Tech giants are trying to control everything along the artificial intelligence (AI) value chain.
For a long time, unlike competitors such as Microsoft and Google, Amazon has not extended its business "tentacles" too much into the development of artificial general intelligence (AGI). But Amazon is determined to change this "shortcoming". According to an internal memo leaked on June 29, local time, the company has "poached" five co-founders and other employees of AI startup Adept to join Amazon's AGI autonomous team.
This is just one part of the "all-rounded" map for tech giants. In fact, the biggest tech companies like Amazon, Google, Meta, and Microsoft are all developing their own AI chips, hoping to eventually sell them to businesses that use their cloud services. And Nvidia, the largest supplier in the field of AI chips, has launched its own cloud service business.
A research report released by UBS in June predicted that over time, the AI market will be monopolized by an oligopoly of vertically integrated "AI foundries." These "AI giants" will cover the entire AI value chain, from data center infrastructure and computing resources to generative AI algorithms to end-end applications.
Will AI be a game between big tech companies? Liu Jiren, founder and chairman of Neusoft Group, told the first financial reporter that he does not believe that the AI market will be completely monopolized by giants.
"One of the distinguishing features of the future development of AI is its continuous evolution, which will lead to continuous iteration. One company may be leading the way today, and another company may be rising rapidly tomorrow. "It's like the college entrance examination, every year the best students come out." Every company has a process of continuously improving its AI capabilities. ”
A well-rounded AI manufacturer
Currently, the strategy of big tech companies has shown a trend towards vertical integration, not only dabbling in cloud services and AI algorithm development, but also designing custom AI chips and wanting to have AI-powered applications.
The cloud service platform provides a powerful infrastructure for AI model training and deployment. In this area, the tech giants have gained a solid advantage. Amazon Web Services (AWS), Microsoft Cloud Computing (Azure) and Google Cloud (GoogleCloud) dominate the global cloud services market, accounting for 31%, 25% and 10% respectively, according to a report by market analyst firm Canalys in the first quarter of this year.
In addition, tech giants are developing and controlling the hardware that underpins AI. In November last year, Amazon unveiled its own second-generation Trainium chip. In the same month, Microsoft unveiled its self-developed AI image processing unit (GPU), named Maia100, at its annual developer conference. In May, Meta announced plans to develop an AI chip tailored to its needs. Google also officially unveiled its new sixth-generation tensor processing unit (TPU) chip at this year's developer conference.
Naveen Rao, vice president of generative AI at Databricks, a U.S.-based data governance vendor and former vice president of Intel's AI product group, said: "Theoretically, if [tech giants] can achieve high enough production and reduce costs, these companies should be able to offer better products than Nvidia." ”
Based on their strong layout capabilities at the infrastructure layer, "big technology companies" are also "blossoming" in the development of large language models (LLMs) and AI applications. Google, for example, has used its TPU to develop large language models and its own AI products, including its online chatbot GoogleBard, now Gemini.
AI startup Cohere and others are building similar technologies with Google's cloud services. At the end of last year, Amazon invested $4 billion in AI startup Anthropic to strengthen its competitiveness in the AI field. Microsoft, for its part, has enhanced its AI capabilities through a partnership with OpenAI and the integration of AI capabilities into its Azure cloud services. At present, Microsoft's AI solutions are widely used in enterprise customers.
Vertically integrated "AI foundry" oligopoly
The UBS report argues that the economies of scale of cloud computing and generative AI suggest that companies that dominate these businesses will also dominate the development of AI applications. This vertical integration could lead to a small number of large players controlling the majority of the AI market, creating a fully vertically integrated "AI foundry" oligopoly.
The report explains that the economies of scale of cloud computing and generative AI are enormous. Large-scale operations bring economies of scale, making it cheaper and more efficient to develop and deploy AI technologies. This further strengthens the position of large enterprises, as it is difficult for small companies to compete with them in terms of cost and capacity.
For example, the cost of building and maintaining a state-of-the-art data center is high. Vertically integrated companies can spread these costs across a wider range of products and services, reducing the overall cost per unit of AI capability.
