Enterprise AI spending is spiraling out of control, with Uber burning through its entire 2026 AI budget by April and JP Morgan analysts warning that corporate token bills are unsustainable.

The biggest AI companies in the world are about to go to war over pricing, and the timing could not be more awkward.

OpenAI is reportedly considering significant reductions to the token prices it charges developers and enterprises, according to the Wall Street Journal. The cuts would be a direct response to Anthropic's rapid capture of corporate customers, particularly through its breakout coding agent Claude Code. But neither company has turned a profit, both have filed confidential IPO paperwork with the SEC, and Chinese open-source alternatives already offer frontier-quality AI at a tiny fraction of the cost.

The question is no longer whether AI pricing will come down. It is whether the two most valuable AI startups on Earth can survive the fall.

Why OpenAI Is Considering Price Cuts Now

The pressure on OpenAI is mounting from multiple directions.

Anthropic's annualized revenue reportedly grew more than fivefold in under six months, jumping to an estimated $47 billion by May 2026 from roughly $9 billion at the close of 2025. The catalyst behind that explosive growth was Claude Code, the company's AI coding agent that became a must-have tool across enterprise engineering teams. The talent pipeline has followed the money, with former OpenAI founding member Andrej Karpathy joining Anthropic in late May. Q2 2026 marked Anthropic's first profitable quarter, a milestone that OpenAI has yet to reach.

Meanwhile, OpenAI's own financial position tells a very different story. The company posted a negative 122% adjusted operating margin in Q1 2026, effectively spending more than double what it earned on every dollar of revenue. ChatGPT's share of global generative AI web traffic declined by nearly 24 percentage points between May 2025 and April 2026, falling to just 53.7%, according to Decrypt.

For the first time, more companies tracked by the Ramp AI Index are paying for Anthropic than for OpenAI, a shift that would have been unthinkable just a year ago.

Sam Altman acknowledged the tension publicly. At a recent event, he said OpenAI would find "a lot of ways we can help people get more value for less spend," according to the Wall Street Journal.

The IPO Problem With Cutting Prices

Both companies are heading toward the public markets at historically high valuations. OpenAI's most recent private valuation hit $852 billion in March 2026, a dramatic leap for a company that turned down a $97.4 billion acquisition offer just months earlier. Anthropic closed a record-breaking $65 billion Series H funding round in late May that pushed its valuation to $965 billion, surpassing OpenAI for the first time.

Deliberately slashing the price of your core product right before going public is an unusual strategy. Wall Street typically wants to see revenue growth and a clear path to profitability, not a margin-crushing price war.

The counterargument is that lower prices could accelerate adoption, lock in more developers on OpenAI's platform, and ultimately drive higher aggregate revenue through volume. But as one Bloomberg analysis warned, mutual price cuts could be devastating for both firms, given that neither has demonstrated sustainable profitability.

Together, OpenAI and Anthropic are projected to spend nearly $65 billion in 2026 alone on computing, training, and operations.

DeepSeek and the Open-Source Floor

This is where the narrative gets uncomfortable for both Western AI labs.

DeepSeek, the Chinese AI company that disrupted the industry in early 2025 with its open-source models, has already demonstrated that frontier-quality AI does not have to cost what OpenAI and Anthropic charge. The gap is so extreme that it has even spawned underground markets reselling Western AI access at steep discounts, further eroding the pricing power of U.S. labs. The official pricing difference is staggering:

DeepSeek V4 Flash costs $0.14 per million input tokens and $0.28 per million output tokens
OpenAI's GPT-5.5 costs $5 per million input tokens and $30 per million output tokens
Anthropic's Claude Opus 4.7 costs $5 per million input tokens and $25 per million output tokens

That makes DeepSeek roughly 36 times cheaper on input and over 100 times cheaper on output compared to GPT-5.5, according to AI Pricing Guru. For enterprises processing hundreds of millions of tokens monthly, the savings are measured in the hundreds of thousands of dollars.

Open-source inference providers compound the advantage. Because Chinese labs release their model weights freely, inference providers do not pay licensing fees. As Tommy Shaughnessy of Delphi Ventures noted in a widely shared analysis, "The model is the single biggest cost an inference provider has, and they get it for free."

As long as China's AI labs keep open-sourcing frontier-grade models, the floor on AI pricing will keep falling toward zero. Any margin recovery at OpenAI or Anthropic becomes what Shaughnessy called "a math problem with no clean solution."

