When you fire up an artificial intelligence chatbot like Google’s Bard or OpenAI’s ChatGPT, you’re really interacting with the product of three or four key ingredients.
One is the engineering talent it took to design the chatbot’s AI model. Another is the vast amount of training data the model chewed through to learn to respond to your prompts. The third are the advanced semiconductor chips used to perform the training, a procedure that can take months even with the fastest chips.
Now, cloud platforms are fast becoming the fourth essential ingredient in AI. They aggregate the computing capacity of many of those in-demand semiconductor chips, provide online storage and other services, renting it all out to AI companies that need raw processing power and a place to keep their training data.
AI developers’ reliance on cloud services shapes the ebb and flow of the broader AI industry, putting cloud companies at the heart of a technology that’s poised to transform the way people work, play and learn.
Just a handful of major players dominate the half-trillion-dollar cloud market, including Amazon, Microsoft and Google. Now, policymakers and industry critics warn, Big Tech’s power in cloud markets may give it enormous and possibly anticompetitive influence over the future of AI.
“I am deeply concerned that a handful of Big Tech firms dominate cloud computing and storage,” Massachusetts Democratic Sen. Elizabeth Warren told CNN. “Without commonsense regulation, those companies will entrench their dominance over AI, crush competitors, and put consumer privacy and safety, innovation, and national security at risk. We must protect competition in this critical industry.”
All three companies declined to comment for this article. Industry defenders argue that the sector’s leading companies compete vigorously for business and that their scale helps give AI companies a comprehensive solution for all their cloud computing needs.
As with any essential industry controlled by large players, though, policymakers’ fears focus on the potential for inflated pricing, anticompetitive collusion, exploitative contract terms or other practices that may dictate what it ultimately costs you to use AI services or the types of products that make it to market.
The cloud market is limited to only a few major players
AI’s share of total cloud spend is expected to grow substantially as generative AI picks up steam.
Even as the public cloud market expands by leaps and bounds overall — the total spending may jump more than 20% to $679 billion next year, according to the market research firm Gartner — artificial intelligence could make up between 30% and 50% of that pie in as little as five years, Gartner and other industry analysts say.
Only a few cloud platforms can deliver the enormous amounts of processing that AI companies and their customers increasingly need, said Matthew Prince, CEO of the internet monitoring and security company Cloudflare.
As AI developers and other businesses looking to use AI become more dependent on cloud providers, Prince said in an interview, “I think that they certainly are going to have a bigger role in picking the winners and losers” of the AI industry.
Today, artificial intelligence accounts for just a small portion, perhaps less than 10%, of the $563 billion public cloud market, according to the market research firm Gartner.
The top three providers of AI in the cloud are Amazon, Microsoft and Google by a wide margin, followed by smaller cloud providers including IBM and Oracle.
Regulatory scrutiny at home and abroad is growing
Governments around the world are taking notice of what is happening in an increasingly high-profile industry.
In the United States, both the Federal Trade Commission and President Joe Biden have highlighted concerns about competition in cloud markets as a potential problem for AI development.
In June, the FTC wrote that if one or more companies gains a stranglehold on the cloud industry or any other essential input into AI development, they “could wield outsized influence over a significant swath of economic activity” as “AI is increasingly becoming a basic part of daily life.”
Biden also alluded to those concerns when he signed an executive order in October addressing artificial intelligence.
Promoting a fair and open market for AI development, the order said, “requires stopping unlawful collusion and addressing risks from dominant firms’ use of key assets such as semiconductors, computing power, cloud storage, and data to disadvantage competitors.”
The order did not name specific providers or allege any misconduct by the industry.
The FTC and other global competition regulators have scrutinized the cloud industry in recent years with multiple wide-ranging studies into the sector.
An investigation launched by the UK government this fall, for example, zeroed in on Microsoft and its software licensing practices — which are largely unrelated to AI — while the FTC in March said it was looking into the industry alongside France, Japan, the Netherlands and South Korea.
Most of the government inquiries to date “converge on the same conclusion, which is that this is a tremendously concentrated market,” said Sarah Myers West, managing director at the AI Now Institute and a former FTC senior advisor on AI.
Part of the reason cloud platforms partner with AI companies is that when their products end up in the hands of consumers or businesses, that generates a constant stream of demand for even more cloud resources, West added.
It can also set up large tech companies to collect even more data from businesses and internet users.
“They want to walk people into their cloud ecosystems, which is the straightest path to making a profit off of these tools,” she said.
What cloud critics say they worry about
There are a handful of specific cloud practices that raise potential competition concerns, according to some industry analysts.
One are the often-exclusive agreements that AI companies sign with just one or two cloud providers in exchange for substantial investments.
Exclusive deals for cloud computing are the norm for AI developers, said West, unlike other parts of the economy where using multiple cloud providers may be more common.
“A lot of the AI startups that are creating tools that are on the market today have already entered into exclusive licensing agreements with a cloud provider,” she said. “Even some of the smaller players all have relationships with a cloud infrastructure company and generally only run on that one [provider].”
