South Korea's AI chip stocks saw a notable pullback on Monday. Industry collaborations are still progressing, as the market begins to reassess the valuation pace of the AI supply chain.
1｜South Korean AI chip stocks pull back as market reassesses supply chain premiums
The Korean stock market plunged sharply in early trading on June 8. According to Livemint, citing Bloomberg, the KOSPI dropped as much as 8.8%, falling about 15% from its recent high and nearing a technical correction. Samsung Electronics slid as much as 11%, while SK Hynix fell 10%. This was not a typical sector rotation, but rather a simultaneous sell-off of core assets driving Korea’s AI trading. Samsung and SK Hynix are global hubs for HBM and DRAM supply, and have been the primary drivers of the Korean stock market’s rally over the past several months.
More interestingly, the positive developments were not absent. On the same day, Reuters reported that NVIDIA announced partnerships with SK Hynix, Naver, and Doosan for AI data centers. NVIDIA’s Newsroom also confirmed that NVIDIA has signed a multi-year technology collaboration with SK Hynix to advance next-generation memory services for global AI factories. The industry logic remains strong, yet the stock declined first. The market’s reaction indicates that certainty in the AI supply chain is no longer sufficient to support valuations alone—capital is now prioritizing the unwinding of crowded trades and excessive expectations.
(Source: Livemint / Bloomberg / Reuters / NVIDIA Newsroom)
2｜OpenAI is turning ChatGPT from a chat window into an action gateway
TechCrunch reports that OpenAI is still moving forward with its ChatGPT super app plan, with insiders directly stating, "Chat is dead." This statement does not mean the chat function is disappearing, but rather that ChatGPT can no longer be limited to just a Q&A interface—it needs to integrate coding tools, agents, file handling, app connectivity, and monetization capabilities into a single entry point. On the same day, 36Kr also mentioned in its morning briefing that ChatGPT will undergo its largest upgrade since launch.
This signal appears to be a product iteration, but beneath the surface, it’s a battle for entry points. Over the past two years, users have treated models as a smarter search box. Now, OpenAI wants to make it the default layer of workflow. The question has shifted: whoever owns user context controls tool invocation, payment methods, and the placement of third-party apps. If ChatGPT becomes a super app, model capabilities are merely the foundation—the real value lies in controlling the entry order.
(Source: TechCrunch / 36Kr)
3｜NVIDIA expands its AI factory partnerships from HBM to power and industrial equipment
NVIDIA and SK Hynix have announced a multi-year technology partnership focused on advancing next-generation memory to support the global AI factory buildout. Reuters and Yonhap report that the collaboration centers on next-generation memory required for AI data centers. 36Kr also cited Jensen Huang’s statement on the same day, noting that NVIDIA has already been purchasing billions of dollars worth of chips from SK Hynix annually, with procurement this year set to increase significantly, and the partnership extending beyond two years.
Another line comes from Doosan. According to the NVIDIA Blog, NVIDIA is collaborating with the Doosan Group on physical AI, robotics, and AI factory infrastructure, with Doosan Heavy Industries exploring the use of gas turbines, steam turbines, small modular reactors, and hydrogen fuel cells to power AI factories. AI factories are shifting from chip procurement to system-level integration. Whoever can secure access to HBM, power equipment, industrial automation, and materials will be closer to controlling the next-generation computing infrastructure.
(Source: NVIDIA Newsroom / NVIDIA Blog / Reuters / 36Kr)
4｜Trump calls for Israeli restraint; Iranian ceasefire back on the negotiating table
On June 7, Iran launched missiles at Israel, marking the first direct attack since the ceasefire on April 8. Both Axios and AP trace this escalation to the same chain of events: Iran responded with missile strikes following Israel’s strike on Beirut, renewing pressure on the fragile ceasefire. Al Jazeera’s live coverage also showed Iran and Israel continuing to threaten each other in a cycle of retaliation and counter-retaliation.
