Mark Cuban Says OpenAI Is ‘Sh*tting Away’ Its Money on AI Infrastructure: Says ‘They’ll Never Get’ a Return on Their Trillion-Dollar Spending Spree

Mark Cuban is not an AI skeptic. He thinks workers and companies that ignore artificial intelligence are already falling behind.

But when it comes to the staggering sums being poured into the biggest AI labs, Cuban’s view is much harsher.

Asked by Alex Kantrowitz on the Big Technology Podcast about the return needed to justify the massive investment going into companies like OpenAI, Cuban did not hedge.

“They’ll never get it,” Cuban said. “Yeah, they’re just sh*tting away that money.”

The comment is striking because Cuban’s broader view of AI is overwhelmingly bullish. He has argued that companies should be using large language models now, that workers need to understand AI agents, and that the technology is already creating an advantage for early adopters. But his optimism about AI’s usefulness does not extend to every business model in the industry.

In Cuban’s view, there is a major difference between AI as a transformative technology and AI as a capital-intensive race among foundational model companies trying to justify enormous infrastructure spending.

“It’s not that AI is not going to work,” Cuban said.

His issue is not with technology. It is with economics.

Read more: AI is “one of the largest industries ever” ...and it’s already disrupting this $1T market

OpenAI has reportedly raised one of the largest private funding rounds in history, while the broader AI industry is racing to secure chips, data centers, cloud capacity, and power. The logic behind the spending is straightforward: more compute allows companies to train more powerful models, serve more users, and build the infrastructure needed for a future where AI becomes embedded into work, search, software, consumer devices, and enterprise systems.

The Math Doesn’t Make Sense

Cuban is not convinced the math works.

He pointed to Apple (AAPL) as an example.

“Look at Apple, right?” Cuban said. “They haven’t spent next to anything, but they’ve got a foundation where they can just plug and play into their devices.”

That comparison gets to the center of Cuban’s argument. Apple does not need to win the foundational model arms race in the same way an AI lab does. It already owns the customer relationship through the iPhone, Mac, iPad, Apple Watch, and its broader ecosystem. If AI becomes a feature embedded into devices, apps and operating systems, Apple can potentially benefit without spending like a frontier lab trying to build everything from scratch.

Read more: This Pre-IPO stock is up 4,000% already 

That is the kind of position Cuban appears to prefer: distribution, integration, and leverage over brute-force infrastructure spending.

He also argued that some of the current fear around data center capacity may be overstated.

“There’s a lot of FUD being put out about the spending,” Cuban said, referring to fear, uncertainty, and doubt. “‘We’re going to spend a trillion dollars because we need all this data center capacity.’”

Cuban’s counterargument is that compute will not remain as expensive or constrained as today’s spending plans imply.

“The ability to process gets faster, cheaper, and it’s going to happen faster and quicker than people expect,” Cuban said. “So I think a lot of the numbers that they’re throwing out there aren’t going to come to fruition.”

That is why, he added, “they’re full of sh*t, it’s not gonna happen.”

Cuban’s skepticism lands at a moment when the AI boom is increasingly defined by infrastructure. The conversation has moved far beyond chatbots. AI companies and their partners now need chips, servers, energy, cooling, land, data centers, networking equipment and long-term cloud commitments. The industry is not just building software. It is trying to build a new industrial layer for the internet.

The problem, as Cuban sees it, is that spending at that scale requires extraordinary returns — and no one knows yet whether the foundational model business can deliver them.

“Those who have just gone all-in, some of them are spending more cash than they have available,” Cuban said. “And I understand why. I’m just not convinced for all of them it’s going to work.”

Market Structure

His uncertainty comes down to market structure.

“We don’t know if the business of foundational models — the ChatGPTs, Geminis, Grok, Claude, et cetera — is going to be like the streaming industry, where there’s one leader and a bunch of players make money,” Cuban said. “Or search, where there’s effectively one company.”

That analogy may be the most important part of Cuban’s critique.

If AI foundation models look like streaming, then several companies could survive. Netflix (NFLX) may be the category-defining name in streaming, but Disney (DIS), Amazon (AMZN), Apple (AAPL), YouTube, Warner Bros. Discovery (WBD), and others have all found ways to participate, even if profitability varies. A market like that could support multiple major AI labs, multiple subscription products, multiple enterprise platforms and multiple infrastructure winners.

But if AI foundation models look like search, the outcome could be much more brutal.

Search became one of the most valuable businesses in technology, but it did not create a balanced field of equally powerful winners. Google (GOOG) (GOOGL) captured the dominant position, and the economics of search became heavily concentrated. If foundational AI follows that pattern, then billions or even hundreds of billions of dollars could be spent by companies that ultimately become also-rans.

That is Cuban’s warning: the technology can be world-changing while still producing a punishing investment outcome for many of the companies building it.

“It’s going to be very difficult,” Cuban said. “I don’t think it’s going to pay off the way they expect.”

Cuban has repeatedly embraced new technologies early, but he has also been deeply attentive to business models. Cuban did not become a billionaire merely by believing the internet would be important. He built and sold Broadcast.com during a moment when streaming media became valuable to a much larger internet platform. His fortune came from understanding both the technology shift and the exit window.

That matters because Cuban is not dismissing AI the way old-guard executives once dismissed PCs or the internet. He is making a more precise distinction. AI adoption may be inevitable. AI value may be enormous. But that does not mean every company spending like it is guaranteed to own the future will earn an acceptable return.

In fact, his argument suggests the opposite: the more obvious AI’s importance becomes, the more aggressively companies may overspend to avoid missing it.

Could It Be a Bubble?

That is how bubbles can form around real technologies. The internet was real, but many dot-com companies failed. Fiber networks were important, but the early-2000s telecom buildout still created massive overcapacity. Streaming changed media, but not every streaming service became a great business. Electric vehicles are real, but not every EV startup justified its valuation.

Cuban appears to see a similar risk in AI infrastructure. The models may keep improving. Businesses may keep adopting them. AI agents may become useful. But the returns from the foundational model layer may be less certain than the spending suggests.

The biggest winners may not be the companies spending the most on model training. They could be companies that already have users, devices, enterprise relationships, proprietary data, workflows, or distribution. They could be infrastructure suppliers. They could be chipmakers. They could be cloud platforms. They could be software companies that use AI to make existing products more valuable without absorbing the full cost of building frontier models.

That is why the Apple example matters. Cuban is pointing to a company that can benefit from AI by integrating it into products people already use. Apple does not necessarily need to convince the world to subscribe to a standalone model. It can make AI part of the device experience.

For OpenAI and other model companies, the challenge is different. They must turn massive infrastructure spending into durable revenue, defensible margins and long-term customer dependence — all while competing against each other, open-source models, hyperscalers, device makers and enterprise software companies.

That is a much harder path than simply proving AI is useful.

Cuban’s comments are likely to resonate because they capture the strange split in the current AI debate. On one side, AI tools are becoming more capable and more embedded in work. On the other, the financial commitments required to stay at the frontier are becoming so large that even believers are asking whether the industry can possibly earn its way back.

Cuban’s answer is blunt: not everyone can.

And for the companies spending as if the future is already theirs, he thinks the bill may be far larger than the payoff.