Author: Ejaaz
Compiled by: TechFlow TechFlow
TechFlow Summary: This article is a direct rebuttal to recent pessimistic views on xAI. The author systematically demonstrates, from four dimensions—computing power, data, distribution channels, and physical AI—why xAI might overtake all competitors by 2026.
The core argument is straightforward : while others are still discussing model architecture, Musk is already building his own power grid, airlifting gas turbines, and feeding data with Tesla robots. This is an analysis with a clear stance and is worth reading.
The main text is as follows:
I've seen too much criticism of xAI lately, and this article is meant to clarify things.
I will break down this judgment systematically: xAI's upcoming Grok 5 is not just about catching up with its competitors, but about surpassing them.
Don't forget, we're talking about a company that's only been around for two years. But they built the world's largest supercomputer in 122 days (which normally takes four years), have 600 million monthly active users, and possess something that no other AI lab has—a physical platform (yes, autonomous robots).
Enough with the small talk, let's get straight to the point.
Musk is building his own power grid.
By 2026, xAI's advantage in computing power will be overwhelming. Their current real-time computing power (approximately 500,000 GPUs) is more than Anthropic and Meta combined.
And that's not all. Including Colossus I and II, he plans to have 900,000 GPUs online by the second quarter of 2026. With the recently announced Colossus III (yes, another new data center has already started construction), it is expected to reach 1 million GPUs, with a total investment of $35 billion.
How can anyone else catch up with this scale?
But the question isn't just about how much money they spent or how much hardware they used, but how they did it. See this tweet:
Elon directly airlifted gas turbines to power the data centers because the power grids in Tennessee and Memphis simply couldn't handle the load. These turbines alone could power an additional 600,000 GPUs.
He chose to completely bypass the entire state's power grid (a process that typically takes several years) to speed up model training. In addition, he deployed approximately 250MW of Tesla Megapack energy storage to address situations where the grid couldn't keep up with demand during peak hours.
This forward-thinking approach, coupled with its extremely fast execution speed, is giving xAI a significant computing power advantage over its competitors.
You have to understand, this involves regulatory approvals, talent recruitment, and logistics operations—no one has ever done it on this scale before. xAI not only did it, but made it look effortless.
If the assumption that "more computing power = stronger model" holds true (which currently seems to be the case), then the rumored 7 trillion parameters of Grok 5 will be a monstrous model upon its release. For comparison, Grok 4 had 3 trillion parameters, meaning this is more than double.
Nvidia CEO Jensen Huang talks about Grok 5:
"Elon has already mentioned that the next cutting-edge model, which is the next version of Grok, namely Grok 5, will be a model with 7 trillion parameters."
The infrastructure race is no longer in doubt.
There is no longer any infrastructure expansion war, because xAI has already won. Their strategy is "do it first, worry later," and unless other labs catch up, xAI's models will remain ahead.
X's "X Factor": Unlocking Personal AI
xAI has the strongest computing power of any other AI, but top-tier models still require massive amounts of data.
And it's not just any data. AI labs are increasingly realizing that real-time data is the key to unlocking personalized AI—an AI that deeply understands your desires and goals, and executes them for you before you even think of them.
Google's newly released "Personal Intelligence" product is the clearest signal that models are ultimately heading in this direction. However, xAI has a unique advantage here that Google lacks:
A social media platform that feeds them over 100 million posts every day.
This means that over 100 million texts, images, and videos can be used to train Grok to achieve:
- Real-time trends and breaking news
- Large-scale understanding of viral spread, trends, and human behavior
- Real-time sensing of global cultural pulse
Other models can only tell you what happened, but Grok can tell you both what happened and how people feel about it—and faster than anyone else.
This ability is valuable.
If we assume that users gain 10 times the value from a tailored AI model compared to a general-purpose model, then X's competitive advantage will be very difficult to overcome.
Not only data, X's distribution capabilities are also phenomenal:
- 70 million daily active users
- 600 million monthly active users
- Each post has an "Ask Grok" button next to it.
It's easy to imagine that xAI will eventually integrate real-time prediction services for markets, shopping, banking, dating, and more into a single app, all powered by Grok.
Currently, most model labs are valued based on the number of GPUs, smart benchmarks, and reputation. xAI has it all, and also has the opportunity to penetrate multiple internet monopolies—and don't forget their goal is to become the "all-purpose app."
Today, X's recommendation algorithm is powered by Grok, which analyzes each post to make recommendations. Tomorrow, it will provide personalized intelligent services for each user.
Grok is clearly not just a standard large language model, and its valuation should reflect this.
Physical AI Advantages: xAI is the most forward-thinking lab
It's no secret that robots will have a huge impact on the world in the next five years. The technology has finally matured.
From manual labor in factories to "last mile" delivery, from fast food chains to top surgeons, all will be assisted or completely replaced by robots.
A viral video from Boston Dynamics over a decade ago has now snowballed into a fleet of self-driving cars and (surprisingly) impressive humanoid robots. Frankly, when you think of those two things, only one company comes to mind: Tesla.
A car that drives better than a human is no longer a fantasy. The latest v14.2.2.3 update is technically a better driver than you. Once the regulations are passed, you'll see self-driving Teslas picking up people everywhere.
Similarly, a humanoid robot that can carry your shopping bags and carefully polish your mother's delicate porcelain is becoming a reality. Optimus will begin mass production and enter homes and factories by the end of this year.
What does this have to do with xAI?
Two things:
- Machines need a brain to drive them, and Tesla uses Grok.
- Grok needs diverse data sources to understand the world around it, and this data comes from Tesla's robots.
This symbiotic relationship gives xAI an almost unfair advantage over its competitors. I think Google is the only company that can compete at this level, but they are still lagging behind.
Grok is already driving Tesla vehicles today—the latest update lets you have Grok drive you somewhere while playing music and telling you about Roman history.
Similarly, Grok is now receiving video data from Tesla's cameras, distance data from Tesla's sensors, and more, helping it understand the physical laws of the real world, visual perception, and navigation.
All of this data can now help it become stronger in other areas, such as generating video content that is more physically accurate.
It must be acknowledged that Musk is playing a five-dimensional game. He's not just building a large language model; he's creating an entire ecosystem for AI to live and function.
As I write this, I admit that all of this sounds wonderful, but it's also extremely ambitious... This leads to the final part of the article:
Yes, there are risks here.
Everything carries risk. Perhaps managing five companies simultaneously is Elon's limit, six would be too many... but I have my doubts. If there's anyone in this world who has repeatedly proven himself capable of silencing his doubters time and time again, it's him.
I don't care if you call me crazy—what he has already accomplished is itself extremely impossible.
I believe there are three key risks:
Elon Musk, the king of controversy , is a familiar face in the headlines. He's currently embroiled in a $130 billion lawsuit with OpenAI and is under investigation by regulators in the EU and India. Who knows what outrageous things he might do to ruin the whole vision?
Execution risks – xAI burns through about $1 billion a month, which is a huge bill. And Elon Musk oversees five companies (not including Starlink).
Scaling Laws – xAI has staked everything on "more computing power = stronger model," but this assumption could be overturned if a new and better training architecture is discovered. Andrej Karpathy has repeatedly stated that he believes large language models are not the final form.
That's all! I think people have been unfairly criticizing xAI's efforts to push the frontiers of intelligence lately, and seem to have forgotten that they are still a force to be reckoned with.
Hopefully, this article will change your perspective. Thank you for reading.
