Why Google Chose India for Its Biggest AI Data Centers Outside America — AB Explained

[00:00:00] Speaker 1: Is it just me or is there a bit of AI fatigue going on right now? All you hear on the news is that AI is going to take people's jobs away. On social media, there are so many fake AI-generated videos flooding your feeds that you cannot even tell what's real anymore. Hell, at this point, some people genuinely seem to believe that I'm not even real, that all this is fake. The point is, AI is so deeply ingrained in our everyday lives now that we cannot go five seconds without running into it. Almost everybody I know, including my parents, uses tools like ChatGPT or Gemini every single day. But have you ever stopped to think about what actually happens the moment you type a question into ChatGPT? Because that query doesn't just disappear into thin air. It must travel somewhere to a physical place, right? Maybe you've heard the word the cloud thrown around, like your photos and your work are all saved in the cloud. But even the cloud has to live somewhere. I don't know, like in some building that has to stay powered 24 hours a day, seven days a week. Those buildings are called data centers. And right now, tech companies are racing to build them at a scale the world has never seen before. There are already more than 10,000 data centers operating around the world. And Google, one of the most powerful tech companies on Earth, announced in 2025 that it was spending 15 billion dollars to build its largest AI data center campus outside of the United States. The location? India. Specifically, a coastal city in India called Vishakhapatnam, or Visakh for short. Construction in Visakh only began in April 2026, right around the time data center projects in the US were facing growing pushback from local communities over power, water, and land use. Hey, I'll be honest. Before I started looking into this, I had no idea what a data center actually was. I mean, don't get me wrong. I've seen the photos and videos of those gigantic buildings, supposedly using a shit ton of water and electricity. But I never really understood what's actually inside them, what they do exactly, or why do they need to be so enormous, and why they're suddenly causing so much controversy. So let's figure this thing out together. The goal here is that by the time you finish this video, you'll have a much deeper understanding of what a data center is and why Google is building its largest one outside the US and India of all places, especially at a time when a significant number of planned data center projects in the US are being delayed or canceled. There are so many reasons we found, and one of them has to do with India's recently introduced data protection law. Speaking of data protection law, if the government of India is going this far to prevent large US tech companies from transferring Indian user data to foreign servers, that's a good enough reason for you to think about protecting your own personal data online, even if you're not an Indian citizen. In this age of AI and tech companies scraping all the data online to train their models, think about how much personal information you've been handing out over the internet over the years. Your name, your address, your payment details. It's highly likely that your personal data has already ended up in the wrong hands. Right now, an entire industry of data brokers, people search websites, and background check companies are quietly scraping the internet to build highly detailed profiles on you. And that's exactly where scammers and identity thieves go to buy your personal data, use it to impersonate you, open credit cards in your name, and steal your money. Imagine someone else taking out a loan under your stolen identity, and you don't even realize it until it's too late. That's why you need Incogni. Incogni doesn't just remove you from data broker sites. Through their custom removals feature, Incogni targets the specific websites and platforms where your personal data is being sold and traded. They contact these companies on your behalf, demand that your information be deleted, and actually verify that it's been removed. They handle everything so you don't have to spend months doing it yourself. Think of it as taking yourself off the internet, the same internet where your data is being bought and sold every single day without your permission. If you want to protect your data from being exploited, try Incogni by scanning this QR code or using the link in the description. Use the code ASIANBOSS at incogni.com slash ASIANBOSS to get 60% off an annual plan. That's 60% off to protect your identity and remove yourself from the internet. I highly recommend taking advantage of this offer. Let's get back to the deep dive. So let's start with the most obvious question, which is, what is an AI data center? Actually, at a more fundamental level, you could even ask, what the hell is a data center without the AI component? You know, when I first thought about a data center, what I imagined was some sci-fi supercomputer with a single glowing brain. Maybe that's just me. Anyway, to understand what's actually happening inside these AI data centers and why they need so much space, let me take you back to the origin of data centers so we can see how they evolved over time. And here's the thing, from mainstream coverage, you might think the first data center only appeared in the early 2000s along with the dot-com boom. But from our research, you can actually trace their origin all the way back to 1943 when the US Army first commissioned what would become the world's first computer. In 1943, the United States was in the middle of World War II and the US Army had a problem. Artillery was