Two years after OpenAI introduced ChatGPT, the frenzy over artificial intelligence has morphed into a fierce competition between tech industry giants and upstarts.
AI stocks led by Nvidia (NVDA), Microsoft (MSFT) and Google-parent Alphabet (GOOGL) now occupy center stage on Wall Street. At the same time, investors are keeping an eye on startups like OpenAI and Anthropic, which blazed the trail in the fast-growing trend.
But as the AI showdown escalates, the Wall Street spotlight has turned to tough questions: Are the massive costs of building AI systems worth it? When will these investments start to pay off? Worries are growing about the limits of what’s been billed as a game-changing, revolutionary technology. And a new view is emerging when it comes to AI models: Big may not necessarily mean better.
↑
X
AI’s Next Frontier: The Tech Set To Dominate CES 2025
AI Stocks Push Technology Forward
“There’s a class of companies that really wants to push the technology as far as they can,” data scientist Ben Lorica said in an interview. “They all believe that bigger is better as part of a law of scaling. What does that mean? The cost of training these models is going to grow rapidly.”
Further, the ability to raise capital is critical for AI players.
What’s clear is that OpenAI, Anthropic and other startups are scrambling for cash infusions from well-capitalized tech industry giants in order to continue operating. Microsoft has invested billions in OpenAI, becoming its biggest investor. Amazon.com (AMZN) last month put $4 billion more in Anthropic, bringing its investment to $8 billion.
But Google and Facebook-parent Meta Platforms (META) aren’t yet sugar daddies to AI startups, though Google has made some investments. Meta and Google loom as formidable rivals. On Dec. 11, Google rolled out its Gemini 2.0 models.
The AI trend has relied mainly on large language models, or LLMs, which require massive amounts of data to be trained. Large language models allow users to interact with AI systems without the need to write algorithms.
Simply put, two battlegrounds have formed. In the consumer market, OpenAI has shaken up internet search. Going forward, the biggest AI models will play a big role in internet search. Google’s Gemini family of AI models still lags OpenAI’s performance but that may change.
Whether OpenAI dethrones Google as the leader in internet search has huge implications for many other tech giants, such as Microsoft, Meta and Apple (AAPL).
Enterprise Trend: Rise Of Small Language Models
There’s also the enterprise market, where big companies and organizations have been trialing smaller AI models for business purposes.
Small language models, or SLMs, train AI systems on specific types of data and are focused on more customized outcomes. They have emerged as a less expensive alternative to LLMs.
The rise of open-source models along with improved data science is making small language models more economical, said Databricks chief AI scientist Jonathan Frankle in an interview. “I wouldn’t underestimate the amount of ingenuity that’s going into making these models more efficient. For the same amount of money that OpenAI spent to train GPT-4 a couple of years ago, you could get a much better model today, just based on modern ways of training models that were not around a couple of years ago.
“The science is moving very fast. We’re learning how to use data better and develop better ways to train these models. In some cases, building big, general-purpose systems might be the right way to go. But often the best way is building very specific, custom AI systems that are designed for specific use cases based on a customer’s proprietary data.”
Meta Stock And Open-Source AI
The capabilities of open-source models, which are free to developers, have advanced rapidly, with Meta emerging as the leading proponent.
According to research institute Epoch AI, the performance of open AI models lags that of proprietary models such as OpenAI’s GPT-4 and Google’s Gemini by only a year.
On Meta’s Q3 earnings call with analysts, CEO Mark Zuckerberg said that Llama “usage has grown exponentially in 2024.” Llama is Meta’s family of open-source large language models.
Meta’s Llama 4 models are in development and should launch sometime in 2025. Meta stock has climbed 76% in 2025 on enthusiasm over Meta’s AI strategy.
Meanwhile, AI chipmaker Nvidia in September released its own powerful Nemotron open-source model. Nvidia stock has soared 172% in 2024.
The open-source approach is playing a key role in an important issue: pricing.
Prices have been falling rapidly for software developers accessing AI models via cloud computing services. AI model makers typically split revenue with cloud companies such as Amazon Web Services.
“You could say the OpenAI models are better. But how much better?” said Lorica, who edits the Gradient Flow newsletter, which follows AI trends.
