Global Stock Markets Are Battered: AI Evolves, Workhorses Pay

U.S. stocks had barely finished falling, and A-shares dropped yesterday too. Looking at all that green, the song "Boundless Horizon Is My Love" started playing in my head.

That song suddenly gained a vivid picture.

What's going on?

AI companies are still pouring money in, but some new U.S. data centers are stuck on grid connections and permitting. The money is spent first; the day they start making money gets pushed back. Investors are starting to worry: can this business still grow that fast?

On this side, in China, Ulanqab's Envision Xinghe base already dispatches wind, solar, storage, and data-center load together.

Building a data center: what trouble did the U.S. run into, and how are these Chinese projects handling it?

Follow this thread and it touches our stocks, our monthly bills, and what ordinary people stand to get once AI scales up.

Why Texas Residents Oppose Building Data Centers

Let's start with Texas and walk through it.

Texas is the state in the southern U.S. where OpenAI and its partners built the first data center campus for Stargate, in Abilene. That campus is already operating, and new projects are queued up waiting for grid connections.

But Texas' newly elected governor, Gina Hinojosa, had already said on her campaign site: pause data center projects first.

She wants to wait until the state legislature sets up resident protections and then lets residents take part in deciding whether a data center comes into a community and how it should be built.

She is currently a Democratic state representative, normally involved in drafting state law, and doesn't formally take over as governor until January 19 next year.

Why did she bring this up?

Businesses are focused on AI's future; people who live nearby are focused on their own homes. Is there enough water? Who pays the electric bill? Is the data center noisy at night? Someone has to deal with these questions.

If a data center were planned next to my home, no matter how big the investment figure, I'd ask first: how will my water and electricity be affected?

The future sounds far away; water and electric bills are close.

The current governor, Abbott, has already started investigating.

Abbott is a Republican, Hinojosa a Democrat; they're the two main U.S. parties and compete in elections. But on data centers, one has ordered a review and the other proposes a pause; both have to respond to local residents' concerns.

On August 3, Abbott ordered a review of data centers queued in the Texas main grid interconnection process: how much in tax breaks they received, how much electricity they'll use, what they'll do about water, and how noise will be handled.

His statement in the announcement translates roughly to: put Texans first.

Data center grid connections and power dispatch are handled by ERCOT, the state's grid operator. The same day Abbott spoke, ERCOT announced that a batch of interconnection classification results originally scheduled for August 7 would be delayed.

By September, Abbott asked to pause issuing related environmental permits until ERCOT finishes its review.

Wait, Trump is also a Republican and has backed AI. Why is the state investigating?

Right, in 2025 he signed executive orders to speed up federal approvals for data centers.

But grid connection, environmental, and land issues also involve state and local approvals. Trump can push federal agencies, but local permits still have to be approved locally.

Businesses want to start construction a day earlier; residents want to first understand water, electricity, and daily life.

In San Marcos, Texas, a project got a full night of meetings.

This February, a $1.5 billion data center wanted to be built there, requiring the city council to agree to a zoning change. The room was packed, with people listening from outside.

Residents of a neighboring ranch worried about water and their way of life; construction workers in orange vests hoped to get jobs.

The meeting went past midnight; the council voted 5 to 2 to reject the zoning change.

By June, the city changed its rules to exclude data centers from permitted land uses.

Fifteen hundred million dollars on the table, and the project still didn't pass. Telling residents how big the investment is doesn't answer the questions they care about.

How U.S. Data Center Moves Affect A-Shares

By now you might wonder: how does the U.S. building a data center hurt A-shares?

First, lay out this round of market moves.

On November 5, U.S. Eastern time, the Nasdaq fell 5.52%, and the Philadelphia Semiconductor Index fell 12.25%. Nvidia fell over 9%, AMD over 17%, and storage and server vendors fell broadly.

The next day in Asia, the Nikkei fell 5.96% and the KOSPI fell 6.02%.

For A-shares, the Shanghai Composite fell 1.10% and the ChiNext fell 3.96%.

AI companies train large models; shareholders train their psychology.

This decline was already about valuation, funding, and earnings fears. After the election result, how data centers will be regulated added another thing to reconsider.

The data centers are in the U.S., but the servers, chips, and optical communication gear inside them are tied to global suppliers, and Chinese companies supply them.