one of the most devastating weapons in that war, but firing it accurately required precise calculations. Wind speed, shell weight, air resistance, the curvature of the earth. These calculations had to be compiled into what was called ballistic tables, worked out by hand by teams of mathematicians. The ballistic tables, all those pages of hand-calculated numbers, were literally data, not unlike a modern-day Excel spreadsheet. The only difference is those hand-calculated numbers on paper were analog data as opposed to digital data. And calculating a single trajectory by hand could take days. The Army needed something faster. So they commissioned a team of engineers at the University of Pennsylvania to build something they call the Electronic Numerical Integrator and Computer, or ENIAC. When it was completed in 1946, ENIAC filled an entire room, 30 feet by 50 feet, roughly 1,500 square feet, weighed 30 tons, and consumed 150 kilowatts of power. It cost roughly $400,000 to build, which is about $6 million in today's dollars. What they had built was a computer so powerful and demanding that it needed its own room, its own power supply, its own cooling system, and its own security. If you're wondering how it worked, a mathematician would punch the variables like wind speed, shell weight, and elevation angle onto a paper card and feed it into the machine. ENIAC would process it and spit out another punched card with the answer. A calculation that previously took a human mathematician anywhere from 20 to 40 hours. That took 30 seconds. So what do you think happened when calculations could be done that much faster? It created an output problem. The computer was printing out so many physical papers that they started piling up. Suddenly, they had a data storage problem. The army's solution was to fill entire rooms with filing cabinets. And eventually, the problem got so big that the army moved ENIAC itself from the University of Pennsylvania to a military facility in Maryland. Data in, data processed, data stored under the same roof. Sounds like a data center to me. Though at the time, it was just called a computer room. And over the next five decades, computers kept getting smaller, cheaper, and faster. The room-sized ENIAC gave way to IBM mainframes that still filled rooms but were far more powerful and far more reliable. Then came mini computers the size of a refrigerator. And eventually, companies like Apple and IBM created something small enough to sit on a desk, what we now call a personal computer or PC. But here was the problem. As computers became smaller and faster, they produced way more data. And the more data they produced, the bigger the storage problem became. Bear in mind, this was still before the internet existed. So every major company, your bank, your airline, your hospital, had to build its own dedicated room somewhere inside its own building, usually in the basement, just to house all of its machines and store all of its data. Every company was essentially running its own private servers. They still weren't officially called data centers, just server rooms or computer rooms. Then the internet exploded in the 1990s dot-com boom, and everything changed overnight. Before the internet, a computer's data was used internally, accessed by its own employees during business hours. After the internet, the same data now had to be served to millions of strangers simultaneously around the clock. A bank's website alone requires storing webpages, processing login requests, and handling transactions, all in real time for anyone in the world. The volume of data didn't just grow. In a relatively short span of time, the amount of data that needed to be stored exploded, and no office baseman could handle that. So companies started building massive dedicated warehouses just to house their servers. And somewhere along the way, the tech industry stopped calling them computer rooms. They became data centers. Actually, nobody made an official announcement from what we could find. I guess the name just stuck. Here's a fun fact. The world's first recognized modern data center was opened by a company called Exodus Communications in Santa Clara, California in 1996, right at the start of the dot-com boom. and it was just 15,000 square feet, roughly the size of a large supermarket. Exodus was renting out space and connectivity in a model called co-location. Companies like Hotmail, Yahoo, and eBay would buy their own server computers, physically ship them to the Exodus building, and rent a small cage section, literally a chain-linked fence enclosure, to house their machines. Even a tiny startup called Google rented its very first cage inside that same facility in 1999, seven feet by four feet, about 30 computers on shelves. That little cage was Google's entire data center. Exodus then provided the industrial power, the cooling systems, and most importantly, a direct high-speed fiber connection to the internet that all these computers couldn't get from their office. Then, of course, the internet bubble burst and Exodus filed for bankruptcy in 2001. But the dot-com bubble had accidentally overbuilt an enormous amount of data center infrastructure. When the bubble burst, all of their real estate, fiber cables, building, server rooms became available for almost nothing. The cables were still in the ground. The cooling systems were still humming. So companies like Google snapped them up for pennies on the dollar. By the way, take a guess how much electricity the Exodus data center was consuming at the time. The whole thing ran on somewhere between one and five megawatts of electricity. If you're like me, you might have no idea what that means in practice. One megawatt can roughly power 800 to 1,000 average