The Tokenmaxxing Hangover

The price war discussion is arriving at a moment when enterprises are already in crisis over AI costs.

The industry coined the term "tokenmaxxing" to describe the practice of burning through as many AI tokens as possible, often without measuring return on investment. What started as an aggressive adoption strategy has turned into a budget nightmare:

Uber burned through its entire 2026 AI coding budget by April, driven largely by Claude Code usage
Microsoft cancelled internal Claude Code licenses for employees in several product divisions, redirecting engineers to its own GitHub Copilot CLI
Amazon reportedly had employees spinning up meaningless AI tasks to inflate usage stats on internal leaderboards
One unnamed company reportedly received a Claude invoice exceeding $500 million for a single month after neglecting to configure usage caps
JP Morgan analysts published a note titled "AI Bills Are Out of Control," warning that corporate token spending was unsustainable

According to J.R. Storment, executive director of the FinOps Foundation, companies started calling in April saying they were already three times over their entire 2026 token budget. In response, the Linux Foundation announced plans for a Tokenomics Foundation to bring cost discipline to AI spending.

Palantir CEO Alex Karp went further, comparing the tokenmaxxing phenomenon to an addiction during his AIPCon appearance.

What Happens Next

The AI industry appears to be entering a new phase where pricing power, not model performance, will determine market share.

Google has already moved aggressively, slashing its consumer AI Plus subscription from $7.99 to $4.99 per month. If OpenAI follows through with deep API price cuts, Anthropic will likely match them, creating exactly the kind of deflationary spiral that investors fear.

For developers and enterprises, cheaper tokens would be welcome. But for investors evaluating IPO valuations north of $800 billion for companies that have never turned a sustained profit, the math becomes increasingly difficult.

The irony, as several analysts have pointed out, is that DeepSeek proved this was coming over a year ago. The company showed the world that frontier-quality AI could be built and served for a fraction of what Western labs were charging. OpenAI and Anthropic are now fighting a pricing battle that open-source models already won.

The only scenario that changes this trajectory, according to Shaughnessy's analysis, is if China's labs reverse course and go closed-source, which would remove the cost floor that open-source models have established. So far, there are no signs of that happening.

FAQs

Why is OpenAI considering cutting its token prices?

OpenAI is weighing price reductions because Anthropic has been rapidly winning enterprise customers, particularly through its Claude Code coding agent. Anthropic's annualized revenue reportedly grew from $9 billion to $47 billion in just five months, and more companies tracked by the Ramp AI Index now pay for Anthropic than for OpenAI. The proposed cuts are a defensive move to retain market share ahead of both companies' expected IPOs.

How much cheaper is DeepSeek compared to OpenAI and Anthropic?

DeepSeek's models are dramatically cheaper than Western alternatives. DeepSeek V4 Flash costs approximately $0.14 per million input tokens, compared to $5 per million for both OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7. On output tokens, DeepSeek charges $0.28 per million versus $30 for GPT-5.5 and $25 for Claude Opus 4.7. This makes DeepSeek roughly 36 to 107 times cheaper depending on the token type.

What is tokenmaxxing and why does it matter?

Tokenmaxxing is a practice where employees and teams consume as many AI tokens as possible, frequently with no measurable productivity gain to show for it. The term gained traction in 2026 after companies including Amazon, Uber, and Meta deployed internal leaderboards that tracked token consumption as a proxy for productivity. The resulting spending explosion has pushed many enterprises to rethink their AI budgets, with some exceeding their annual token budgets within the first four months of 2026.

Could an AI price war affect OpenAI's and Anthropic's IPO plans?

Yes. Both companies have filed confidential IPO paperwork with the SEC, with OpenAI valued at $852 billion and Anthropic at $965 billion. A price war could compress revenue and margins at precisely the moment when public market investors expect growth and a path to profitability. OpenAI is not expected to reach profitability until approximately 2030, while Anthropic projects breakeven around 2028.

Will AI token prices keep falling?

Multiple factors suggest prices will continue declining. Open-source Chinese models provide frontier-quality performance at a fraction of Western pricing, and inference cost improvements driven by hardware advances are expected to reduce costs further. Gartner estimates that by 2030, inference on large models will cost AI firms nearly 90% less than in 2025. However, analysts warn that cheaper per-token prices may not translate to lower total bills, because agentic AI systems consume far more tokens per task than traditional models.

OpenAI Price War With Anthropic Could Reshape AI Costs, but DeepSeek Already Set the Floor - Memeburn