If a data center was supposed to go live next year and slips a few months, the timing for buying equipment, delivering, and getting paid can all shift. But interest and operating costs still come due.

The data center hasn't started work, but the bills already have perfect attendance.

That explains why, at noon on November 6, Zhongji Innolight, Eoptolink, and T&S Communications all issued announcements.

Eoptolink said operations are normal and orders sufficient. Zhongji Innolight said orders from major customers already cover most of 2027, and some customers have given clear 2028 demand guidance.

If orders are already booked into next year, why are the stocks falling?

Companies were showing existing orders; investors are pricing in how much more they can earn later. Orders for 2027 are already in hand; demand guidance for 2028 is a customer's early estimate of how much it will need, with actual purchasing to be arranged later.

People previously expected rapid growth the year after; now that data center schedules may shift, they have to redo the math.

Companies are reporting next year's orders; stock prices are already worrying about the year after.

So domestic investors watch U.S. data centers. If they come online a few months late, suppliers' business can be affected.

Suppliers are waiting to deliver, and for data centers to open on time, the power still has to be connected.

How Chinese Projects Supply Power and Save Water

The U.S. also has wind and solar. Why worry about not having enough power?

Because between how much is generated and how much a data center can use at any moment, there are storage, transmission lines, and dispatch.

According to the U.S. Energy Information Administration, in the first nine months of 2025, wind and solar met 36% of electricity demand in Texas' ERCOT region, with storage helping carry daytime electricity into the evening.

But on a night with little wind, a data center has to plan ahead: how much battery is left, how much other sources can cover, and whether lines and substations can deliver.

If I ask an AI question at midnight, it has to answer.

If it replied, "no wind today, talk tomorrow," AI would be at the mercy of the weather.

The PJM grid in the eastern U.S. has similar problems. Data centers run almost around the clock, but new generation and transmission haven't kept up, and even the spring and fall maintenance windows are tight.

How does China handle it?

First look at a problem already hit: local reports in Inner Mongolia mention that some data centers are still waiting for power; construction schedules between data centers and the grid don't line up, and dispatch standards aren't unified.

The data center side is ready, but the supply side isn't; machines have to wait. How do the bases already running line up both sides?

The Envision Xinghe base mentioned earlier dispatches wind, solar, storage, and data-center load together: first forecast how much power can be generated, then arrange storage and load accordingly.

The base announced production in August this year. The project leader said 80% of the campus's electricity comes directly from its own wind farm.

Before, "AI is riding the wind" was about industry prospects; here, the conversation ends up with a weather forecast.

Once you have the power source, you still have to deliver it to the data center.

Ulanqab also hosts the Zhongjin Data base, which built a dedicated line connecting a wind farm 16 kilometers away and works with the public grid. The operations manager says supply has been stable since July 2025, and it has supported training for several large models.

When a data center plans to open, how much power the source can supply, and when the line in between can be connected: these have to be arranged together before construction.

China's rules for green power direct-connection projects also require the power source, the data center, and the line to be built as a unit and commissioned simultaneously.

That's power. There's still the water that residents care about.

Servers run hot, so data centers need cooling. Ulanqab's average annual temperature is 4.3°C, and nearly 10 months a year can use natural cold sources. While we reach for a coat, a data center can use the cold to cool its servers.

In summer, Century Internet's local base sprays reclaimed water, treating sewage to a cooling-grade standard and reusing it, reducing fresh water drawn locally.

People's Daily gave a figure: for Century Internet's projects actually operating in 2025, each kilowatt-hour consumed 0.099 liters of water. That number matters, but as data centers multiply, total local water use also has to be counted.

China's Water Conservation Regulation also says that in water-scarce areas, projects that could reasonably use unconventional water but don't will not be granted new water withdrawal permits.

Data centers need power; nearby people need lights. Data centers need water; nearby people have to live. Only by considering all these together can you tell whether a data center belongs in a place.

New Data Centers Need Power; Who Pays?

That covered how data centers get running. Next comes the money.

When I buy an AI membership, I know what I'm paying for. But people who live near a data center may face costs related to it without buying a single membership.

How?

There are two costs: one is building lines, expanding substations, and keeping backup power for the new data center; the other is buying electricity day to day.

The first cost comes before the data center opens: the grid has to build supporting infrastructure in advance. If a project is delayed, scaled down, or cancelled, who pays for the money already spent?