American homes. So five megawatts would power around 4,000 to 5,000 homes simultaneously around the clock. So even at this early stage, you can see data centers were already getting pretty big, powering most of the internet companies you ever heard of. But after the dot-com crash, a company most people knew as an online bookstore quietly launched a service called Amazon Web Services in 2006. And it changed the entire concept of what a data center was for. There's actually a popular myth about how this all started. That Amazon had to build a bunch of spare servers sitting idle after the holiday rush and decided to rent them out. Not true, according to the people who actually built it. What really happened is that running a massive e-commerce operation meant Amazon had to deal with these huge, unpredictable spikes and traffic and they got really good at building infrastructure that could flexibly scale up and down on demand. At some point, somebody at Amazon realized other companies would pay good money for exactly that kind of flexibility. That idea became the cloud. Instead of physically shipping your own servers to someone's building, you just logged onto Amazon's website and rented a slice of their physical server over the internet at a fee. And here's what made it so clever. Using software, Amazon could take one physical server and divide it into up to six separate virtual servers simultaneously. The result was that client companies didn't need any hardware at all. They just paid for what they used. Google and Microsoft eventually built their own versions. And the money was extraordinary. That simple idea of renting spare servers eventually grew into a $128 billion a year business for Amazon, generating more than 60% of all of Amazon's profits in 2024. Of course, this meant that Amazon had to keep building more and more data centers because what happened next caused server demand to explode. Three things happened actually. First, the smartphone. Apple launched the iPhone in 2007 and today, there are nearly 5.8 billion smartphone users on Earth, more than 70% of the entire planet. before smartphones, the internet was mostly text and simple web pages. After smartphones, everyone had a high-resolution camera in their pocket generating photos and videos. WhatsApp alone sees roughly 7 billion photos shared every single day. Every one of those photos has to live somewhere, right? Second, video streaming. Netflix launched a streaming service in 2007 as well. YouTube already existed, but exploded in scale. Today, more than 500 hours of videos are uploaded to YouTube every single minute. A single 4K movie can consume anywhere from 7 to 100 gigabytes of storage depending on quality. Multiply that by hundreds of millions of subscribers streaming simultaneously and you start to understand why data centers had to keep getting bigger. Third, automatic cloud backup. Apple launched iCloud in 2011 and Google Drive came along in 2012 before your photos lived on your phone or your hard drive. After, every photo you took automatically copied itself to a data center somewhere without you doing anything. Every photo and video ever took. The average person went from roughly 300 digital interactions per day in 2010 to nearly 5,000 by 2025. Every single one generated data that had to be stored somewhere. So yeah, Amazon just kept building and building and building. Google and Microsoft have followed with their own cloud services and data centers. By the early 2020s, a single hyperscale data center was pulling up to 100 megawatts of power. Remember our 4,000 to 5,000 homes from the Exodus era? A 100 megawatt cloud data center powers roughly 80,000 to 100,000 homes simultaneously. That's a lot. But still, no one cared. The data centers weren't all over the media and there was no visible public backlash. More importantly, maintaining the data centers was somewhat manageable because engineers had gotten really good at squeezing more computing out of the same amount of power. And that's part of why almost no one outside the tech industry was even aware these cloud data centers existed. Then AI arrived. In November 2022, a US tech company called OpenAI launched a product called ChatGPT, an AI tool that completely changed the way the general public views AI and made Google search feel outdated, like flipping through an old phone book or something. Within five days, it had a million users. And within two months, it had a hundred million, the fastest adoption of any consumer product in history. Suddenly, everyone, including myself, started using ChatGPT for pretty much everything. But do you know what exactly happens the moment you type a question into ChatGPT? The moment you hit send, your question gets broken down into small units called tokens. A short, common word like dog is roughly one token. A longer, more complex word might get split into two or three tokens. Those tokens get sent as a standard encrypted data packet through your internet connection the same way on email travels and into a data center. There, thousands of specialized chips called graphics processing units or GPUs run your tokens through layers after layers of mathematical calculations until they produce a response. That process is called inference. We did a full explainer on how inference actually works in our previous video, so check that out after this one. To give you a quick recap, AI tools like ChatGPT, Claude, and Gemini are what's called large language models. Massive neural networks trained on essentially all the texts and images they could scrape from the internet. Inference is how they predict the most likely next word, not unlike how a human uses past knowledge and pattern recognition to answer a question, except they're doing it entirely through mathematical calculations, billions of times per second. So can you imagine how much computing power this requires? According to the International Energy Agency, a single ChatGPT query consumes about 10 times more electricity than a regular Google search. 