Few areas of technology have ever appeared to advance as quickly as artificial intelligence. New capabilities are emerging almost every month, often making systems released only a short time ago appear surprisingly limited. This is why we continually review our online courses as new developments emerge, ensuring that they remain relevant in such a rapidly changing field.
With enormous investment taking place in the United States, China and elsewhere, the race is on to develop the most capable AI systems. However, none of this progress would be possible without the chips providing the immense computing power behind them.
To understand where AI may be heading, it is worth looking back at the history of the chip, what today’s AI processors can already do and what the next generation could mean for work in the office.
The First Commercial Microprocessor
In 1971, Intel introduced the 4004, widely recognised as the first commercially available microprocessor. It contained approximately 2,300 transistors, had been designed for a Japanese calculator and could complete around 92,000 instructions every second.
That was an extraordinary achievement. A single piece of silicon could perform calculations that previously required much larger electronic systems.
Today, Nvidia’s Blackwell AI processor contains 208 billion transistors and can complete quadrillions of specialist calculations every second. It supports artificial intelligence capable of reading documents, writing software, recognising images and communicating in human language.
In just over half a century, the chip has progressed from helping an office worker calculate a figure to potentially performing a substantial part of the office worker’s job, processing information at speeds no person could possibly match.
The Chip Enters the Office
The Intel 8088 processor used in the original IBM Personal Computer in 1981 contained approximately 29,000 transistors. It helped move computing away from specialist departments and onto ordinary office desks.
By 1993, the Intel Pentium contained 3.1 million transistors. Computers could run increasingly sophisticated spreadsheets, databases, word processors and multimedia applications.
These machines transformed office work, but they still depended on people providing detailed instructions. A spreadsheet could calculate thousands of figures, but it could not decide which figures mattered or explain what they meant.
The computer remained a tool operated by an employee. It did not understand the work.
Gaming Technology Helps Create Modern AI
The next major change came from graphics processors developed for computer games. Games were demanding increasingly realistic images, movement and three dimensional environments, requiring chips capable of performing large numbers of calculations simultaneously.
Games such as Wolfenstein 3D and Doom introduced millions of players to increasingly ambitious computer generated environments. Quake took the next major step in 1996 by creating a genuinely three dimensional world, helping to demonstrate the advantages of dedicated graphics acceleration and encouraging players to install increasingly powerful graphics cards. Games such as Crysis later pushed this competition even further, with players regularly upgrading their computers to handle more advanced visual effects.
This growing gaming market financed the development of processors that could complete thousands of similar calculations at the same time. Researchers eventually realised that the same ability was ideal for training artificial neural networks.
In 2012, AlexNet was trained using two Nvidia GTX 580 graphics cards originally designed for gaming. It learned to recognise objects from more than one million images and defeated competing image recognition systems by a substantial margin. Nvidia
The same technology eventually supported generative AI. Chips developed to create virtual worlds inside computer games became essential to artificial intelligence capable of interpreting the real one.
What Makes an AI Chip Different?
Modern AI processors are designed to complete the enormous number of mathematical operations required by neural networks.
Nvidia’s Blackwell architecture contains 208 billion transistors across two closely connected pieces of silicon. It provides around 10 petaflops of specialist AI performance. One petaflop represents one quadrillion mathematical operations every second. Nvidia
A chip cannot independently write a report or make a business decision. It needs an AI model, software, memory and access to relevant information. However, the chip provides the mathematical engine that makes these abilities possible.
The Intel 4004 helped a calculator produce an answer. Blackwell can support AI that examines information, identifies patterns and explains the result in ordinary language.
What Can AI Do in an Office Today?
Connected to the correct model and company information, modern AI systems can read contracts, examine accounts, summarise meetings and search thousands of documents.
They can extract information from forms, compare transactions, translate correspondence, draft reports and write computer code. AI agents can also open applications, retrieve information and complete a sequence of connected tasks.
Work that might take an employee several hours can sometimes be completed by AI in seconds. A person must still check its accuracy, but the amount of human time required has already changed.
The biggest restriction is no longer always processing speed. It may be whether the AI has access to reliable information and permission to act.
