fr en es pt
Contact the author rss astronoo

Updated April 09, 2024

Artificial intelligence: the explosion of gigantism

Artificial intelligence: the explosion of gigantism

Description of the image: Artificial intelligence - The explosion of gigantism.
Carried away by its success, this great leap in quality comes at the cost of a colossal burst of energy.
This image was generated by artificial intelligence.

AI: Figures that make you dizzy!

Natural language processing (NLP), powered by the “Transformer” brick, is at the origin of the great success of artificial intelligence. However, other factors have enabled the emergence of AI, in particular the increasing availability of data, the storage power of data centers and the exponential increase in computing power. Indeed, the development and growing use of AI is accompanied by a growing demand for data, storage capacity and computing power.
Carried away by its success, this great leap in quality comes at the cost of a colossal burst of energy.

• The first source of this gigantism lies in the explosion of available data.
The rise of Internet technologies generates a volume of data that is growing exponentially.

The gigantic size and growth of internet data is dizzying.
The figure is so large that estimates are impossible or very imprecise.
According to some reports (Google Cloud Platform, AWS, Microsoft Azure, IDC, Gartner, Forrester Research, and others), the amount of data stored on the Internet in 2020 was somewhere between 40 and 50 zettabytes (1 byte is synonymous with 1 character or 1 byte). In 2021, it was around 60 to 70 zettabytes. In 2025, the likely figure is around 180 zettabytes. This quantity represents 180 billion 1 terabyte (TB) SSDs. The terabyte has become the storage size of the data media we use every day.

* 1 zettabytes (ZB) = 10^21 bytes, which is equivalent to 1,000 exabytes or 1,000,000 petabytes (PB) or 1,000,000,000 terabytes (TB).

In 2024, the estimate for the amount of internet data generated each day varies depending on the source, but it is generally around 3.5 exabytes. To put this daily amount of data (3.5 x 10^18 bytes) into perspective, we can compare it to 625,000,000 2-hour HD movies (≈ 4 GB of data).

According to a study conducted by Digital Universe, the data sphere doubles approximately every two years.

Power Symbol Power Symbol
10^0 1 10^1 da (deca)
10^2 h (hecto) 10^3 k (kilo)
10^6 M (mega) 10^9 G (giga)
10^12 T (tera) 10^15 P (peta)
10^18 E (exa) 10^21 Z (zetta)
10^24 Y (yotta) 10^27 R (ronna)
10^30 Q (quetta)

• The second source of this gigantism lies in hyperscale data centers (the largest data centers in the world).
The largest storage capacity is concentrated mainly among large players such as Amazon Web Services, Microsoft Azure, Google Cloud, Meta Platforms, Equinix, Digital Realty.
According to Omdia's report, there are approximately 8.6 million data centers worldwide in 2024.
Of these, around 600,000 are hyperscale data centers owned by cloud giants like Amazon, Microsoft and Google. Although difficult to estimate, it is likely that global data storage capacity doubles every two years.

Running data centers consumes a lot of energy and the growth of storage could put a strain on power grids. By 2025, data center power consumption is expected to reach 2% of global power consumption.

• The third source of this gigantism lies in the computing power of machines, measured in FLOPs (Floating-point Operations Per Second).

In 2024, processors (CPUs) can reach a computing power of 1228 GFLOPs, or approximately 1 teraflops like the processor Intel Core i9 -13900K.
Graphics coprocessors are even more powerful for performing specific tasks or mathematical calculations. These hardware accelerators (GPU, FPGA) can reach a computing power of 100 teraflops, like the NVIDIA GeForce RTX 4090 Ti card which works with 18,176 cores.

This is the reason why the fastest supercomputers in the world exploit both the processing power of CPUs and that of GPUs. Thus, supercomputers have reached petascale power (several million billion operations per second).

In 2024, HPE's Frontier in the United States has reached exascale power.
* 1 exaflops is equal to a billion billion operations per second.

Exponential growth cannot last very long!

1997 © − Astronomy, Astrophysics, Evolution and Ecology.
"The data available on this site may be used provided that the source is duly acknowledged."