10 times just for one question. Now, multiply that by 100 million users asking questions simultaneously around the clock every single day. And remember, before ChatGPT, we were still in the cloud data center era. OpenAI didn't own a single data center when ChatGPT launched, so they had to rely on infrastructure that Microsoft had built for them. As a side note, Microsoft invested a billion dollars in OpenAI in 2019 and then another 10 billion in January 2023. Just to handle the AI training for ChatGPT, Microsoft had to specifically build a custom supercomputer with 10,000 specialized GPUs, which became one of the five most powerful computers in the world at the time. There was simply no way regular cloud servers powered by central processing units or CPUs could handle this. Just for your reference, GPUs are up to 100 times faster than CPUs for AI workloads. And even with all that, in the early days of ChatGPT running on the GPT 3.5 model, the system still crashed. Do you remember the now infamous message? ChatGPT is at capacity right now. Training GPT 4 alone required 25,000 of these GPU chips running nonstop for 100 days straight at a cost of $100 million. When GPT 5 launched in 2025, OpenAI needed over 200,000 GPUs just to serve responses to users, 20 times more chips than the original supercomputer Microsoft built. And that was just for inference, just for answering prompts. And now in 2026, we are already in the GPT 5.5 era, where ChatGPT alone processes 2.5 billion prompts every single day. That's roughly 29,000 questions every second. OpenAI has already committed to scaling to millions of GPUs over the next few years. That's just for OpenAI. We haven't even touched Google yet. Talk about the escalation here from 10,000 GPUs to millions of GPUs in just a few years. Can you see how that requires data centers at a scale that almost nobody was prepared for? I say almost because there was one company that everybody seemed to write off as soon as ChatGPT launched, but in hindsight was better positioned and prepared than any other company with more of the foundational infrastructure already in place than its competitors. That company was Google. As it turns out, Google already had the largest AI research team in the world. And here's the wild part. Google had essentially built ChatGPT before ChatGPT even existed. One of the most brilliant researchers was a man named Noam Shazir. He was actually one of the core authors of the legendary 2017 Google paper that invented a transformer, the literal underlying math that powers all of modern AI today. While at Google, Shazir built two revolutionary conversational chatbots, Mina and Lambda, that could hold sophisticated conversations on virtually any topic. Shazir was so convinced of Mina's potential that he wrote a memo to Google leadership predicting it would eventually replace Google search. But Google refused to release it citing safety concerns. So in 2021, Noam Shazir left Google to start a new AI chatbot startup called Character AI. Then OpenAI launched ChatGPT in 2022 and Google completely panicked. In August 2024, just three years after Noam Shazir walked out the door, Google paid a staggering $2.7 billion just to license Character AI's technology on the condition that Shazir would return to the company. So he went back to Google DeepMind not just as a newly minted billionaire, but also as a technical lead of the entire Gemini project. When I visited the main Google campus, I heard stories from their engineers about how this billionaire guy just walks around in his old cab like a regular dude working harder than anyone else because he's just so passionate about the work. That's pretty cool. Anyway, many industry insiders point to his return as the exact moment Gemini finally woke up and started getting competitive with or even better than ChatGPT. Now, at this point, you might be wondering ChatGPT required Microsoft to build an entirely new AI data center for its training. So where was infrastructure for Gemini? Where are Google's AI data centers? Well, Google already had data center clusters all over the U.S., including their very first one in Oregon opened back in 2006. But the two most important ones for AI training are in Columbus, Ohio and Council Bluffs, Iowa. The Iowa campus was one of Google's earliest facilities first built in 2009, originally meant for cloud workloads like Google Search, Google Maps, and Google Docs. For Iowa specifically, what Google had to do was take a hybrid approach, retrofitting those old cloud buildings from around 2017 while simultaneously building brand new AI-specific facilities right next to them from 2022 onward. Ohio, on the other hand, is a cluster built specifically with AI in mind from the start and is now the largest Google's AI training facilities, the physical place where Gemini essentially lives. Now, as far as retrofitting, you might think you can just walk into one of these old legacy buildings and swap out the old chips for new ones. You can't because overhauling a cloud data center for AI is a brutal engineering challenge. First, you have to rip out and reinforce the floors because AI racks are much heavier. Take one of the most common AI racks right now, the NVIDIA GB200 which packs 72 GPUs into a single liquid-cooled unit. It weighs 3,000 pounds and costs up to $3 million just for one single rack. A standard cloud server rack from a decade ago weighed around 500 pounds. So the floor of an old cloud