Oil companies compete for supercomputing capacity

Oil companies own 20 of the 500 most powerful supercomputers in the world, according to TOP500a project created by academics and research centers that ranks the 500 machines with the best performance in the Linpack test. The survey considers the June 2026 edition of the TOP500. Companies may operate other equipment that has not been subjected to the test prepared by the organization and, therefore, does not appear on the list.

Just the Petrobras has 7 supercomputers listed. Technology is considered strategic. Supercomputers are mainly used to locate and manage oil reservoirs.

Despite the high number of HPCs (English acronym for High Performance Computer), the sum of the total capacity of the machines in Petrobras’ custody is lower than that of 2 of its competitors. Eni has 3 supercomputers on the list, but it has processing power far above that of other oil companies: 1,085 petaflops. The company also owns the HPC7 and the HPC6 which are among the 10 most capable in the world, in 6th and 8th place in the ranking, respectively.

A ExxonMobil appears next, with two machines that add up to 190.3 petaflops. Petrobras ranks 3rd among companies in the sector, with a combined capacity of 121.6 petaflops.

In addition to Eni, ExxonMobil and Petrobras, SaudiAramco —most profitable oil producer in the world—, the French TotalEnergiesthe Thai PTTEP and the British BP —whose HPC is in the United States— have supercomputers listed among the 500 most powerful on the planet.

Together, the equipment of the 7 oil companies identified in the ranking totals around 1,493 petaflops of maximum processing capacity. The amount is equivalent to the processing capacity of around 7.5 million conventional notebooks.

ESSENTIAL STRUCTURE

Petrobras’ seismic processing specialist, Daniel Thomé de Paula, explains that HPCs are an integral part of the oil exploration infrastructure because drilling a well requires high investments. Before starting operations, companies need to reduce uncertainty about what lies beneath the surface and assess whether there is an exploratory opportunity that justifies the cost of drilling.

At sea, specialized boats emit sound waves that pass through the water and penetrate underground. The signals are reflected by different layers of rocks and captured by sensors. The material collected, however, is made up of billions of records that, in their raw form, cannot be directly interpreted by geologists and geophysicists.

Supercomputers apply mathematical models to this data to reconstruct the path taken by the waves and transform them into images of the subsurface. With this, technicians can identify geological structures, evaluate potential reserves and decide whether to recommend drilling a well.

Daniel compares the process to a dolphin. He explains that the animal hunts using echolocation. The dolphin makes a “clique“, he says, which reflects on the fish and returns to the mammalian position. Upon hearing the click, the animal processes “in your head” and understands that that click is related to a fish at a certain distance, going in a specific direction.

“This transformation of the click into a mental image of the dolphin is what the supercomputer does: it transforms this seismic signal into an image”, says the expert.

After the discovery, the technology continues to be used to monitor the behavior of reservoirs. As oil is extracted, fluids move, water accumulates and internal conditions in the field change. Simulations help decide where to drill new wells, where to inject water and when to change the development plan.

This processing is only possible because supercomputers bring together several high-capacity servers, which in turn are equipped with cutting-edge processors, large volumes of memory, graphics cards and high-speed communication networks.

According to Petrobras HPC consultant Guilherme Silva Vilela, the machine divides mathematical problems into smaller parts and then distributes them among the computers that make up the system. Each piece of equipment processes a portion of the calculation and communicates with the others so that, in the end, the results are gathered.

“A supercomputer is a combination of several computers, all of them powerful, that communicate so that each one can solve a part of the problem and, in the end, achieve the expected result in a timely manner.”, he states.

Daniel Thomé de Paula (top), a specialist in seismic processing, and Guilherme Silva Vilela, a high-performance computing consultant at Petrobras, work on the development and operation of supercomputers used by the state-owned company

According to experts, Petrobras has invested in this type of infrastructure practically since the beginning of its activities. The state-owned company’s first digital processing center was created in the 1960s. Before that, the company already maintained an analog processing structure focused on exploration.

The company also carried out the first tests with GPUs for seismic processing in 2006 and 2007 and put its first production machine with graphics cards into operation in 2008. According to Guilherme, the change multiplied processing capacity by around 10, at the same time as reducing the amount of equipment needed and energy consumption.

Today, Petrobras focuses a significant part of its resources on designing supercomputers capable of handling gigantic volumes of seismic data. In addition to the 7 machines listed by TOP500, the company also has other smaller HPCs.

Daniel also states that the ranking is not the appropriate way to compare supercomputers aimed at the purposes proposed by Petrobras.

The Linpack test, used by TOP500, measures the speed in solving complex systems of linear equations, with double precision calculations that favor constant communication between all the machine’s servers. Seismic processing, however, requires above all a high capacity for reading and writing files, in addition to generally using simple precision calculations and more concentrated communication between nearby servers.

“It’s as if you designed a 4×4 car and put it on a Formula 1 track to see what its top speed is. It’s not your land, it’s not what it was intended for“, it says.

INVESTMENTS IN THE SECTOR

Processing is expensive and requires periodic renewal of the infrastructure. Harpia, Petrobras’ largest supercomputer, went into operation in 2025 to single-handedly replace Fênix, Atlas and Dragão. With a performance of 75.2 petaflops in the Linpack test, the machine occupies 37th position in the June 2026 edition of TOP500.

According to Guilherme, the company’s supercomputers have a planned life cycle of around 6 years. Before each replacement, Petrobras evaluates the expected processing demand for the entire period, considering the algorithms it intends to execute, the new exploratory areas and the need to expand or just maintain existing capacity.

The size of the investment also depends on economic factors. Daniel states that the price of a barrel of oil influences how much spending on exploration equipment is justified. The greater the expected return from new discoveries, the greater the economic space to expand computational capacity tends to be.

The decision also considers the price and availability of components. According to experts, instability in the hardware market and the scarcity of parts, such as memories, can make systems more expensive or limit the size of machines. Therefore, even with constant demand for more processing, the company seeks to size the equipment to meet needs without generating financial waste.