The end of AI is really electricity! Google (GOOGL.US) is planning a new AI infrastructure layout in the US energy hinterland

The Zhitong Finance App learned that US tech giant Google (GOOGL.US) is actively considering building a data center in Lee County, New Mexico. This latest AI infrastructure plan can be described as showing the increasingly important site selection logic of tech giants in AI infrastructure expansion — in addition to chip procurement, stabilizing power supply, energy costs, water resources, and community support also determines whether the project can be implemented. Lee County is located in the western Permian Basin. Rich natural gas resources and local gas power generation infrastructure may provide energy conditions for large-scale computing facilities.

However, Google is currently still evaluating the project and carrying out community communication, and has yet to disclose the construction scale, commissioning time, and specific power supply plans. Google is considering building a data center in Lee County, highlighting the importance that the expansion of AI computing power attaches to energy supply conditions. Abundant local natural gas resources and existing gas power generation infrastructure provide potential energy support for new data centers.

AI computing power infrastructure extends to the energy hinterland, and Google is looking for the next cornerstone of growth

Judging from the AI computing power industry layout logic, as AI reasoning and intelligent applications increase the demand for continuous computing, large cloud computing vendors urgently need to simultaneously plan chips, computer rooms, and power supply capabilities. Judging from the economy of data center site selection, abundant fuel resources and existing gas power generation infrastructure may provide energy appeal for Google's proposed projects. This is why Google promises to bear the full cost of its own energy use and required infrastructure.

The spot price of natural gas in the region once fell to a negative value due to excessive gas supply, reflecting the abundant supply of natural gas in the region. Furthermore, Google management also sees community communication as an important part of project development.

“Our goal in Lee County is simple: to be a trusted long-term partner to do the right thing to support the place we call home,” Google said in a blog post introducing the proposed project. “This includes managing critical water resources responsibly, covering 100% of the costs of the energy we use and the infrastructure we need, and being active in community affairs.”

According to financial data previously disclosed by Google, Google Cloud's revenue in the second quarter increased sharply by 82% year on year to about 24.8 billion US dollars; the company disclosed during the same period that the Gemini model processed 22 billion API terms per minute. These data all showed that with the huge increase in computing power brought about by the popularity of AI applications around the world, Google's demand for AI computing power resources can be described as continuing to expand.

Google said it is aware that many communities are resistant to data center development, so it plans to start a conversation with local residents first.

The county is located in the western Permian Basin, which is rich in oil, gas, and natural gas liquids. The local area has considerable natural gas power generation capacity and related infrastructure. However, over the years, oil and gas extraction companies in the region have burned natural gas through torch burning when focusing on extracting more valuable oil. Natural gas spot prices in the region have fallen to negative values due to an oversupply of natural gas associated with shale oil extraction.

With the advent of the AI inference era, global demand for AI computing power continues to expand at a blowout

Strong AI computing power demand support linked to the AI computing power industry chain level has been clearly reflected in the strong performance of industry chain leaders, South Korea's continued record semiconductor exports, and long-term capacity agreement arrangements.

According to South Korea Customs data, semiconductor exports reached 16.5 billion US dollars, up 270% year on year; “AI chip superpower” Nvidia's revenue for the second quarter of fiscal year 2027 as of July 26 reached 96.2 billion US dollars, up 106% year on year, of which data center revenue was 89 billion US dollars, up 117% year on year; the company's revenue guide for the next quarter was the median value of a record $108 billion revenue range, fluctuating 2% up and down.

Anthropic, which is preparing for a record IPO in the US stock market, is also expanding the supply of computing power through long-term agreements. It announced in April of this year that it will invest more than 100 billion US dollars in AWS-related technology over the next ten years to obtain up to 5 gigawatts (GW) of new capacity to train and operate Claude; another agreement signed with Google and Broadcom involves several gigawatts of next-generation tensor processing unit (TPU) capacity, which is expected to be launched one after another starting in 2027. Together, the above data shows that infrastructure demand at the AI computing power level has already generated large-scale actual procurement and supplier revenue. These long-term commitments have increased the visibility of future chip, computer room, and power demand, while also making supplier delivery capacity, financing costs, and actual customer usage become key variables in AI investment returns.

The market rebound also highlights the strengthening of the bullish logic of strong AI computing power demand for the entire computing power industry chain. In mid-August, South Korea's KOSPI index rebounded more than 20% from its July low; as of September 9, the Philadelphia Semiconductor Index had risen for the fifth consecutive trading day, with a cumulative increase of nearly 6% during the period. However, the market then weakened again. On September 14, KOSPI closed down 3.26%, and half of the fee fell by about 5.9%. Remarks about the slowdown in AI development and pressure on interest rates once again affected valuations.

The GPT-6 Astra model recently launched by OpenAI and the RSI technology path focused on by AI leaders are expected to become the two core driving forces driving the exponential expansion of AI computing power demand, namely the AI big model with better performance, the use of a wider range of AI application tools, and the next generation AI training path with stronger computing power requirements, which are an important basis for the continued growth in AI computing power infrastructure demand.

The investment significance brought by Astra is to improve the success rate and economic viability of complex tasks, so that companies are willing to deploy more agents and handle more professional tasks; Wall Street financial giant Morgan Stanley's recent emphasis on “shifting from demand debates to physical supply constraints on AI themes” is a new round of AI computing power resource demand expansion mechanism brought about by Astra, the most advanced model. The statement by the OpenAI product manager that demand is unprecedented and that the company may suspend new Pro subscriptions can be described as an important sign that AI computing power service capacity is under pressure recently.

Judging from the underlying technical logic, models such as Astra with computer operation and complex task execution capabilities may expand AI requirements from a single round of question-and-answer to an agentic workflow (Agentic Workflow) that runs continuously: a single user delegation may trigger multiple rounds of reasoning, file processing, tool calls, and result verification. If the range of tasks that can be completed, the number of users, and the frequency of use increase simultaneously, inference requirements may form a new impetus for expansion.

Looking further, AI inference infrastructure must simultaneously address computational, memory, and service delays. Prefill (Prefill) processes the input context and can usually form strong parallel computing requirements; decoding (Decode) gradually generates output, making it more susceptible to memory bandwidth limitations in many interactive scenarios. Long contexts and a large number of concurrent requests can also increase the capacity and access pressure of the KV Cache. Actual bottlenecks change with the model architecture, batch processing scale, and caching strategy. This means that Google's economic goal in expanding its data center is to allow accelerators, high-bandwidth memory (HBM), networks, and power supply systems to work together to increase the amount of inference services that can be delivered per unit cost on the premise that response time requirements are met. The potential advantages of regions rich in energy resources need to be converted into profits through long-term reliable power supply resources, reasonable AI infrastructure construction costs, and high utilization rates; this is also an investment logic that truly corresponds to “the end of AI is electricity.”