At the beginning of the widespread use of mobile phones in the 1990s, Nokia achieved an undisputed leading position as a mobile phone manufacturer. In the early years of this century, it had a global market share of 40%. Nokia was the most recognised brand, with the widest range and best distribution in the world. Its position seemed unbeatable. In 2007, Apple launched the iPhone, a smart, touchscreen mobile phone with the ability to use countless applications (apps). A year later, Google’s Android appeared with a multitude of apps, which was adopted by all Asian manufacturers. As a result, Nokia, which disregarded Apple’s innovation, was left behind and ended up almost disappearing as a mobile phone manufacturer.
Today, Nvidia is the leading supplier of the chips needed for artificial intelligence (AI). In the last five years, its sales have multiplied by 20 and its valuation by 14. It operates with profit margins of 75%. Its stock market valuation has exceeded £4 billion (with twelve zeros). Over the next two years, the United States has committed investments in new data centres (computer macro farms) worth more than £500 billion. Approximately 60% of the investment in a data centre corresponds to the necessary semiconductors. Data centre availability rates in the United States are at minimum levels, at only 1.6%. In addition, three-quarters of the data centres currently under construction already have guaranteed leases. As a result, demand for Nvidia semiconductors is guaranteed for the next few years.
In recent days, Nvidia’s share price has been penalised by the emerging threat from Google’s TPU (Tensor Processing Units) chips. The announcement of a contract between Meta and Google to purchase billions of Google chips for its data centres from 2027 onwards, considering them cheaper and more efficient than Nvidia chips, indicates that something may be changing in the market for advanced AI chips, which has so far been dominated by Nvidia.
Google created the TPU to solve an internal problem back in 2013. According to its own calculations, the demand for AI computing, especially for voice searches, would multiply. This would require prohibitive investments in new data centres. At that time, Google relied on standard chips (GPUs). Traditional GPU (Graphics Processing Unit) chips were considered too slow and expensive. Therefore, they decided to design their own systems (ASIC – Application Specific Integrated Circuit) and their own chips specifically for AI.
Without going into technical details, TPUs consume much less energy than GPUs because data flows between units without having to return to memory. The GPU chip moves data back and forth between memory and the computing units for the calculation performed. Google created TPUs to solve an internal problem, not to market them to third parties.
Although TPUs are less versatile than GPU chips, they consume much less energy and perform better for specific tasks, the use of TPUs significantly reduces the cost of research for each AI query compared to GPUs. Although Google’s TPUs are efficient and outperform Nvidia’s GPUs for certain tasks, their wider adoption by the market has a serious problem. Google has developed its own ecosystem for internal use, without considering external sales. As a result, most companies use two or three of the major cloud providers: AWS, Microsoft Azure, and Google Cloud. These three clouds use Nvidia GPUs. To use only Google TPUs, it is necessary to use exclusively the Google cloud.
By way of comparison, to use an Apple computer programme, you must do so on an Apple computer; it cannot be used on a computer running Microsoft Windows. Meta has just announced that it will lease part of Google’s cloud with TPU chips and purchase billions of these chips from 2027 onwards for its own data centres.
This announcement does not seem to be equivalent to the announcement of the first iPhone in 2007. In any case, it shows that technological development can unseat a temporary leader if they do not react in time with new innovations. Sources at Google estimate that they could take up to 10% of Nvidia’s revenue. Although Nvidia is by no means at risk, its current margins of 75% seem hardly sustainable if competition increases.
Nokia helped launch and popularise mobile telephony, although it later lost its leadership and virtually disappeared as a mobile phone manufacturer. However, the success and evolution of mobile telephony itself thanks to smartphones is undeniable. Something similar is happening with AI. Not all companies will be winners. Nvidia may be one of them, but technological developments and competition, such as that represented by Google’s TPUs, make it difficult for it to maintain margins like the current ones.