Five Times More Power Is Already Arriving
Nvidia’s newer Rubin processor contains 336 billion transistors and is designed to provide approximately 50 petaflops of AI inference performance. Blackwell provides approximately 10 petaflops for the same type of calculation. Nvidia
That represents around five times the theoretical performance in a remarkably short period.
Complete Rubin systems are also expected to offer up to ten times better inference performance for every watt of electricity. Nvidia says the cost of producing each unit of AI output could fall to around one tenth of its previous level. Nvidia
Not every AI application will suddenly become ten times faster. Performance also depends on memory, networking, software and the model being used. Nevertheless, advanced AI is becoming faster and less expensive to operate.
How Powerful Will AI Be in Five Years?
Future processors will combine several pieces of silicon, faster memory and improved communications within tightly connected packages.
TSMC, the world’s largest contract chip manufacturer, has outlined a route towards packages containing as many as one trillion transistors by 2030. AMD has set a separate target of improving the energy efficiency of complete AI computing racks twentyfold between 2024 and 2030. AMD
An advanced chip system in 2031 could therefore perform many times more useful AI work than today’s processors while reducing the cost of each completed task.
The most important improvement may not be how quickly AI writes one document. It could be the number of AI agents that a business can afford to operate continuously.
The Office Operated by AI Agents
Most businesses currently use AI as an assistant. An employee asks it to summarise a document, research a subject or draft an email.
By 2031, more powerful chips could support teams of specialist AI agents. One might monitor correspondence while another checks financial transactions. A third could prepare management information, while another searches for mistakes or regulatory problems.
These systems could update reports as new information arrives, complete work overnight and alert employees only when human judgement or approval is required.
The United Kingdom Government has published scenarios in which AI systems could automate many digital workflows as capably as people by 2030, although important limitations may remain. UK Government
This would not necessarily create an office without people. It could create an office in which fewer people supervise much larger quantities of automated work.
Faster AI Could Reduce Office Employment
AI does not need to become conscious or equal human intelligence before it affects employment. It only needs to become capable and inexpensive enough to perform commercially useful work.
If AI agents can complete research, administration and preliminary analysis for a small fraction of the current cost, businesses will reconsider how many people they recruit for those roles.
Management could also change. If AI monitors activity, prepares reports and allocates routine work, one manager may eventually supervise tasks that previously required several employees and additional administrative support.
The greatest employment effect may come from falling computing costs. Some work that AI can technically perform today remains too expensive or unreliable to automate fully. A fivefold or tenfold improvement could change that calculation.
From the Calculator to Super AI
In 1971, a chip helped someone perform a calculation. During the 1980s, chips placed computers on office desks. Today, they support artificial intelligence that can read, write, analyse and communicate.
Within five years, advanced chips may support affordable teams of AI agents capable of completing connected areas of office work with limited supervision.
Beyond this sits the possibility of Super AI, sometimes described as artificial superintelligence. This would be an AI system that exceeded human ability across almost every intellectual field rather than merely completing particular tasks faster than people.
There is no agreement over whether Super AI will be achieved, when it might arrive or whether present technology can ever produce it. Chips alone will not provide the answer. New models, improved memory, better reasoning and major scientific discoveries may also be required.
However, if AI chips continue improving at anything close to their current rate, the computing power available in 2031 will give researchers opportunities that simply do not exist today.
Whatever happens, this technology is moving in only one direction. AI is becoming faster, more capable and less expensive to operate. The consequences for work may be difficult to predict, but they will be impossible to ignore.
Brace yourself, because the next five years could make the previous fifty look remarkably slow. And if you think we are simply repeating the computerisation of the 1980s, think again. Those machines followed instructions entered by people. AI can increasingly interpret information, make decisions and complete work that once required human judgement. It is an entirely different animal.
Continue Learning About the Future of AI
AI Tuition Hub offers more than 110 courses covering artificial intelligence, its practical uses and its possible effects on business and society.
Many of our courses also examine the future of AI, including the development of AI agents, automation, robotics, leadership, education and the changing workplace. They are designed to help learners understand not only what artificial intelligence can do today, but what it may be capable of doing next.
Published: 15th August 2026.