BlockBeats News, June 25th, a Goldman Sachs strategist believes that Wall Street's AI trading is entering a more complex phase: the market still believes in the AI investment cycle, but no longer lumps all AI companies into the same valuation framework.

Over the past year, investors have been most willing to buy into direct beneficiaries of AI infrastructure on-chain. Nvidia, TSMC, and some semiconductor equipment and server suppliers have benefited from large cloud computing companies' continued capital expenditures. As long as Amazon, Alphabet, Meta, and Microsoft continue to purchase chips, servers, and data center capacity, hardware companies' revenue expectations will remain supported.

However, the hyperscalers bearing these expenditures themselves have not seen equally strong stock price performance. The market is rewarding the "receiving end" but remaining cautious about the "spending end." Investors are increasingly concerned about whether these tens of billions of dollars in AI investment can ultimately be converted into profit, free cash flow, and shareholder returns.

This is what Goldman Sachs refers to as the AI trade being like a "stretched rubber band." One end sees hardware suppliers' orders and profit expectations being repeatedly raised; the other end sees large tech platforms under increasing capital expenditure pressure. As long as AI demand continues to grow, this structure can be maintained. However, if the market begins to doubt the return on investment, or if the cloud giants suggest that AI spending growth is peaking, related stocks may be repriced.

Goldman Sachs is not bearish on AI but believes that the AI trade has moved from thematic investing to a return validation stage. The market is no longer just asking "who is involved in AI" but is starting to inquire about "who can truly make money from AI."

For Nvidia, TSMC, and the AI equipment chain, the greatest risk is not the disappearance of demand but demand growth no longer exceeding expectations. For Amazon, Alphabet, Meta, and Microsoft, short-term pressure comes from excessive capital expenditures; however, if AI costs decrease or AI products bring in clear revenue, they may instead become beneficiaries in the next phase.

A bigger variable is the AI cost curve. If China, Japan, or other regions can develop and operate high-performance models at a lower cost, the high capital expenditure path of U.S. tech giants could be called into question. The market previously assumed that leading AI necessarily required more chips and larger data centers, but model efficiency improvements and alternative chip development could weaken this logic.

Therefore, the AI narrative is not over, but the buying logic is becoming more refined. The key in the next stage is no longer just whether AI demand exists but who can turn AI investment into real cash flow.