Wall Street firms: Kimi K3 is not a repeat of the "DeepSeek moment", but instead strengthens demand for computing power.
After the launch of Kimi K3, the U.S. semiconductor sector declined as markets feared a "DeepSeek Moment 2.0". However, UBS, Nomura, BofA Securities, and Citigroup all agree that the model does not reduce demand for computing power, but may instead drive further expansion of demand for AI infrastructure.
Kimi K3 features 2.8 trillion parameters, a 1 million-token context window, and supports always-on inference, native multimodality, and Mixture of Experts (MoE) architecture. Institutions note that its scale and long context will increase KV cache occupancy, boosting demand for HBM, server DDR5, enterprise-grade solid-state drives (SSDs), cloud infrastructure, and high-speed interconnects.
Citigroup dubs this trend "another Jevons Paradox", explaining that improvements in model efficiency and cost could lead to more applications and token consumption. UBS points out that open-source models typically rely more on memory and storage due to their longer context windows; Citigroup estimates that large-scale deployment of K3 may require supernodes composed of more than 64 GPUs. Nomura, meanwhile, argues that global competition in large language models (LLMs) will push leading labs and hyperscale cloud platforms to continue investing, benefiting AI infrastructure segments including TSMC, NVIDIA, storage vendors, optical module suppliers, and data center operators.
That said, BofA Securities warns that if the pace of model efficiency gains continues to outstrip workload growth while actual usage fails to expand in lockstep, AI infrastructure construction could still see a notable pullback.
3 minutes ago
Sell-off in chip stocks pauses, real test shifts to tech giants' earnings reports
Overnight, U.S. chip stocks rebounded from their declines, with the Semiconductor ETF climbing roughly 1.3%. Major players including Nvidia, AMD and Micron generally recovered, though the Nasdaq remained nearly flat. The earlier concentrated sell-off and leverage rebalancing pressures have temporarily eased. The stabilization of U.S. equities has provided a brief respite for South Korean stocks and East Asian storage and equipment supply chains, but deleveraging inertia and high-volatility holdings have not yet been fully digested. Upcoming earnings reports from Alphabet, Tesla, Intel and others will re-test AI demand, capital expenditure and profit realization capabilities. Therefore, whether the rebound can continue depends on whether the earnings results can support the previous high expectations. (Jinshi)
3 minutes ago
JPMorgan Chase CEO: Investors are underestimating market risks, and he will not buy stocks or long-term U.S. Treasuries at present.
JPMorgan CEO Jamie Dimon said investors are underestimating the geopolitical and fiscal risks facing the global economy, and at current prices, he would not buy the overall stock market nor long-term U.S. Treasuries. Dimon noted that the Russia-Ukraine war, Middle East conflicts, strained China-U.S. relations, and rising military spending amid expanding government deficits could eventually hit markets. While the global economy is more resilient due to reduced energy dependence, this does not rule out sudden market downturns. Persistently large U.S. fiscal deficits could eventually push up interest rates, as bond investors will demand higher returns to hold government debt. He added that even if inflation falls to the Fed’s 2% target, the 10-year U.S. Treasury yield could stay between 4% and 4.5%, leaving limited upside for long-term Treasury prices. On stocks, Dimon said he would consider buying individual high-quality stocks, but not the broader market at current valuations. The S&P 500 has risen nearly 10% so far this year. Turning to AI, Dimon compared the current investment boom to the early days of the internet. He believes massive AI spending could ultimately pay off overall, just like the internet, but the returns and timeline will “definitely not be as people expect”. He pointed out that early internet-era giants like Yahoo and Netscape later faded, while eventual winners like Google and Facebook emerged later.
3 minutes ago
Robinhood Chain ecosystem token PONS briefly surged past $39 million in market capitalization, hitting a new all-time high.
According to GMGN monitoring, Robinhood Chain ecosystem token PONS briefly hit an all-time high market cap of over $39 million, and is now trading at $34 million, up 110% in 24 hours with around $10 million in trading volume over the same period. PONS is the native platform token of Pons, a token-launching platform on Robinhood Chain. The platform supports creating and issuing fixed-supply tokens, uses collected WETH fees to repurchase PONS, and directly burns PONS fees. It is viewed by some community members as the "pump.fun" of Robinhood Chain.
3 minutes ago
Margin balance in South Korean stock market falls to its lowest level since April.
According to data from the Korea Financial Investment Association, as of July 16, the margin balance used for stock financing has fallen to 33.4 trillion won (about $226 billion), the lowest level since April 15. The figure is 13% lower than the peak of 38.6 trillion won recorded at the end of June. Additional data indicates that South Korean retail investors’ enthusiasm for stocks may be cooling. Per the Korea Financial Investment Association, as of July 16, investor deposits dropped to 108.1 trillion won, down from the high of 139.7 trillion won on June 4. (Jinshi)
3 minutes ago