The key turning point is in Washington. Axios reports that Trump asked Netanyahu not to retaliate against Iran’s missile attack, to avoid disrupting U.S.-Iran negotiations. The U.S. is no longer acting merely as an ally, but managing risk between Israel’s military actions and the window for Iran talks. The variables of a Middle East war are shifting from the battlefield to the negotiating table, and markets are now re-pricing Iranian war risks alongside AI valuations and rising interest rates in the same risk category.
(Source: Axios / AP / Al Jazeera / Fortune)
5 | AI security failure leads to litigation, product promises begin to be tested by the court
Ars Technica reports that a survivor of the Nashville high school shooting has sued AI firearm detection company Omnilert. The plaintiff alleges that the system failed to identify the weapon prior to the shooting, despite the company’s prior marketing claims that it could detect firearms before shots were fired. Local media WSMV further reported that the lawsuit also disputes limitations such as camera placement, distance, angle, lighting, and weapon visibility.
The significance of this news lies not in the failure of a single system, but in AI security products entering a test of their liability boundaries. In the past, AI companies often packaged their product capabilities with metrics like accuracy, real-time detection, and automated alerts. When incidents actually occur, courts will ask: Under what conditions is the system effective? Were its limitations disclosed during sales? Did the school develop a false sense of security because of it? Once AI moves from the lab into public safety applications, errors are no longer just about model metrics—they become issues of product liability, procurement review, and public institutional trust.
(Source: Ars Technica / WSMV)
It's also worth knowing ↓
Agentic AI has made coding faster, but the bottleneck in software engineering has shifted. VentureBeat reports that after code generation speeds increased, requirements definition, system integration, testing, and deployment have become the new bottlenecks. AI is not eliminating software engineering—it’s pushing software engineering toward systemic governance. (Source: VentureBeat / TechRadar / arXiv)
White House AI advisor Sriram Krishnan will step down at the end of June. According to The Washington Post and TechCrunch, he plans to leave his government position and found an external organization to continue shaping technology policy. The institutionalization of U.S. AI policy still lags behind the pace of industry expansion. (Source: Washington Post / TechCrunch)
After Notion restored Anthropic's access, application-layer dependency issues resurfaced. A single interruption in a model provider's access can disrupt the workflows of many users. Enterprises' risks in using AI are shifting from price and performance to supply chain resilience. (Source: TechCrunch)
JPMorgan says Strategy may need to rebuild its U.S. dollar reserves. The Block reports that market focus has shifted from how much bitcoin Strategy holds to whether it has sufficient cash buffers to pay dividends and maintain investor confidence. Bitcoin treasury companies are being reevaluated as credit structures. (Source: The Block / JPMorgan)
Adyen won the Gov.UK Pay contract, replacing Stripe. The switch in government payment system providers illustrates that payment infrastructure is not static after a single integration. The public sector is re-evaluating who should manage the national payment interface. (Source: Finextra)
Huachao Shenkong has completed a hundred-million-yuan seed round, betting on AI-powered ultrasound brain-computer interfaces. The capital narrative around brain-computer interfaces is expanding from invasive chips to medical imaging, ultrasound, and AI-driven diagnostic pathways, as early-stage projects seek more clinically viable entry points. (Source: 36Kr)
AI content creators are becoming increasingly difficult to identify. The Verge reports that AI virtual creators are evolving from obvious experiments into harder-to-distinguish platform personas. Content platforms will next need to address not just labeling generated content, but the very authenticity of identity. (Source: The Verge)
Quantum Bit reports that China's open-source AI long-video framework has entered the global top tier. The focus of competition among Chinese AI video tools is no longer just on generating a few seconds, but on engineering capabilities such as long-video consistency, latency, and super-resolution. (Source: Quantum Bit)
Click to learn about the open positions at BlockBeats
Welcome to the official BlockBeats community:
Telegram subscription group: https://t.me/theblockbeats
Telegram group: https://t.me/BlockBeats_App
Official Twitter account: https://twitter.com/BlockBeatsAsia