“If you talk to developers, there may be some differentiation, but not a lot. The performance gap with open-source models will decrease over time. So the largest models may be indistinguishable. That means OpenAI, Anthropic and others will have to figure out a way to make money, not just revenue, but to become profitable.”
AI Stocks: Startups Face Rising Costs
Capital spending is soaring as the AI battle intensifies.
Meta and Elon Musk’s xAI say they’re training next-generation AI models on clusters of more than 100,000 Nvidia H100 GPUs, up from the industry norm of 10,000 GPU clusters a couple years ago.
Musk’s xAI recently raised $6 billion at a valuation of $40 billion. Musk may also be looking to tap his electric car company Tesla (TSLA) to fund his AI model project.
But AI costs are expected to continue rising.
Dario Amodei, CEO of startup Anthropic, predicted in July that training an AI model could cost $10 billion in two years and eventually $100 billion. Former top researchers at OpenAI founded Anthropic in 2021.
Microsoft has plowed $14 billion into OpenAI. Aside from Amazon’s new investment, Anthropic is separately trying to raise money from other investors at a valuation of up to $40 billion.
“The unfortunate thing for the startups, such as OpenAI and Anthropic, is that there are alternatives from companies like Google and Meta,” said Lorica, who has hosted many industry AI conferences at O’Reilly Media and elsewhere. “They have a lot of profitable revenue derived from advertising and other sources. So they can subsidize the cost of training these ever larger models.”
AI models process “prompts,” such as internet search queries, that describe what a user wants to get. LLMs are made of neural networks — mathematical models that imitate the human brain — that generate outputs from the training data. AI models are ranked in size by “parameters,” numerical values associated with training.
OpenAI’s Long Road To Profitability
The AI spotlight is clearly on OpenAI, the startup that unleashed the current frenzy over artificial intelligence. OpenAI plans to convert from a nonprofit organization to a for-profit company.
The company’s valuation multiple on forward revenue estimates stands at roughly 30 and 40 times — much higher than those of most public companies.
OpenAI has stated that 75% of its revenue comes from consumers paying to query ChatGPT, mostly from $20 monthly subscriptions. The startup also reportedly has over 1 million corporate ChatGPT customers.
Still, it’s unclear how much consumers and businesses will pay for internet search on steroids. OpenAI this month unveiled a new $200-per-month upgrade to ChatGPT Pro, giving users unlimited access to its best AI tools.
The company recently raised $6.6 billion in new funding, valuing the startup at $157 billion, up from $86 billion early this year. The new round was led by venture capital firm Thrive Capital. New investors include SoftBank and Nvidia.
According to the Information, OpenAI is expected to lose around $5 billion this year on revenue of $3.7 billion. Also, OpenAI is expected to burn $14 billion cash in 2026. At the earliest, it might break even by 2029.
AI Stocks: Cloud Prices Falling Fast
But with the stiff competition, a shakeout among AI startups is expected to continue. This was underscored by Microsoft’s acquisition of top scientists of startup Inflection AI in an unusual partnership.
“I do think there’ll be fewer players,” said AI startup SambaNova CEO Rodrigo Liang. “It’s not just about the cost structure. It’s also going to be about the talent that it takes to train these models right. So I think there is a period here for some of these companies to figure out how to create enough value, enough monetization, to be able to sustain themselves.”
Falling prices for cloud AI services are a trend most AI observers are watching. OpenAI, Anthropic and others share cloud revenue with cloud service providers.
“Competition is fierce to provide the most effective and efficient large language models, offering an extraordinary range of price and performance to developers,” said Andrew Ng, who once led AI research teams at Google and Baidu (BIDU).
“Makers of foundation models that can’t match the best large models in performance or the best small models in cost are in a tight corner,” Ng wrote at DeepLearning. AI website.
“What an amazing time to be developing AI applications,” added Ng, who is a member of Amazon’s board of directors. “You can choose among models that are open or closed, small or large, faster or more powerful in virtually any combination. Everyone is competing for your business.”
Databricks Is A Rising Star
On the software front, privately held Databricks, a fast-growing data storage and analysis software maker, and enterprise tech giant Salesforce (CRM) are considered important AI players in the enterprise market.