data centers would literally crack under the weight of something six times heavier. Speaking of the floors, old cloud data centers had their fiber optic cables running on the raised floor tiles. Think of those cables as the nervous system connecting all the CPUs running at roughly 10 to 100 gigabytes per second. But when you are building a brand new AI native data center, the fiber is usually installed overhead, suspended from the ceiling in metal cable trays. These cables are super high speed running at 400 to 800 gigabytes per second. Fast enough for every single GPU in the building to communicate with every other one simultaneously during a training run. A single AI data center campus can contain hundreds of miles of these lines. So why not just run them through the floors? Because the floor space is already completely taken up by the liquid cooling infrastructure. And that cooling system, as you're about to see, is a massive engineering nightmare in itself. Think about how we used to handle heat. Traditional cloud data centers rely on industrial-scale air conditioning. Kind of like a massive fridge keeping a room chilled. But AI racks produce up to 35 times more heat than traditional cloud server racks. A single AI chip at full load runs at up to 180 degrees Fahrenheit or 83 degrees Celsius at its surface, which is hard enough to burn your skin on contact. And just one rack containing 72 of those chips generates the same continuous heat as 80 household space heaters packed into a box the size of a refrigerator. At that density, standard air conditioning is completely useless. The air simply cannot move fast enough to stop the chips from overheating. So tech companies like Google had to switch to liquid cooling. Basically water. Why? Because liquid is over 3,500 times more efficient at absorbing and moving heat than air. Here's how it works. Cold water gets pumped through pipes running along the floor directly into the server racks, flowing into flat copper-cold plates clamped directly on top of each chip. The water absorbs the heat, turns warm, and gets pumped back out of the building to massive external cooling towers to shed that heat before looping back in. Round and round, 24 hours a day. And that loop consumes an enormous amount of water. A single large AI data center can consume up to 5 million gallons of water every single day just for cooling. Okay, so we've covered the main physical components of an AI data center, right? GPU racks, fiber cables, cooling systems. You're looking at an absolutely fucking massive operation. These new AI native data centers are way bigger than shopping malls. They're even bigger than major airports. Google's site Invisag alone covers 601 acres. That's roughly 450 football fields of land. And to run an empire that size, you need raw power, electricity. We're not talking about megawatts anymore. We're talking about gigawatts, where one single gigawatt equals 1,000 megawatts. Each of Google's major AI clusters, like the ones in Ohio and Iowa, consists of four to six individual data centers that total around one gigawatt per cluster. To put that into perspective, one gigawatt is enough to power roughly 800,000 standard homes simultaneously. That is double the number of households in the entire city of Seattle. So between its two main U.S. training clusters alone, Google draws roughly two gigawatts of continuous power every single day of the year. Since these data centers consume such an astronomical amount of power, you will think companies like Google would have just built their own power supplies from day one, right? Well, up until now, no. They just plugged straight into the public grid because it was cheaper and easier. The issue here is that the American grid was simply never built for this kind of demand, and that creates two problems. First, electricity bill for every household in the region started going up. In effect, the average American household was subsidizing data center power demand that most of them didn't even know existed. Second, even the existing public grid doesn't produce enough power for new AI data centers. In states like Virginia, getting a new data center connected to the grid now takes up to seven years because the regional grid operator has to upgrade local infrastructure to handle the extra load. As of late 2024, the total capacity of new power projects waiting in line to connect to the grid exceeded the entire install generating capacity of the American power grid. So while they wait for public grid connections and to protect against power outages, every data center keeps massive diesel generators parked just outside the building, each one the size of a large truck existing purely for emergencies. if the grid fails even just for one second during an AI training run, these kick in automatically. But some companies tired of waiting years for grid connections have gone further and built their own on-site power plants using natural gas turbines, which are basically jet engines bolted to the ground. Elon Musk's AI data center in Memphis is the most extreme example. Running over 30 of these turbines continuously, Google still relies primarily on the public grid combined with renewable energy contracts, though they're betting on nuclear power for the future. And look, I'm going into all of this detail to give you some idea of how challenging it is to build an AI data center in the United States right now. Years of waiting for grid connections, rising electricity bills for local communities, and mounting environmental regulations. And that's even before they start exploring nuclear power options. So what is it about AI data centers that local communities are complaining about exactly? And why does Google think India is