US-China Chip Shock Triggers Twin KOSPI-KOSDAQ Circuit Breakers... Park Tae-woong: "AI's Golden Time, Details Decide the Outcome"

The KOSPI index widened its intraday loss to as much as 10%, falling into the 6,700 range, while the KOSDAQ dropped more than 8% to 702 points, triggering circuit breakers on both markets. The won-dollar rate also fell to around 1,464 won, making the session's directionality hard to read. Foreigners sold roughly 2.6 trillion won in KOSPI cash and futures combined, extending their selling streak to a third straight session, and also sold about 50 billion won in KOSDAQ. In the futures market, selling was particularly heavy at around 1.1 trillion won.

The decline stemmed from two main factors. First, reports that China can produce DUV lithography equipment domestically sent Netherlands-based ASML plunging around 8% intraday — trading-halt territory — while its US-listed shares fell about 5%. Hosts noted, however, that "capable of production" differs from "mass production and delivery," and that the report still needs fact-checking. Second, concerns over circular financing intensified after Nvidia said it would guarantee $50 billion in funding for SK Hynix and $250 billion for OpenAI, sending Nvidia's CDS premium sharply higher and its stock down about 4% at the close.

Nvidia's decline pushed Apple back to the top of global market capitalization. Apple, with lower AI investment exposure, has relatively outperformed over the past month; its market cap now stands at about $4.9 trillion versus Nvidia's roughly $4.7 trillion.

This week brings a heavy earnings calendar. Tuesday: HD Hyundai Electric, Hanwha Systems, HYBE, Daewoo E&C. Wednesday: SK Hynix (morning release and earnings call), Samsung C&T, HD Hyundai Heavy Industries, HD Korea Shipbuilding & Offshore Engineering, Korea Aerospace Industries, Krafton, GS E&C. Thursday: Samsung Electronics' confirmed results (with segment breakdown), Samsung Electro-Mechanics, LG Energy Solution, LG Electronics, Samsung SDI, POSCO Holdings, SK Innovation, Amorepacific. Friday: Hanwha Aerospace, LG Chem, Hyundai E&C, Ecopro BM.

According to a Wall Street Journal report, the Trump administration finds itself caught between two conflicting policy goals — price stability and US semiconductor development — as Apple pushes to use Chinese-made memory chips while Micron objects. Apple argues that, given the global memory supply shortage, it must be able to use Chinese memory to keep consumer prices in check. Micron counters that letting Chinese memory makers into the market risks repeating the collapse of the US steel industry, and contends that Apple's low-margin demands are themselves a cause of the shortage.

Panelists described Apple's approach as a "Rockefeller strategy" — waiting while rivals invest and burn out, then swooping in at the end. They noted, however, that this no-investment approach also carries a negative precedent: Rockefeller-style underinvestment once ceded US oil sovereignty to the Middle East and helped trigger the oil shocks. In other words, what is optimal for a company is not necessarily optimal at the national level.

Shipbuilding, coming off 2024 — widely regarded as the best year since the pandemic — is seeing order volumes and order values in H1 2026 run at almost the same pace, signaling the strong upcycle continues. Recent orders are concentrated in "mainstream" cargo-ship segments such as containerships, where demand is solid and pricing power is strong. Beyond commercial vessels, expectations are also building around defense exports and offshore data centers as new growth drivers.

On profitability, Korean shipbuilding's historical peak operating margin was 16.6% in a single quarter, recorded by HD Hyundai Mipo on vessels ordered in 2007-2008. In the current cycle, forecasts suggest Samsung Heavy Industries could achieve upward of 20%, implying profit levels nearly double the prior historical peak.

The source of Korean shipbuilding's competitiveness was identified as R&D investment that never stopped even during downturns. By contrast, Japan cut back on additional investment and design-workforce expansion after the 2000s and stagnated, while Korea kept investing in R&D through downcycles and continued to accommodate demanding requirements from overseas buyers — an accumulated advantage now underpinning its competitiveness. By country, the fourth-ranked player, Italy, holds only about 1% share, underscoring the clear Korea-China duopoly; excluding China, there is effectively no competing nation.

Park Tae-woong, chairman of the Green Book Forum, joined to assess the controversy over AI investment. Citing Roy Amara's law — that the effect of a technology is overestimated in the short run and underestimated in the long run — he argued AI is following the same path nanotechnology took in the 1970s: an initial wave of hype, followed by disappointment, and now pervasive adoption across every industry. He contrasted this with the metaverse, where a serious look at the underlying technology — fundamental discomforts in servers and infrastructure — allowed the bubble to be identified early.

He noted that the market currently mixes bad and good signals. The bad signals: the circular revenue structure between Nvidia and OpenAI, data-center buildouts falling behind schedule due to power shortages and local resident opposition, and cash flow concentrating toward GPU and memory manufacturers. The good signals: Anthropic's operating profit is surging even after absorbing these rising costs; 2026 data-center construction volume is seven times that of 2025, so even at a 70% completion rate the absolute volume is far larger; and Chinese open-weight models (such as Kimi) are achieving strong performance even on older GPUs while memory usage is actually increasing.