Therefore, based on the interconnectedness of these technologies, it is likely that companies that dominate one layer of the AI value chain will extend their influence to other layers. For example, large cloud computing providers such as Amazon Web Services (AWS) have large computing resources that are critical for training and running large AI models. These resources can be used to develop advanced AI applications that create a competitive advantage that smaller businesses can't match.
And large companies are more emboldened to invest in innovation. Significant investment in AI infrastructure and research has driven the rapid development of AI technology, according to the study. Vertically integrated companies tend to be the leaders of these investments, positioning themselves at the forefront of AI innovation. For example, at the end of March this year, it was reported that Microsoft and OpenAI planned to spend more than $100 billion to invest in a data center project that would include an AI supercomputer called Stargate and be equipped with millions of dedicated server chips to power OpenAI's AGI.
Ultimately, vertically integrated companies will have a significant competitive advantage, enabling better performance, lower costs, and faster innovation cycles. For example, they can optimize the entire AI stack, from hardware to application, ensuring that all components work together seamlessly. Companies that control more of the AI value chain can also better protect their intellectual property and become less dependent on external suppliers.
However, Liu Jiren told the first financial reporter that the generation and creation process of AI capabilities is not transparent enough, and people have not fully understood the principles behind it. "The development of AI is very complex," he said. Especially when computing power becomes cheaper and there are more and more algorithms, there will be a situation where a hundred flowers bloom. Just like we have so many apps today, it's because of a large number of users who use smart devices, and the communication cost is cheap. The same will be true for AI in the future. I believe that in the future, AI will become a platform with a large number of algorithms and basic computing power, and users will have more choices. ”
Antitrust regulators have not been idle
Direct takeovers of small companies in related fields are a common means for tech giants to make up for their "shortcomings" as soon as possible, but this is often also prone to triggering antitrust scrutiny by regulators.
In the past year, tech giants have increasingly achieved control over AI products, technologies or key personnel through means such as injecting capital into AI startups, establishing strategic partnerships, and "acqui-hires", while also avoiding being restricted by antitrust regulations due to excessive control.
For example, Amazon did not buy Adept outright, but "generously" accepted Adept's core executives and most of its employees. In March, the founder and key team members of the startup InflectionAI were also "poached" by Microsoft to join the latter's newly formed division, MicrosoftAI, and the two companies also signed a technology licensing agreement.
But even so, the tech giants' monopoly propensity for the AI value chain has caught the attention of regulators. Microsoft's poaching of InflectionAI, for example, has not escaped surveillance by the Federal Trade Commission (FTC), which is investigating whether Microsoft is avoiding antitrust scrutiny in this way.
Since 2019, Microsoft has invested more than $13 billion in OpenAI in exchange for some of its profits, technology, and the latter's use of the former's cloud services. Since the end of last year, the EU's antitrust watchdog, the UK's Competition and Markets Authority (CMA) and the FTC have successively revealed that they are reviewing the relationship to clarify whether it is a de facto "merger" and whether it violates antitrust laws.
In April, the CMA also announced in a statement that it would open "invitations for comment" on Microsoft's relationship with French company MistralAI, Microsoft and InflectionAI, and Amazon and Anthropic to assess their compliance with UK M&A rules and explore the competitive impact of these deals.
In a survey report, the CMA said that the partnerships of the major players in the AI base model market could exacerbate monopolies through their value chains. In the report, the CMA lists an "interconnected network" of more than 90 partnerships and strategic investments, including Google, Apple, Microsoft, Meta, Amazon, and Nvidia, among others.
In May, the U.S. Department of Justice (DOJ) also announced increased focus on competition in the AI space. The department disclosed that it has launched a number of investigations into whether AI companies use common officers or directors.
However, it is often difficult for regulators in Europe and the United States to prove this within the framework of traditional mergers and acquisitions. Last month, the CMA said it had decided not to conduct an investigation into Microsoft's partnership with MistralAI under the merger provisions of the UK's Companies Act 2002.
(This article is from Yicai)