[Digital Today reporter Chi-gyu Hwang (황치규)] Concerns about excessive investment in AI infrastructure swept through the tech sector. It was a week when declines in AI-related stocks stood out. Apple, which has invested relatively less in AI, contrasted with the trend by reaching a $5 trillion market value after Nvidia.

ㆍAI infrastructure stocks stumble...Apple, which invested less in AI, tops $5 trillion market value

AI data centres are in short supply now, but some observers say that, over the long term, excess facilities could deliver a major shock to the stock market. There are also signs investors are turning their attention to safer U.S. Treasury investments amid AI bubble worries.

ㆍ"AI data centres are in short supply now...over the long term, concerns grow about an overinvestment shock" ㆍMoney flows into U.S. Treasuries on AI bubble fears

There is also talk that revenue growth at Anthropic, which is pursuing an IPO, is slowing. Anthropic posted third-quarter revenue of $16.8 billion, but growth has slowed compared with earlier this year. Anthropic received a 'BB+' credit rating, one notch below investment grade, from global credit rating agency Kroll Bond Rating Agency (KBRA).

ㆍAnthropic revenue growth slows...did profits also fall after blocking AI imitation? ㆍAnthropic gets 'BB+' from KBRA, nearing investment grade

Tech companies moved to actively rebut the bubble argument. The AWS CFO said current AI spending is sustainable amid concerns about an AI investment bubble. Dell Technologies CEO Michael Dell (마이클 델) said the problem is not demand but a shortage of memory chips. He said enough memory chips are needed to deliver ordered servers on time, but supply is short. On next year's outlook, he also said, based on the current situation, "it will get worse."

ㆍAWS CFO: "AI spending is sustainable...server investment can be recovered within 3 years" ㆍ[Ditup People] "The AI bottleneck is not demand but memory...it will get worse next year"

A trend is also emerging of U.S. companies actively jumping into the open-weight AI model market, long led by Chinese companies. With big tech firms Nvidia and Meta releasing open-weight models, attention is on what changes this could bring to industry dynamics.

Meta released an open-weight model, 'Muse Spark Open', with 816 billion parameters. Meta also teamed up with Apple, with which it had an awkward relationship, as it rolled out the open-weight AI model Muse Spark Open.

ㆍU.S. big tech targets open-weight...Meta launches a major push after Nvidia ㆍMeta unveils frontier AI model 'Muse Wonder', targeting Anthropic and OpenAI ㆍMeta forms local AI alliance with Apple, once an uneasy partner...runs Muse Spark Open on Macs ㆍNvidia releases open model 'Nemotron 3.5' for AI agents, targeting everything from servers to the edge

FriendliAI, a South Korean AI inference cloud company, and open-source AI development platform Ollama will hold an event on Nov. 4 and 5 to meet major South Korean companies and developers and share ways to adopt and use open models in practice.

ㆍFriendliAI and Ollama join hands to expand open-model adoption by South Korean companies

Below is a roundup of other moves by companies in South Korea and abroad around AI.

Microsoft unveiled a new AI model, 'MAI-Base-2', and stressed it is stronger than OpenAI's and was developed using a quarter of the computing resources.

ㆍMicrosoft AI chief: "Developed an AI model stronger than GPT-5.6 with a quarter of the computing power"

AWS introduced an open-source stack, 'Physical AI Toolchain on AWS', for building intelligent machines.

ㆍAWS unveils open-source stack 'Physical AI Toolchain'

Anthropic will absorb confidential computing and AI specialist Opaque through an acquihire. OpenAI is in talks to hand off its shelved video-generation AI 'Sora' to a startup, according to a report.

ㆍAnthropic effectively acquires confidential computing AI startup Opaque ㆍOpenAI's missing Sora to return as a film production-focused AI?

Google DeepMind developed a fraud-detection AI model that runs on smartphones with Singapore's Government Technology Agency (GovTech Singapore).

ㆍGoogle DeepMind develops Singapore-tailored fraud-detection AI model that runs on devices

Kling AI officially launched its new video-generation model 'Kling 4.0'. The company said the upgrade improved support for video production requiring high dynamic range and bold camera movement as part of efforts to enhance realism, controllability and professionalism. OpenAI updated ChatGPT's voice feature based on its voice model 'GPT-Live-1.1'.

ㆍKling AI launches video AI model Kling 4.0, generating 30-second videos at once ㆍOpenAI improves ChatGPT voice feature...adds 2 Korean voices

Amazon unveiled 2 new products of its AI device 'Bee', which listens to and records conversations. The products are a wearable worn on the wrist and a credit card-sized voice recorder.

ㆍAmazon unveils 2 always-on recording AI devices 'Bee'...wearable and smartphone-attachable

Dell Technologies will expand features of the 'Dell AI Data Platform' to help companies secure reliable, up-to-date context from their own data and provide it to applications and AI agents.

ㆍDell Technologies expands AI data platform to provide data context to AI agents

ITCEN CTS, an IT infrastructure company under ITCEN Group, will work with Lotte Innovate to expand its artificial intelligence data centre business, starting with operations of the Ansan Global Cloud Center (GCC) scheduled for completion in 2027. ITCEN Entech, led by CEO Jang-ho Shin (신장호), launched an AI-based learning service 'AI Essay Coach' (essay-coach.ai) for students preparing for the 2027 university entrance exam essay track.

ㆍITCEN CTS speeds up expansion of AI data centre business ㆍITCEN Entech launches 'AI Essay Coach'...steps up expansion from SI to education SaaS

Neocloud company VesselAI unveiled a plan to secure 8 AI data centre hubs nationwide by 2028 and to introduce 200,000 GPUs in phases by 2030. AI infrastructure company MangoBoost introduced a serverless AI inference platform, 'Mango Inference'. VesselAI will provide global inference infrastructure for Upstage's large language model 'Solar Pro 4'. Mango Inference is an inference service provided on MangoBoost's full-stack AI infrastructure, allowing use of AI models without complex infrastructure build-out or configuration.

ㆍVesselAI enters global neocloud race with fluid computing technology ㆍVesselAI provides global inference infrastructure for Upstage's 'Solar Pro 4' ㆍMangoBoost launches AI inference platform 'Mango Inference', competing on efficiency

As personal AI agents for individual users spread, whether to open the door to these agents has emerged as a key agenda for website operators. For now, more companies across financial and consumer goods sectors appear to be locking out AI agents.

ㆍ"Not yet"...from finance to e-commerce, companies increasingly block external AI agents ㆍMajor financial firms increasingly block access by personal AI agents..."concerns about fund outflows"