Kimi's announcement of its new model caused a brief "DeepSeek effect" fear in the market. The very low API prices reinforced the idea that the same tasks could now be accomplished with fewer GPUs and HBMs. This concern led to sharp sell-offs in GPU, memory, and data center stocks. However, the picture presented by Kimi K3 shows not a weakening of hardware demand, but rather a much faster growth in the use of artificial intelligence. BofA also points out that Chinese open models like Kimi K3 offering API prices up to 80% lower than Western models may not necessarily mean that hardware costs are falling; it's more of a pricing and market share strategy.

According to BofA's OpenRouter data, global token consumption has increased by an average of 6% per week since 2025 and by 9% since the beginning of 2026. In July 2026, approximately 70% of the tokens used belonged to Chinese models. Cheap Chinese models not only create alternatives to Western models; they also make previously expensive tasks economically viable, thus increasing overall usage.

Kimi K3 is one of the clearest examples of this. Moonshot AI announced that demand exceeded expectations after the model's release, that existing clusters were nearing their limits, and that the company faced "unprecedented computing challenges." Demand grew so rapidly that new subscriptions were temporarily suspended, and existing capacity was allocated to paying users. In short, lower pricing didn't eliminate the need for GPUs; it simply led to more users flocking to the system and capacity filling up much faster.

This is where Jevons' paradox comes into play. When a technology becomes more efficient and cheaper to use, total expenditure and resource consumption don't necessarily decrease; they often increase as use cases expand. Lower token prices mean more users, longer conversations, more code generation, and more complex agent tasks. Even if the cost of a single query decreases, the total computing need increases as the number of queries and workflows where the model is repeatedly called grow faster.

Moreover, while efficient, Kimi K3 is not a small model. With a total of 2.8 trillion parameters, each system running the model requires approximately 1.4 TB of HBM, according to BofA. Thanks to the MoE architecture, even if only a portion of the parameters are active in each query, maintaining model weights in memory and serving a large number of users simultaneously requires significant GPU and HBM capacity. The repeated calling of the model by coding and agent tasks also increases the inference load. More detailed technical data that Moonshot AI will share on July 27th will provide a clearer picture of the model's true hardware requirements.

Looking at the overall picture, interpreting low API prices as "less hardware is needed now" can be misleading. As the cost of use decreases, demand grows, and the supply of compute is struggling to keep up. The K3 case shows that efficiency hasn't ended hardware demand; on the contrary, it has spread the need for GPUs, HBM, and data centers to a wider user and application base.

On the other hand, another indicator for the AI ​​market will be the earnings reports of Alphabet and Tesla, which will be released after the close of US markets tonight. On the Alphabet side, growth in Gemini usage, the contribution of AI to search and advertising revenue, Google Cloud demand, and data center investments will be closely monitored. Alphabet, which increased its cloud revenue by 63% year-on-year in the previous quarter, is expected to maintain the same pace this quarter. Statements regarding increased capital expenditures or continued capacity constraints could indicate strong demand for GPUs, TPUs, HBMs, and data centers.

Tesla is expected to increase its capital expenditures to over $25 billion by 2026, driven by investments in AI, robotaxi, and robotics. This figure is approximately three times higher than last year, and the market will focus on when these investments begin generating revenue. The messages both companies provide regarding AI demand, investment expenditures, and capacity needs could affect not only Alphabet and Tesla shares but also the entire AI ecosystem, including Nvidia, Broadcom, Micron, and data center companies. While strong demand and positive investment signals could support an uptrend in the sector, statements suggesting that investments will be slow to generate revenue or that spending will slow could put pressure on stocks.