Microsoft and Google Want to Dominate the AI Economy. Open-Source Models Are a Threat.

Meta Platforms is launching an open-source artificial-intelligence model in partnership with Microsoft. It's a sign of increasing competition in the AI sector which could limit the share of revenue won by the giant technology companies.

Companies such as Meta Platforms (ticker: META) and International Business Machines (IBM) are offering access to AI models allowing the companies to create their own software. They hope they can win over customers from Microsoft (MSFT) and Google-parent Alphabet (GOOGL), which disclose lesser details about how their own AI models operate.

Meta's new AI language model, named Llama 2, is being released as free open-source software, which means it can be used, modified, and shared. It will be available to companies for the first time via Microsoft's cloud-computing platform and Amazon.com's (AMZN) Web Services and other providers.

Similarly, IBM recently launched Watsonx - a platform allowing its clients to access a range of AI models trained for different tasks in partnership with open-source AI start-up Hugging Face.

"We believe this will result in models that will require less data ingestion with a platform that is easier and potentially more cost effective to manage for high-quality Generative AI results," analysts at J.P. Morgan wrote in a research note on IBM.

There's no guarantee that the IBM or Meta models will gain widespread traction with commercial customers, and investors don't seem too fussed for now. Microsoft stock rose to new heights this week as shareholders concentrated on stronger-than-expected pricing for its AI-supported software. Meta shares barely moved on the AI partnership announcement.

However, there's reason to believe Microsoft and Google's current dominance won't last. A leaked memo from a Google researcher, published on industry blog SemiAnalysis, stated the company had "no moat" or durable competitive advantage in AI and said the same was true of OpenAI, citing the threat from faster and more customizable open-source models.

One potential vulnerability for Microsoft is that it depends on an independent company , ChatGPT developer OpenAI for an increasingly important part of its business. That might not be tenable in the long term.

"I think it's likely that they're going to try to buy OpenAI even at a very high price, it would still make sense for a company like Microsoft," Pedro Palandrani, director of research at Global X, a provider of thematic exchange-traded funds.

OpenAI was last valued at around $29 billion. News website Semafor has reported that Microsoft will get three quarters of OpenAI's profits until it recoups its initial investment and then assume a 49% stake in the company.

Microsoft didn't respond to a request for comment from Barron's on the terms of its arrangement with OpenAI or its future policy toward other open-source AI models.

One limiting factor for those hoping to challenge Microsoft and Google is cost. Microsoft's Chief Technology Officer Kevin Scott warned at a recent developers' conference that building an AI rival from the ground up was "economically not viable."

The problems can be glimpsed in academia, previously the center of most AI development. In 2022, there were 32 significant industry-produced machine-learning models compared with just three produced by academia, according to Stanford University's AI Index report. That was put down to the gap in resources between academia and industry.

"Professors have refused literally eight-figure salary jobs to keep being professors," said Adnan Masood, chief architect of AI at technology-services company UST Global and a visiting scholar at Stanford.

However, companies could choose to buy rather than build. Software company Databricks struck a deal to acquire generative AI start-up MosaicML in June. Databricks CEO Ali Ghodsi said the goal of the deal valued at roughly $1.3 billion--was to "democratize AI."

Some also have the money to simply poach talent. Tesla (TSLA) CEO Elon Musk recently launched his ex-AI business and has recruited Igor Babuschkin, formerly a scientist at Alphabet's DeepMind, to helm the effort. Musk has previously criticized OpenAI as having moved to a closed-source model under Microsoft's influence.

The future of AI models is beginning to look more like a pick-and-choose buffet than a set menu dominated by Microsoft and Google. That doesn't mean the tech giants will necessarily lose their leading role. They have key advantages in terms of money, cloud-computing power, and talent, which could maintain their lead in commercial products. But it does mean they face more competition for the wave of AI spending coming over the next few years.