OpenAI, Meta and SpaceXAI at war to make enterprise AI dramatically cheaper

There is fierce competition among OpenAI, Meta Platforms ($META), and SpaceXAI (NASDAQ: SPCX) on prices of enterprise AI, not just the scores for the models. They launched their latest models this week, promising improved performance but reduced prices to win the corporate scrutiny.

This year, some firms had their employees work with AI all day long and battle over who can spend the most tokens. This behavior came to be called tokenmaxxing. The finance departments got a hold of the invoices. Employers placed restrictions after providers like Anthropic switched from subscription models to pay-per-use.

OpenAI cuts token use as companies question their AI bills

OpenAI says GPT-5.6 can handle more work with less data processing. Tokens are the small pieces of information an AI system reads and produces. Using fewer lowers the cost of coding, research, support, and other jobs run at scale.

OpenAI Chief Executive Officer Sam Altman discussed that focus during a Thursday interview with CNBC, where he said:

“GPT-5.6 is 54% more efficient on a genetic coding tasks. And also, you know, as good or better as the other best models out there. And we’re really seeing people now start to care about efficiency, understand their spend, get a great ROI. So this is a great step forward for us.

Sam said the model runs faster. He added that business clients are now studying what they pay and what they receive back. “Every enterprise now is thinking about spend and the value they’re getting in exchange for AI,” he said.

OpenAI wants those customers to see it as a stable supplier with strong returns, Sam said. “We want to be the best, most dependable, most reliable, most best ROI partner for enterprises.”

The issue followed him to Sun Valley, where he met partners and customers. Sam said this is the first year AI costs have become a major boardroom topic. “Everyone’s asking what we can do to help reduce spend or increase value.”

SpaceXAI offers cheaper computing power whereas Anthropic leases its entire data center

Elon Musk’s SpaceXAI is making a similar promise with Grok 4.5. The company says the model uses tokens twice as efficiently as rival products. Organizations are now able to compare cost of task with speed and intelligence.

Anthropic competes with Elon’s AI business, yet it is also one of SpaceX’s biggest customers. Users on X recently suggested Elon could remove Anthropic from SpaceX servers whenever he wanted and hurt the company. Elon rejected that idea and wrote, “Not my style.”

He also reversed an older prediction. “I was clearly wrong about Anthropic,” Elon wrote Thursday. He was referring to a September 2025 post where he said, “Winning was never in the set of possible outcomes for Anthropic.” Reports already placed Anthropic first in enterprise AI market share at that time.

Anthropic signed a huge infrastructure contract in May. It agreed to buy 300 megawatts of computing power from Colossus 1 near Memphis, Tennessee. That amount covers the data center’s full output. Musk’s xAI joined SpaceX in February, placing the agreement inside the combined SpaceXAI operation.

Anthropic will pay $1.25 billion each month through May 2029. The contract is worth about $40 billion in revenue for the SpaceXAI unit. Google, owned by Alphabet ($GOOGL), also booked SpaceX infrastructure. Its deal runs through June 2029 and costs about $920 million per month.

Elon said Anthropic had not taken a reckless risk by using a competitor’s hardware. “They are obviously currently the leader in AI,” he wrote. He called Mythos/Fable the strongest released model and said Anthropic would likely have Mythos 2 ready soon. He also said he would never disconnect the company in a way that caused serious harm.

GPT-5.6 reached customers several weeks late because it went through a government approval review. Sam called the process productive and named Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and Director Cairncross as officials involved.

He said it was OpenAI’s first time using that route, so the company learned how to handle the next review. Sam expects the next one to run more smoothly and said the government team showed strong technical knowledge.

“You know, this is our first time through it. So there are things we will learn about how to make it better for the next one, which we’ll get going on soon. And I think it’ll be a much smoother process. But the government’s technical capability was impressive to me,” said Sam.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.