the answer? Let's explore that next. Here is the biggest controversy besides rising electricity costs that is turning local communities against AI data centers. We're talking about a more visceral problem, one that directly damages people's health. Constant, relentless noise. Through our research, we can track this down to three distinct sources. The first source is inside the data center building, the server fans. Every single server rack has its own built-in cooling fans running non-stop, 24 hours a day. If you're standing inside one of these facilities near the server areas, noise levels reach up to 96 decibels, loud enough that workers are legally required to wear ear protection. Think of standing next to a lawnmower that never turns off. Here is the second source, the cooling tower fans outside the building. These massive industrial fans, some stretching up to 30 feet or 9 meters in diameter, spin continuously to cool the water before it loops back inside. They generate about 70 decibels of noise at a distance of 50 feet or 15 meters. By the time you get 200 feet away, which is the minimum distance new data centers in Virginia are required to keep from residential homes, that noise drops to about 60 decibels. Now, Virginia has more data centers than anywhere else in the world, and 200 feet means residents are living within 60 decibels of constant noise around the clock. 60 decibels is roughly the level of busy street. So data center operators love to point at their decibel meters and say, look, the noise is within legal limits. And technically, they're right. But here's what those measurements miss entirely. Standard decibel meters largely ignore infrasound, low frequency vibrations below the range of human hearing. You don't hear infrasound with your ears. You feel it in your chest, your bones, your internal organs. So it's entirely possible that the server fans noise bleeding through the walls of the data center and the massive cooling tower fans combine to emit a constant low frequency pressure wave that never stops. Residents near these facilities report chronic dizziness, nausea, and a complete inability to sleep because their houses are subtly vibrating. And all of this happens before the situation gets even worse. Enter the third source, the gas turbines. I already mentioned that data centers are increasingly bolting natural gas jet engines to the ground to generate their own power on site. At close range, they produce over 100 decibels, as loud as an ambulance siren right next to your ear. And that sounds travels much further than you'd expect. So when residents living next to one of these facilities say it sounds like a commercial jet permanently taken off in their backyard, they're not exaggerating. That is literally what it is. So when you hear completely conflicting accounts about data center noise, they could all be telling the truth. The data center experts and operators keep pointing to the acceptable level of noise from the cooling tower fans while locals are complaining about the low frequency vibrations and the roaring jet engines of the power generators. And all of the noise, the grid delays and the community pushback is what gets us to India. So why India? Because, let's be honest, the outside perception of India is that it's a really hot country. Here is a city of Visakh sitting on the eastern coastline of India on the Bay of Bengal, about 700 miles southeast of Mumbai. The temperature in Visakh peaks at 97 degrees Fahrenheit or 36 degrees Celsius in May and humidity climbs to 83% during monsoon season. Does that sound like the right environment for data centers to you? Before Google's announcement, Visakh had zero AI data center infrastructure. It's more known for its naval base, its steel plants, and its shipyards. But what it does have is a coastline, which matters a lot. The fiber optic cables that carry the internet across oceans need to physically land somewhere on a coast. So Google is basically turning Visakh into a brand new global internet gateway. It also has a willing state government and space. Lots of space. Oh, and here's a fun fact. Google CEO Sundar Pichai grew up in Chennai, about 370 miles from Visakh, in the same region of South India. He left India to the United States on a scholarship to eventually run Google. and he's now returning to the same region he grew up in to turn it into a powerful AI hub. The state government has committed to a combined 5 gigawatts of data center capacity in Visakh alone. That is three times everything India built in the past 30 years. But Pichai's personal connection to the region is just one part of the story. We identified quite a few compelling reasons for Google's decisions, but let me give you the two that might not be talked about as much. Reason number one, India generates roughly 20% of the world's digital data. That comes from nearly 900 million internet users, people searching on Google, watching YouTube, messaging on WhatsApp, and increasingly using AI tools like Gemini. So you can see why the Indian market is super important for Google's global expansion beyond the US and Europe. And yes, there is the Chinese market, but China is completely closed off to Google. China has its own powerful domestic AI tools like DeepSeq and Seedense 2.0. So India is really the last major untapped AI market at scale. And this is not just Google seeing the opportunity. In the same week that Google announced its $15 billion Visakh investment, Microsoft committed $17.5 billion to India, its largest investment anywhere in Asia. Amazon Web Services pledged a $35 billion. The AI race for India is very real. And to win it, Google has to serve those 900 million users better. Right now, every time one of those 900 million