Goldman Sachs Traders: Interest Rates Remain Key to Stock Market Rally, Fad of Token Buying Has Peaked

Goldman Sachs One-Delta trading desk head Rich Privorotsky pointed out in the latest market observation that, despite NVIDIA's latest financial report and guidance exceeding expectations across the board, the underlying logic of the AI industry chain is undergoing profound changes. The past fervor of enterprises blindly pursuing the total consumption of tokens has peaked in the first quarter of this year.

NVIDIA's financial report exceeds expectations, but the structure is quietly changing

Rich Privorotsky pointed out that NVDA's quarterly results and forward guidance are astonishing, even activating the Asian hardware supply chain in reverse, leading to a nearly 20% surge in SoftBank and an approximate 8% rise in the Korea Composite Stock Price Index (KOSPI).

However, NVIDIA's stock price fell slightly after the market, reflecting that after market expectations are pushed to the extreme, it has become increasingly difficult to continue to surprise the bulls.

What is more noteworthy is the internal structural change in NVIDIA's financial report: its data center business recorded $7.52 billion in revenue, with the networking business contributing $1.48 billion, surging 199% year-over-year.

This indicates that the important driving force behind NVIDIA's performance this round of exceeding expectations is tilting from pure GPU computing power to network infrastructure.

The bulls believe that this further consolidates the long-term narrative of the "AI factory"; while the bears question that its core logic is drifting, and it faces the risk of being nibbled away by new entrants in the core computing power share.

Farewell to "buying IQ", open-source models breed the "end-side inference" inflection point

Through in-depth tracking of OpenRouter platform data, Cursor benchmark tests, and the GPU rental market, Rich Privorotsky made a highly disruptive judgment: open-source models are becoming an irreversible real trend.

For example, Cursor's self-developed models and smaller specialized underlying architectures not only significantly reduce costs but also have output capabilities that are sufficient to meet the vast majority of programming needs.

This change leads to a major shift in enterprise performance indicators: enterprises are shifting their focus from simply "total token consumption" to "effective output per dollar", which is no longer simply buying "IQ" from large model companies.

As Huang Renxun said, inference has become a new workload, and tokens have become a new type of commodity, which means that the "cost of a single effective action" is becoming the core indicator of the industry. This year's first quarter is likely to mark the peak of "Tokenmaxxing (token maximization)" as a key performance indicator.

The unlocking of this "token operation decentralization" has brought a brand new bull and bear game to the hardware market:

• Bear market logic: Currently, a large number of tokens are only used for reading emails, summarizing documents, searching PDFs, or routing workflows, making a costly top-tier GPU summarize emails, like "using an industrial blast furnace to boil water." As Microsoft rebuilds the Windows laptop roadmap around a dedicated NPU with 40+ TOPS of computing power, and Apple's unified memory architecture makes local inference practical, open-source models of 1B to 8B scale can run smoothly on consumer-grade hardware.

• Bull market logic: After local inference and end-side computing power absorb low-end loads, it can actually liberate the most expensive cloud GPU production capacity on the surface, promoting it to move towards more advanced complex reasoning, multimodal, and agentic workloads, and the global total token consumption will still show an explosive increase. The current price of the GPU rental market is still strong, which just proves that computing power and electricity are still scarce, forcing enterprises to turn to self-hosting or open-source solutions.

Geopolitical and macro risks: U.S. stocks face numerous obstacles to new highs

On the macro level, as the Memorial Day in the United States approaches, Wall Street is closely watching whether the U.S. can find a "downhill road" to ease the situation in Iran before the summer travel peak.

Despite frequent media headlines about "negotiations to reach an agreement," Iran has rejected Trump's demand for enriched uranium, and the actual details are still full of contradictions. However, market behavior shows that fund behavior is currently more afraid of missing the compression and regression of geopolitical risk premiums than fearing a resurgence in oil prices.

Physical crude oil traders show divergence against the market: macro funds believe that crude oil will震荡下行due