo World Congress in Barcelona. ITU calls it the UN’s first system-wide AI governance toolbox for smart cities. It is aimed at mayors, city managers and policymakers — including those who may decide that AI is not the right tool for a particular service.
The guidance gives officials four rules. Start with the urban problem, rather than a technology to deploy. Keep public decisions under meaningful human control. Make safeguards stronger as an application’s potential impact grows. And treat a decision not to use AI as a responsible possible outcome. Across a project’s life, ITU says, officials should ask whom a system might help or harm, how it will be monitored, who is accountable for decisions made with it, and when to change course or stop.
ITU says city halls already use chatbots to answer residents and algorithms to select roads for repair. It also points to software used in administrative decisions, where bias, privacy, human rights and accountability are concerns. UN-Habitat executive director Anacláudia Rossbach emphasised communities in informal settlements and other underserved areas. She called for cities to address data gaps, protect human rights and give residents and civil society a meaningful role in shaping urban decisions.
The release marks ten years of the United for Smart Sustainable Cities initiative. ITU and UN-Habitat are also releasing guidance on subjects including urban AI governance, measuring people-centred digital transformation and security for digital public infrastructure. The central question in the toolbox is one for officials before deployment: whether a proposed system addresses an identified city problem, and what it could mean for the people receiving that service.
Finally, Alphabet, Google’s parent company, built a stake worth nearly five hundred and forty-eight million dollars in Qualcomm during the third quarter, Reuters reported. A filing of Alphabet’s investments put the holding at nearly three million Qualcomm shares at the end of September. The investment comes as Qualcomm, long associated with smartphone chips, pursues AI processors for data centres.
Qualcomm expects annual data-centre chip revenue to reach fifteen billion dollars by 2029. That is a target, not revenue it has already recorded. Reuters reports that the company has spent the past year approaching cloud providers with custom AI processors as they develop their own chips and seek alternatives to Nvidia’s products. Qualcomm introduced its AI200 and AI250 data-centre AI chips last year. It has also signed a custom AI chip agreement with ByteDance, while Microsoft and Meta have placed orders for Qualcomm data-centre processors.
A September agreement permits Amazon to buy up to sixty billion dollars of Qualcomm AI data-centre chips and related products, and includes warrants allowing Amazon to buy Qualcomm stock. The sixty-billion-dollar figure is a ceiling on potential purchases, not a statement of completed sales. Alphabet and Qualcomm already work together: Qualcomm processors are used in phones running Google’s Android system, and the companies have agreed to develop chips and software for automotive AI applications. Google also designs its own data-centre AI chips, known as TPUs.
Qualcomm shares had fallen more than nine percent yesterday after its quarterly results and amid a wider market sell-off. They rose three percent in trading before the US market opened today following disclosure of Alphabet’s stake, according to Reuters. The filing also showed Alphabet increasing its SpaceX holding to about five hundred and fifty-five million shares.
Today’s stories put capital commitments alongside decisions about which capacity to keep, build or use. Nvidia’s Naver payment is complete; Microsoft is ending a rental contract; and OpenAI says it will keep raising capital for infrastructure. In cities, the new UN guidance asks a different question before an AI project begins: whether it is needed at all.