TSM president tooks the topic and raised the perspective to the computing needs of the entire era. He pointed out that the number of tokens consumed is hopping geometrically from the query model to the instruction and action model of agent AI. This shift in underlying logic creates a huge rigid demand for advanced silicon chips.
$英特爾(INTC)$ (Bear Case): The transformation costs are extremely high. The huge loss and negative free cash flow on GAAP's books indicate that its financial constitution remains fragile. The wafer foundry business is still a long way from truly high yield and large-scale profitability, and it remains unclear whether future capital expenditures will drag the balance sheet down
$英特爾(INTC)$ ( Bull Case): The rise of edge AI (Edge AI) and AI agents has brought back computational demands from pure GPUs to heterogeneous architectures with 'CPU + GPUs. Intel's advanced packaging capacity became a scarce resource in the market. With the strategic support of the US government, its geopolitical moat is irreplaceable.
Intel is experiencing painful but necessary "breaking the cocoon". The current surge is a violent repair of valuations from extreme pessimism to a reasonable average. Instead of being the lumbering giant that was slammed by AMD, it's repositioning itself as a key hub of the global technology map using domestic US production capacity and AI margin computing power demand. This is a high risk, high explosive power of geopolitics and computing power dual concept stocks.
$英特爾(INTC)$ The market's craze for Intel is not blindly believing that its subcontracting technology has surpassed TSMC, but rather a fundamental shift in market pricing logic from the GPU-only CPU revival driven by AI Agent. Human society needs not only AI models to generate, but also AI agents to perform, manipulate and retrieve, giving the CPU an irreplaceable hub.
$英特爾(INTC)$ Intel also played several trump cards: signed a plural year contract with Google, Xeon 6 entered its infrastructure, was selected as host CPU for NVIDIA DGX Rubin NVL8 systems, and even aligned with SpaceX, Tesla and xAI. This shows that Intel is no longer fighting alone, but is embedding itself at the bottom of the entire AI ecosystem.
Tesla's post-performance tumbled 3.6% as Musk announced that it would burn more than $25 billion this year, putting negative pressure on short-term cash flows. The cloud giants are stuck in the prisoner's dilemma: Investing in AI now might get punished by shareholders for not paying enough short-term returns, but if not, you might be eliminated entirely
The market doesn't care about the smoke of the Middle East, and capital greed and fear are all focused on AI infrastructure bets. The migration from software to hardware is reshaping the power structure of the entire U.S. stock market
After the emergence of ChatGPT, DeepSeek, Gemini, and Claude one after another, the market was finally forced to confront a fact that an engineer already knew well: the bottleneck of AI is not just computing power, but bandwidth. GPU is not not fast enough, it's not full.
Amazon announced a reinvestment of about $5 billion (20 billion in future) in Anthropic, in exchange for a commitment to use hundreds of billions of AWS computing power and ASIC chips over the next decade. The "revolving investment" model of the giants' left hand investment and right hand cash out earnings has been priced reasonably. This kind of heavy capital game that bypasses the traditional real economy recession and self-produced and self-sales, making the semiconductor sector an independent kingdom free from macro gravity.