Indians uses Gemini, that query has to travel to an AI server in Singapore or the US and back. Engineers call that delayed latency. At AI scale, where each query is 10 times more compute intensive than a regular Google search, that round trip can make Gemini feel slow. Building AI centers in India solves that problem. But there's also a legal dimension that makes this even more urgent. India passed the data protection law in 2023 called the Digital Personal Data Protection Act. It gives the Indian government power to restrict cross-border transfers of Indian user data to foreign servers. In other words, Google could potentially lose the right to process Indian user searches, emails, and payments outside India entirely, unless it builds the infrastructure inside the country. But to me, this begs an interesting question. India passes law to protect Indian citizens' data from being controlled by foreign entities. Fair enough. But the machines processing the data are owned by Google, an American corporation, and therefore accessible to American authorities under U.S. law. So who's really in control of Indian data? Maybe one could argue that the Indian data protection law was never about data sovereignty, and it was merely a convenient way to give Google a legal reason to act faster in investing in India. Something to think about. Reason number two, extraordinary cost and infrastructure advantages. Building an AI data center comes down to two main costs, construction and land. Construction costs are typically measured per megawatt of capacity. In VSEC, that runs about $7 million per megawatt, a 30% cheaper than the U.S., where the same megawatt costs around $10 million. For Google's one gigawatt VSEC campus, that difference alone saves roughly $3 billion in construction costs before a single GPU is purchased. As for the land, in Virginia, it hit an average of $244,000 per acre in 2024 and jumped 23% in a single year as AI data centers competed for the same shrinking pool of available sites. In VSEC, the state government allocated 601 acres as part of the investment deal at a 25% discount on market price, which is already a fraction of Virginia rates. and on top of that, committed roughly $2.45 billion in total state incentives, including water and electricity subsidies over the life of the project. Can you now see the billions upon billions of dollars that Google is saving by building its AI data centers in India? But there's more. In February 2026, India's central government announced a 20-year tax exemption for data centers, zero corporate tax until 2047 for Google and any other foreign companies that route global cloud and AI services through Indian data centers. So everything sounds great for Google, right? I mean, the financial case for building in India rather than the US is not even close. Here's the problem though. Two problems actually. Not for Google but for Indian locals having to live with data centers in their neighborhoods. First, power. Like I said, Google's new AI campus alone will draw one gigawatt, about 3.2% of the entire state's power capacity going to one private data center. And that grid was already struggling before Google arrived. An investigative report found that farms and households surrounding the construction sites are already experiencing regular power cuts. To be fair, Google has committed to powering the campus with 80% renewable energy through new solar and hydro projects along the coastline. But that'll take years to build. The AI data centers need power now. Second, water. Remember we mentioned that a single large AI data center can consume up to 5 million gallons of water every single day just for cooling? India has 18% of the world's population but only 4% of its water resources making it one of the most water-stressed countries on Earth. Vizak's drinking water supply depends on a small number of reservoirs that are already facing serious depletion. One of Google's three construction sites sits just 400 feet or 120 meters from one of those main reservoirs. Does that mean Google will start tapping into that water to source the millions of gallons it needs daily for cooling? Nobody knows because Google's confirmed water sourcing plan for Vizak has never been made public. The water allocation agreement hasn't even been disclosed. But one can reasonably infer where the water will come from when you are building a facility that needs millions of gallons daily right next to the city's drinking water source. Here is where things get very different from the United States. In Virginia, local communities could push back because their local governments were on their side. Town hall meetings were packed. Lawsuits were filed. Projects worth $64 billion were blocked or delayed. In Vizak, all three levels of government from local to state to central made this project a national priority before residents even knew what was happening. When locals raised concerns, they were fighting both the corporations and their own government. And when a video documenting the project's impact on local communities went viral on Instagram with 2.6 million views, the Indian government ordered it blocked within three days. So why would India at all governmental levels approve such plans seemingly at the expense of its local citizens? The uncomfortable truth here is that it brings enormous investment into India and builds the country's reputation as a global AI infrastructure powerhouse. The Indian government clearly believes that these benefits will outweigh the costs experienced by locals. Then there are the jobs the government is promising. The state government promised 188,000 jobs. That sounds like a lot, but most of those jobs are temporary construction work. Once the campus