Will a chip good for one task truly have the mettle to take down the reigning champ of AI chips? Alphabet is certainly bent on proving the affirmative, and it’s leading Tensor Processing Units (TPUs) that are very much at the forefront of this battle of market capitals. Graphics processing unit vendor Nvidia has been the go-to provider for machine AI development and deployment for several years, and its parallel computing capabilities have solidified its marketplace supremacy. However, Alphabet’s vertical integration strategy is slowly threatening to disrupt this marketplace status quo.
TPUs are application-specific integrated circuits designed and built by Google’s parent company, Alphabet, for machine learning applications. Unlike GPUs, which are general parallel processors, TPUs are hardwired for those specific mathematical operations that machine learning applications perform or solve. The seventh generation TPU was unveiled in November, and it’s already in scale deployment, with company Anthropic announcing plans to train their Claude models on anywhere up to one million TPUs. Industry insiders think that TPUs are on an equal footing, or even better, than Nvidia’s latest GPUs for some machine learning applications.
The hardware edge is further strengthened by Alphabet’s dominance over the entire AI software and hardware stack. Right from designing chips to data center implementation, as well as clouds, Google has been able to create a fully integrated ecosystem that requires less dependence on costly hardware provided by companies like Nvidia. The TPUs are deeply integrated with Google Cloud’s AI platform, Vertex AI, making it simple to implement and develop models. For business customers, it enables a fast development cycle and possibly lower costs, which are winning big contracts for Google. Meta’s six-year, $10 billion Google Cloud deal, as well as Apple’s annual commitment to spend $1 billion to use a custom Gemini model to power Siri, are a testament to that.
The distribution moat of Alphabet is just as strong. Its Gemini large language model is integrated into products used by an enormous number of people: Google Search, YouTube, and so on, which means it gains immediate traction that the likes of OpenAI have to work hard to match. The number of monthly active users of Gemini has climbed to a whopping 650 million, reducing the gap with the estimated 845 million monthly active users of ChatGPT. According to Sensor Tower, Gemini’s user count has risen by 30% in three months, beating the 6% rise of ChatGPT, thanks to its recent release of a Nano Banana image generation model.
These technological developments are fueling several revenue engines. In the last reported quarter, the revenue from Google Search rose 14.5% year-over-year, YouTube Ads increased 15%, Google Cloud revenue skyrocketed 34%, and subscriptions, which are Gemini’s revenue sources, increased more than 20%. It is a significant development that the operating income in Q3 is up a whopping 85% to $3.6 billion, while the operating obligations are up 46% to $155 billion. The operating margins in this business are now at 23.7%.
Conditions in the economics of AI infrastructure are also tilting in favor of Alphabet. Leasing the usage of TPUs in Google Cloud enables enterprises to have an alternative solution apart from using Nvidia’s GPUs, which may even go up to $40,000 per unit and are faced with the possibility of shortage. TPUs, which are specifically important within inference-intensive applications, have the potential to provide superior cost-performance ratio compared to GPUs. With AI applications being adopted beyond the boundaries of the technological industry by others such as the finance, medical, and manufacturing sectors, the market potential for such chips is estimated to reach hundreds of billions of dollars. Nvidia’s strength in the CUDA software ecosystem and its existing developer base will continue to sustain it. However, the company’s year-over-year revenue growth of 62% in the previous quarter may face challenges due to the increase in cloud chip initiatives by cloud majors such as Alphabet, Amazon, and Microsoft. Alphabet’s challenge to the company has already been visible in the form of prominent successes in cloud AI infrastructure and partnerships that cut into Nvidia’s richest revenue sources,
The company’s story begins with Jensen Huang, the founder, who has remained focused on the industry’s most essential and rapidly changing area in With an operating income of $127 billion in the past year, Alphabet is quickly approaching Apple’s $133 billion figure and Nvidia’s $110 billion. However, if the current trends are to be believed that is, user growth at Gemini, profitability at Google Cloud, and the use of TPUs at Google Alphabet will be able to surpass both Apple and Nvidia in its earnings. For a technology investor, the future implications appear to be that it is not the company that is shipping the most semiconductors but the one that controls the most comprehensive AI ecosystem.