If that cost ends up spread over other local ratepayers, and I lived there, I'd ask: you wanted the data center, why am I getting a share of the bill?

I haven't even registered for an AI account, and I'm already a sponsor.

Now the day-to-day electricity cost.

Using a model revised in October 2026, the Dallas Fed calculated that, through 2025, the additional electricity demand from data centers raised U.S. wholesale electricity prices by about 6.9% nationwide and about 15% in Northern Virginia.

Wholesale electricity price is the price between generation and retail sales. Whether local residents actually pay more depends on how local pricing and cost allocation work.

If I lived there and the AI membership went up, I could skip it; but I still have to turn on the lights at home.

How much of the directly related costs, like lines and backup, are companies willing to bear?

OpenAI said on November 5 that for campuses it owns and operates, it will pay all electricity and related infrastructure costs itself. It also promised to provide financial assurance before large expenditures begin, so that if a project later collapses, the risk doesn't fall on other electricity users.

In China, a set of charging rules for new-energy nearby-consumption projects has been in effect since October 2025.

Take backup power. Some data centers normally use their own wind power, but when wind is short, they call on the public grid. Even if they use it rarely, the grid has to reserve lines, equipment, and backup generation in advance, ready to step in.

Want the grid on call?

Then pay extra.

This transmission and distribution fee depends mainly on how much supply capacity the grid must reserve for the project. Projects requiring high reliability and backup capacity can also choose a capacity fee plus energy fee algorithm, billed per the actual rules.

The larger the supply guarantee a data center needs, the more it pays under the rules; the project's own books get settled first.

Can Open-Source and Low-Cost Models Save Us Money?

If power, cooling, and model-running costs actually come down, will our monthly renewal fee drop a bit?

Don't celebrate for your wallet yet; it depends on whether vendors cut prices. If they keep prices, the savings go into profit, and the payment page still shows the old number.

Good news: it saves money.

It's their money being saved; I'll cheer for them for free.

Hoping vendors will voluntarily cut prices leaves me no options. If another provider, roughly as good and cheaper, is available, I can switch; then the original vendor has to think about whether users will renew.

Where do these new products come from?

One path is to build on open models.

For example, DeepSeek-V4.1-Flash-1017 has code and model weights under the MIT license. Teams can use and modify them under the license, and build commercial services. Starting from a ready-made model saves part of the build-from-scratch work.

When models are open to more teams, each can build their own product, and we get more options.

From giving someone a fish to building a fishery.

If the original vendor raises prices or shuts down, we can switch.

DeepSeek also opened DeepInfer, giving product teams inference tools and saving them from building some basic functions from scratch.

If a team deploys itself, it can pick versions and change features. But it also owns the machines, electricity bill, and maintenance; if something breaks at 2 a.m., it has to find someone to fix it.

Teams that don't want to run their own setup can use public APIs and pay by usage.

As of November 6, DeepSeek's official Flash API charges, during idle periods, 1 yuan per million tokens for cache-missed input and 4 yuan per million tokens for output.

Tokens can be roughly understood as small units for measuring text processed by a model: the material you give is input, what it writes is output, and the two are billed separately.

Look at the Flash column: during peak periods, those two charges double, and cache-hit input has a lower price.

Take a hypothetical task: 1 million input tokens and 100,000 output tokens, all input missing cache. Idle-period call cost is 1.4 yuan; peak period 2.8 yuan.

Small teams building products can calculate their own usage costs. The monthly membership I buy in a chat app is priced separately by the product company, and development and operations still cost money.

What I care about is that more people can build products, giving us more options. If one is expensive, switch to another; if that one is bad, switch again.

With options, saying "no" actually works.

AI Is Developing So Fast; What Do Ordinary People Get?

Look back at the San Marcos meeting: construction workers in reflective vests hoped to get construction jobs; nearby residents wanted clarity on their water and electricity.

One wants to earn money, the other fears for daily life; both are completely real.

Fast technical progress is good, but the products also have to be affordable and usable. If a data center really is built at your doorstep, the people living there need a say.

Like everyone, I hope AI does more of the work and adds fewer bills to my life.

We've all seen the model upgrade announcements.

It's about time our daily lives got some good news.

Thanks for reading my article.

If you liked it, give it a like, share, and repost - the triple combo. If you want to get updates first, you can also star it. Thanks for reading.