is completed, a hyperscale AI data center pretty much runs itself. A single data center building the size of several city blocks typically employs around 50 full-time permanent staff. That's fewer permanent staff than your local Walmart. And even those jobs will not go to the local farmers whose hillsides are being flattened or the residents living next to the groundwater reservoir. Data centers hire specialized engineers and AI talent, not the local communities absorbing all the costs. So is there a better way to build AI data centers? Apparently, yes. And it's being proposed by Elon Musk. SpaceX literally just completed the largest IPO in history, raising $75 billion. And a significant part of that is earmarked for building AI data centers in space. So how does that work? SpaceX has already filed with the US government to launch up to 1 million data center satellites, each equipped with racks of AI chips and massive solar panels. And the appeal of space is obvious once you think about it. No land to buy. No water to drain from a city's drinking supply. No seven-year wait for a great connection. Instead, you get solar power that is about 36% stronger than on Earth's surface with no interruptions. And rather than cooling towers, the satellites use liquid radiation panels that dump heat directly into the void of space, which sits nearly at minus 454 Fahrenheit or minus 270 degrees Celsius. The temperature difference between the chips and the surrounding void is so extreme that the radiators can shed heat almost effortlessly. Of course, there are challenges. First, the full commercial deployment is only projected to begin in 2028 and beyond. So that's some time away. Second, there are real technical risks because cosmic radiation can silently corrupt AI computations. And in orbit, there is no maintenance crew or no way to swap out a damaged chip. Then there is the latency issue. The round-trip communication delay between orbit and Earth is inevitable to some degree, which could slow down certain AI applications that need instant responses. And it's not like every AI company can launch its own satellites. This means they all need to rely on SpaceX to rent orbital computing power. SpaceX is already charging Google roughly $920 million per month for AI compute access. Now you know why SpaceX's valuation is so high. They are the ultimate landlord in space. Either way, it always seems to come down to this. There are winners and losers when new technology disrupts the economy. And right now, the losers are the little guys with no voice or power. It's just that when it comes to AI, no matter how much we think the industry needs to pump the brakes and slow down a bit, it won't. The market forces want to allow it because that means you will fall behind the competition and lose billions of dollars already invested. So for now, it looks like the construction of new data centers will continue on Earth. More will be built in the US, a lot more will be built in India, and who knows where else. Because a lot of countries, especially in Asia, will line up to provide favorable conditions and incentives like tax breaks and subsidies to attract these giant AI companies. In fact, I'll be surprised if there weren't deals being made right now with communities nearby who have no idea what is about to be built next door. If you found this video insightful and even understandable, first of all, I really appreciate it because this was a lot to research and digest. Sometimes I wonder, hey, why am I the one to have to explain and deliver this type of information when there are so many professional journalists and experts out there? I think what makes us different from other media brands is that we don't pretend to be the subject matter experts ourselves. We do our best to research keep asking questions and really try to get insights that you cannot get from the mainstream media, Google search, or AI chatbots. And because we put a lot of time and effort into our research and storytelling, our uploading schedule can be unpredictable, which is why I'm asking you to subscribe and turn on the notification bell because unless you do, you won't even know when our new video is up. That'll hurt the YouTube algorithm and our views will suffer as a result. So if you agree that our explainers need to be seen by the broader global audience who appreciate authentic insights and news from Asia, turn on the notifications and watch our next video as soon as it's up. Oh, and just a little update. We've been testing our new live stream format and we think we're almost there to start doing it on a regular basis. Maybe weekly, but not sure yet. The goal here is to show you guys in real time how we discover insights for our explainers by talking and asking questions to locals directly. This means we're looking to feature many of our viewers who want to share their perspectives on particular topics we might cover in the future. So if you're interested in participating in our future live stream sessions, just fill out the form in the description with your video submission and I will personally reply. Because the topics are not set yet, we can even brainstorm together on live stream what the next video should be. All this to say, we are dedicated to being the most authentic media brand that gives voice to the people. That's been our approach for the past 13 years and we are working to amplify that effort. So if you're sick of seeing my face all the time, just get in touch and I'll be happy to give you the platform. Let's really cut through the mainstream media noise and be the platform for real people's voices and perspectives. Of course, I'm Stephen Park. Thank you for watching all the way to the end and as always, stay curious.