He identified OpenAI, not the AI sector broadly, as the real epicenter of the bubble debate. Anthropic's B2B revenue surged from an annualized $30 billion in April 2024 to $47 billion in June 2026, putting it roughly double OpenAI's revenue despite OpenAI having 800 million users — yet it is OpenAI, not Anthropic, undertaking massive data-center investment, which he said raises legitimate doubts. He interpreted this through the lens of differing US-China AI strategies: the US is focused on capturing frontier AI (AI surpassing human intelligence) first, while China is pursuing an "intelligent manufacturing" strategy of embedding AI across every industry and releasing open-weight models for free to squeeze the profitability of US frontier AI — with OpenAI as the primary target of that strategy. He noted that even though US investment is ten times China's, the performance gap has narrowed to just 2-3%, and that the listing of ChangXin Memory Technologies has freed China from dependence on state subsidies, giving it independent fundraising capability.

On the sustainability of AI investment, he pushed back on the washing-machine analogy. Just as more washing machines increase total laundry volume (Jevons paradox), AI investment also creates its own demand. He pointed to Tesla's self-driving and Samsung Galaxy's AI features becoming top sales drivers as evidence AI is being embedded into every product, and forecast that once agentic AI takes hold, token consumption could rise to over 150 times current levels. He noted, however, that the main obstacle to agent adoption at scale is the lack of security standards (such as delegating payment information), and that China is already accepting that risk and moving quickly to commercialize agents.

Park Tae-woong interpreted President Lee Jae-myung's recent San Francisco remarks — "we have moved beyond an algorithm race into an era where computing resources, energy, data, and the speed and competitiveness of industrial application determine outcomes" — as a dual message Korea is sending to both the US and China. To the US, it signals that Korea can supplement full-stack capabilities in semiconductors, energy, and data that the US itself lacks. To the rest of the world, it signals: "if the US is too expensive and China can't be trusted, come to Korea." He argued Korea is the only non-China nation with end-to-end capability spanning batteries, high-power substations/transformers, high-voltage DC transmission, solar and wind, semiconductors, large-scale data-center construction experience, domestic NPUs with twice Nvidia's power efficiency (Furiosa, Rebellions, etc.), and its own AI models.

He nonetheless flagged "lack of detail" as the biggest challenge. Government policy, he said, stops at outputs (e.g., "we will do AI across the board") without outcomes — concrete target figures. For example, if AI data centers are built merely as simple rental space without operational know-how, Nvidia captures the GPU value and OpenAI/Anthropic capture the model margin, leaving Korea with only depreciation, electricity costs, and land rent — reduced to a "real estate leasing business." Conversely, filling in the details — using domestic NPUs, improving utilization rates (typically from 35% to 70%), and internalizing operating software — can generate margin at every layer. He noted the US struggles with data-center buildout due to land scarcity and an aging power grid, while Korea holds an advantage by owning its energy infrastructure end-to-end, and that Nvidia's large investments in Naver and SK Telecom stem from this backdrop.

He also addressed the significance of the Samsung Electronics-Broadcom and SK Hynix-Nvidia deals. The Broadcom deal comprises three elements — bulk memory purchases, support for custom GPU design, and 2-nanometer foundry volume — and he assessed that if Samsung Electronics can bring 2nm yields up to TSMC's level, it could secure the large orders needed to offset its accumulated foundry losses. He viewed such deals as effectively long-term supply agreements (LOIs/MOUs) that smooth out the semiconductor cycle. The SK Hynix-Nvidia deal links construction of a 2GW-scale data center with Nvidia chip purchases, and he forecast that if it proceeds via internalized operational know-how rather than simple co-location, Korea could become Asia's AI data-center hub.

As a way to redistribute wealth generated by AI, he proposed sovereign-wealth-fund-style equity participation models. He argued that since Samsung Electronics has received annual tax breaks and R&D support of 2-3 trillion won for the past decade, some share of the resulting profit belongs to the public. Citing Shinan County's solar project, where equity participation paid households an annual dividend of 7.5 million won and drove population growth, he suggested a nationwide equity-participation model in AI and energy funds could serve as a realistic social safety net. He also urged that job losses, already a locked-in future, be taken seriously as a wealth-redistribution issue for the next generation rather than treated merely as a political slogan.

Host Lee Kwang-soo stressed that in a sharp selloff, principles matter more than experience. He noted that the probability of a further decline versus a rebound is, over the long run, exactly 50-50, and that KOSPI's plunge from 9,000 to 6,000 does not itself reduce the odds of additional downside. He argued that even after a sharp fall, investors still need a response that prepares for further declines — that is, an effort to minimize losses as much as possible.

Concretely, he recommended re-establishing a plan for an acceptable loss level — for example, a rule to trim exposure if KOSPI falls from 6,100 to below 5,800. This is not pessimism about the market but a necessary step to preserve the capacity to act again once the market eventually recovers. He also urged investors not to exit the market, since only those who remain can benefit from a recovery.

He also stressed that a price decline and a value decline are different matters. What has fallen now is price, not corporate value — the gap between price and value has simply widened. Still, he reiterated that belief alone should not be used as an excuse to delay action; a real response to the price decline is needed.

He also raised the concern that Korea's stock market has become a "lonely market where only investors remain." A healthy market requires a triangle of investors, companies, and government, but over the past month or two, investors have been struggling alone without shareholder-return efforts from companies or supportive policy from the government. He suggested that concrete shareholder-return measures exceeding market expectations in the upcoming Samsung Electronics and SK Hynix earnings releases would be needed to restore confidence.