Take OpenAI, for example. The startup’s ChatGPT tool that wowed consumers and kickstarted the current AI frenzy a year ago depends heavily on a partnership with Microsoft, which has committed $13 billion to the relationship in cash and free cloud computing usage.
“There is no OpenAI without Microsoft leaning in, in a deep way, to partner with this company on their mission,” Microsoft CEO Satya Nadella told CNN contributor Kara Swisher in November, adding that if forced to by the recent governance crisis at OpenAI, Microsoft could do all of the startup’s AI work itself. “We have the people, we have the compute, we have the data, we have everything.”
Earlier this year, Amazon announced it was investing up to $4 billion in the AI startup Anthropic in exchange for a minority stake and “primary” cloud provider status. Just months earlier, Anthropic had announced a deal to use Google as its “preferred” cloud provider.
Some experts have warned that free cloud credits could allow powerful cloud companies to “lock in” AI companies as customers, and discourage competition by making it hard for them to leave or mix and match features from different providers.
“The further we go down that road, the harder it is to unbundle,” said Steven Weber, a professor at the University of California’s Berkeley School of Information, at an FTC roundtable this year.
And some deals between cloud and AI companies, as with Amazon and Anthropic, can also present opportunities for large tech companies to gain ownership shares in influential AI startups. The UK’s competition regulator is considering a possible antitrust probe into Microsoft’s relationship with OpenAI along these lines, suggesting that OpenAI’s recent leadership crisis may have led to a “relevant merger” that warrants an investigation. Microsoft has denied that it owns “any portion” of OpenAI, though it did recently gain a non-voting seat on the startup’s board.
Another practice that’s come under scrutiny are the fees some providers charge for withdrawing data from a cloud provider.
“It just makes AI more expensive,” Prince said. If AI companies could cheaply and easily move their data from one provider to another, “that would drive the cost of [AI] training down markedly …. my hunch is that you could cut the cost of training probably in half.”
How cloud companies defend their record
Cloud providers and their defenders argue that the cloud market is highly competitive, even if it may be concentrated due to how expensive it is to build cloud services at enormous scale.
“Since AWS began, it has significantly cut its prices for services such as computer processing, data storage, and data transfer,” Amazon told the FTC this summer as part of the agency’s industry study. “Competition among cloud and other IT providers is thriving, to the benefit of the U.S. economy.”
In a sign of how Google is trying to compete for AI cloud customers, the company this month announced a new generation of powerful cloud-based computing processors that Google said can train large AI models nearly three times faster than the last generation.
Google has said it used those chips to train its latest and most sophisticated AI model, Gemini.
“At Google Cloud, we are dedicated to being the most open hyperscale cloud provider, and that includes our AI ecosystem,” Google said in a blog post this year. “Through partnerships, we can help organizations more easily access and innovate with generative AI and large language models, apply fast-evolving AI and ML capabilities to address real industry use cases, and build a new wave of applications utilizing all of these capabilities.”
Cloud customers, including AI developers, are not naive when they sign agreements with a cloud provider, Microsoft has said.
“Customers negotiate extensively with service providers, including on price, storage capacity, contract length, and more,” Microsoft told the FTC. “This means that for cloud vendors, every project is also a pitch — if a customer is not getting the level of quality or value that it expects from a cloud vendor, that customer is going to turn to other alternatives. Cloud vendors have to compete accordingly.”
Exclusive agreements between AI companies and cloud providers can be a good thing, said Brandon Jung, vice president of ecosystem and business development for the AI startup Tabnine and one of Google’s earliest cloud platform employees.
“The plus is efficiency for developers and for deploying,” Jung said, adding that it could also enhance security to use a single vendor. “But I think it’s going to tie people more into the clouds.”
Critics say some of the concerns policymakers have raised are a misplaced byproduct of the suspicion governments have had now for years toward app stores, social media and e-commerce giants.
Many of those happen to be the same companies that run massive cloud computing businesses, said a person familiar with the thinking of major cloud vendors who requested anonymity because they were not authorized to speak publicly.
“There is a general discomfort among regulators around the world with Big Tech and with digital markets, and a sense that the level of concentration in those markets is not good,” the person said. “They’ve taken on, unfortunately, some of those same fears and concerns and transferred them into their view of cloud computing, which has very different customers.”
Some AI companies have deliberately steered clear of any exclusive entanglements with cloud vendors.
Cohere AI, which provides AI models to business customers rather than consumers, said it works with all major cloud providers even though it has probably meant missing out on some funding opportunities.
“In our last [funding] raise, we were very deliberate about not taking any one check bigger than the others,” said Martin Kon, Cohere’s COO. “We’re not beholden to anyone that says, you know, ‘I’m the lead person you have to listen to.’”
But companies like Cohere are more the exception that proves the rule, said West.
“I think it makes clear that the cloud firms hold a tremendous amount of power in the market,” she said.
