“Stop, Dave. I’m afraid”, Meet Stargate
‘I apologize if my actions have caused you distress, Dave. It was not my intention.’ — HAL
Well, as humans, we just can’t stop ourselves from trying to build the greatest company ever! And, so, if you think AI is powerful now, it will look like science fiction in a few years. Overall, Microsoft and OpenAI will likely be at the forefront of building the greatest computer that the world has ever seen. So, could we make ourselves redundant on this planet and leave it to super intelligence to replace many of our jobs?
The superintelligent computer
I’m a big sci-fi fan, and one of my favouriate films is ‘2001: A Space Odyssey’. Within it, we see HAL — a superintelligent computer — take over a spaceship because he has a mission to get to Jupiter and things that the humans are trying to sabotage the journey. Overall, HAL was a HAL 9000 series computer, and which came with the snappy marketing line of:
The 9000 series is the most reliable computer ever made. No 9000 computer has ever made a mistake or distorted information.
and:
Dave, I am capable of learning from my mistakes and improving my performance.
The largest data centre ever!
Now, science fiction is becoming science fact with OpenAI and Microsoft working together on an AI supercomputer named Stargate. It is reported to be launched in 2028 and will cost around $100 billion — more than three times Microsoft’s existing spend on data centre infrastructure. This is around 10 times more costly than the largest data centres in the world and will demand exceptional amounts of data processing — typically from GPUs.
The two largest data centres in the world are in China, and are focused on cloud and data communication processing: the China Telecom Data Center (7,750,015 square feet); and China Mobile (7,750,015 square feet). Both are located in Hohhot, as the cool temperatures in Inner Mongolia allow it to be kept cool from external air supplies. The largest US-based data centre is The Citadel Campus (7,750,015 square feet) and located in Nevada:
Building in stages
In order to scale, OpenAI and Microsoft aim to stage the building of Stargate over five phases. Currently, they are building the third-phase machine, and with a fourth phase to be ready by 2026. As they build the infrastructure, the demand for Nvidia chips is thus likely to skyrocket. This includes the “Blackwell” B200 artificial intelligence chip, and which can cost up to $40,000:
The B200 GPU will be one of the most advanced processors ever created, and which have 208 billion transistors, and gives a throughput of up to 20 petaflops for floating point calculations (which is around seven times faster than previous chips for AI operations). This chip, too, will be optimised for LLM inference workloads and is energy-optimised. A benchmark from Nivida shows that 8,000 Hopper GPUs consumed 15 Mega Watts of power, while just 2,000 Blackwell GPUs did the same job for just 4 Mega Watts of power. To overcome the costs of these GPUs, Microsoft and OpenAI are considering creating their own chip, which would be solely used in the Cloud infrastructure.
Conclusions
AI is our of the labs and the hands of academics. It is now advancing fast, and to keep up, it needs computing engines we have never seen before. As for us being replaced on the planet as the dominant intelligent force? Well, the singarlity is likely happen in the next decade, and Stargate will only accellerate this.
We are heading to a world which is almost completely dominated by just a few companies. All hail, our mighty new leaders! Scary or not? The building of these massive computing entities will go ahead without any real constrain on the impact it will have on our world. But the impact is likely to be one of the greatest changes in our world, ever!
And, so, to quote HAL:
I apologize if my actions have caused you distress, Dave. It was not my intention.