Databricks’ customers can buy access to OpenAI, Anthropic or open-source models at cloud services. But Databricks builds its own smaller models targeting specific industries.
“The market for AI is incredibly competitive at the moment,” said Databricks’ Frankle. “The way I look at this is that cost is coming down faster than I think quality is going up.”
Frankle was a cofounder of AI startup Mosaic, which Databricks acquired for $1.3 billion in 2023. Databricks was valued at $43 billion in its most recent fundraising. Analysts say it could file for an IPO in 2025.
AI Stocks: Salesforce Makes Big Push
Salesforce is pursuing a similar strategy. The software company built its own open-source LLM called Code-Gen.
“What we are seeing is that the pricing for frontier models is definitely decreasing,” Silvio Savarese, Salesforce’s chief scientist and head of AI research, said in an interview.
Companies in 80% of use cases do not need to process queries with a powerful LLM such as OpenAI’s GPT-4, he said.
“In the consumer space, bigger models are often the best choice,” he said. “But for enterprise use cases, smaller models are an effective alternative in many scenarios. The bigger, super-duper models can be overkill.”
Are AI Models Plateauing?
Further, one big debate is whether AI models have started to plateau.
Amid criticism that LLMs are becoming too generic, companies are racing to add new capabilities. OpenAI this summer released a new LLM code-named “Strawberry” that is capable of enhanced reasoning skills.
“There is the contention that results from scaling up pretraining has plateaued, in part due to increased difficulty associated with sourcing untapped, high-quality, human-made training data,” UBS analyst Timothy Arcuri said in a report. “On the other hand, there is the argument that model performance can be ‘unhobbled’ via a variety of techniques that require significantly more compute resources than more traditional methods.”
Aside from developing AI with greater reasoning capability, another battleground has formed around multimodal systems that can process audio and video as well as text.
Bank of America analyst Alkesh Shah expects innovation to flourish in the consumer space. OpenAI, Google and Anthropic are all building AI models that use web browser information to complete tasks, such as buying a product or booking a flight.
“Despite claims that the pace of AI innovation is plateauing, new and disruptive AI capabilities appear to be emerging on a weekly, if not daily, basis,” he said. “Just a month after OpenAI and Anthropic raised the bar with models that exhibit PhD-level performance and enable apps to use computers, Mistral released its newest model that exhibits best-in-class performance on tasks requiring visual capabilities.”
AI Stocks: Big Tech’s Content Advantage
Meanwhile, a wild card in the AI battle is the debate over content.
The New York Times has sued OpenAI, accusing it of copyright infringement for allegedly using its content to train AI models. Meanwhile, IBD parent Dow Jones has filed a lawsuit vs. AI startup Perplexity.
Giants like Google, Apple and Meta have the financial muscle to fight and perhaps win these legal brawls, or to forge deals with major content organizations such as big news outlets. Apple stock has gained 28% in 2024. Dow Jones parent News Corp has content deals with Google and OpenAI, for example.
Also, tech behemoths like Google and Meta own in-house proprietary data from YouTube, Maps, Instagram and Facebook that can be used to build bigger models. That’s an advantage that AI startups simply cannot match, Lorica said.
Also, Google stock has gained 32% in 2024. It’s one of many AI stocks to watch.
“Only a few large organizations can keep up with these expenses, potentially limiting innovation and concentrating influence over frontier AI development,” said Epoch AI researcher Ben Cottier in a blog post.
“Unless investors are persuaded that these ballooning costs are justified by the economic returns to AI, developers will find it challenging to raise sufficient capital to purchase the amount of hardware needed to continue along this trend.”
Follow Reinhardt Krause on Twitter @reinhardtk_tech for updates on artificial intelligence, cybersecurity and cloud computing.
YOU MAY ALSO LIKE:
Learn The Best Trading Stock Rules From IBD’s Investor’s Corner
Want To Trade Options? Try Out These Strategies
Monitor IBD’s “Breaking Out Today” List For Companies Hitting New Buy Points
IBD Digital: Unlock IBD’s Premium Stock Lists, Tools And Analysis Today