Summary of Key Points
The competition in the AI industry has completely shifted from a focus on “model and algorithm development” to a race for “heavy-duty data centers.” American tech giants are spending tens of billions of dollars to build massive data centers (also known as “computing power factories”). However, local residents strongly oppose the construction of these centers due to issues such as rising electricity prices and job losses. These projects rely on complex capital cycles, with the giants often interlocking their interests. NVIDIA, for its part, is securing power and land to guarantee the sales of its GPUs. This race faces two major risks: public opposition and a fragile internal capital cycle.
1. The Shift in the AI Competition: From “Light Software” to “Heavy-Duty Computing Power Factories”
Previously, AI was seen as software operating in the “digital realm” (such as the ChatGPT model), but now it requires physical infrastructure in the “real world” in the form of data centers. These are no ordinary computer rooms; they require thousands of high-end chips, servers, fast networks, and massive amounts of electricity (1 gigawatt of power is equivalent to the consumption of hundreds of thousands of households), as well as advanced cooling systems (liquid or air cooling). The current AI arms race is no longer about algorithms or talent; it’s about who can build the most and largest data centers. OpenAI, for example, invested $36 billion to $80 billion in building a 1-gigawatt data center in Michigan (not including the cost of upgrading the power grid), and the Interstellar Gate project plans to invest $500 billion over the next four years to build even larger data centers. Tech companies that once touted their “light-asset” models (like OpenAI) have now become heavy-duty players, spending substantial sums of money on building these facilities.
2. Why Do Americans Hate Data Centers More Than Nuclear Power Plants?
A Wall Street Journal survey shows that 75% of American voters oppose the construction of data centers near their homes (with 60% strongly opposing it), a higher rate than the 53% who oppose nuclear power plants. The main reasons are practical concerns:
- Rising Electricity Prices: More than half of respondents believe that data centers drive up electricity costs due to their high energy consumption.
- Limited Local Benefits: Data centers provide limited long-term job opportunities, yet governments often offer tax incentives to tech companies.
- Impact on Daily Life: Data centers occupy land, consume large amounts of water, and generate significant noise.
- NIMBY (Not In My Backyard) Attitudes: People don’t necessarily oppose data centers in general, but they don’t want them near their homes. A Reuters survey found that only 14% of residents would be willing to have a data center in their community.
This opposition is bipartisan; for instance, Texas (a Republican state) has halted several data center projects due to power concerns, while Pennsylvania (a Democratic state) has introduced policies to improve environmental standards and community approval processes.
3. Interstellar Gate: A “Capital Cycle Game” with Interlocking Interests
Interstellar Gate is the largest AI infrastructure project in history, led by OpenAI and involving partners such as SoftBank and Oracle:
- OpenAI is responsible for operations, while SoftBank provides financing (with a total investment of $54.6 billion in OpenAI). Oracle builds the data centers and sells computing power to OpenAI.
The Ohio project is even more complex: SB Energy (a joint venture of OpenAI, SoftBank, and NVIDIA) is building an 8-gigawatt data center, with OpenAI signing a 20-year lease and NVIDIA providing chips and $10.5 billion in credit support.
Reuters has questioned whether this is a form of “circular financing,” where the giants hold shares in each other and guarantee each other’s success. The entire venture relies on OpenAI generating enough revenue from external sales (of ChatGPT services) to cover costs such as rent, electricity, and equipment updates. A failure in this model could have serious consequences.
4. NVIDIA’s New Strategy: From Selling Chips to Securing Power and Land
NVIDIA’s initial focus was on ensuring that its GPUs were purchased; now it has shifted to securing the necessary infrastructure for using those GPUs:
- Data centers require land and power, so NVIDIA invests in companies that possess both. For example, it invested $1.5 billion in SB Energy and $2 billion in Lancium (with an additional $1 billion to be invested once power targets are met).
NVIDIA’s strategy is to lock in future GPU demand by securing land with access to power. Once data centers are established, they will inevitably need to purchase NVIDIA’s chips.
5. The Two Major Challenges Facing This Race
- External Constraints: Public opposition can turn into political pressure, potentially leading to the cancellation of data center projects (as seen in Texas).
- Internal Vulnerability: The capital cycle is tied to OpenAI’s external earnings. If ChatGPT or other services fail to generate sufficient revenue, the entire chain could collapse, given the massive investments involved.
The debates happening on the other side of the ocean are not just idle discussions; they reveal the inherent contradiction in the AI industry’s transition from a “lightweight” to a “heavyweight” model. Technological progress requires substantial infrastructure, but this heavy investment touches on people’s daily lives and economic interests. How long this race can continue depends on the ability of the giants to balance money, land, power, and public sentiment.