I agree, we are in bubble territory. Three years in and OpenAI's bleeding $12B per quarter while 95% of users pay nothing. Anthropic breaks even by 2028, Google doubled market share in 12 months. The memory chip comparison nails it - tokens are commoditizing faster than OpenAI can figure out how to charge for them.
But your article misses something: the differentiation already moved up the stack. Claude Code has 42% developer share vs OpenAI's 21% not because of better tokens, but because it's baked into how developers actually work. Switching costs show up at the workflow level, not the chatbot level.
And Anthropic's path to $70B revenue with 80% enterprise proves you don't need a consumer circus to survive - you need customers who'll actually pay. OpenAI's building factories to make a commodity product while their competitors are building the applications that matter.
800M users is a vanity metric when you're burning $1.4T in compute commitments you can't monetize.
Claude Code is the app layer built over the model.
As I wrote,
The true differentiation is how they are applied — the application layer. But as we saw with DeepSeek — which “borrowed” tokens from OpenAI — this doesn’t require massive investment in token-making. Tokens will always be needed, and we’ll always need more of them, but the tokens are like the oil fueling AI applications.
Fair point on the application layer. But Anthropic just acquired Bun yesterday to own the Claude Code stack - $1B run rate in 6 months. And they're prepping for a $300-350B IPO. (God bless them)
it's a company betting they already won the enterprise layer while OpenAI burns cash chasing consumers. OpenAI is spending like they're building memory fabs. Anthropic is spending like they're building SAP.
I agree, we are in bubble territory. Three years in and OpenAI's bleeding $12B per quarter while 95% of users pay nothing. Anthropic breaks even by 2028, Google doubled market share in 12 months. The memory chip comparison nails it - tokens are commoditizing faster than OpenAI can figure out how to charge for them.
But your article misses something: the differentiation already moved up the stack. Claude Code has 42% developer share vs OpenAI's 21% not because of better tokens, but because it's baked into how developers actually work. Switching costs show up at the workflow level, not the chatbot level.
And Anthropic's path to $70B revenue with 80% enterprise proves you don't need a consumer circus to survive - you need customers who'll actually pay. OpenAI's building factories to make a commodity product while their competitors are building the applications that matter.
800M users is a vanity metric when you're burning $1.4T in compute commitments you can't monetize.
Claude Code is the app layer built over the model.
As I wrote,
The true differentiation is how they are applied — the application layer. But as we saw with DeepSeek — which “borrowed” tokens from OpenAI — this doesn’t require massive investment in token-making. Tokens will always be needed, and we’ll always need more of them, but the tokens are like the oil fueling AI applications.
… But many model makers will be wiped out.
Fair point on the application layer. But Anthropic just acquired Bun yesterday to own the Claude Code stack - $1B run rate in 6 months. And they're prepping for a $300-350B IPO. (God bless them)
it's a company betting they already won the enterprise layer while OpenAI burns cash chasing consumers. OpenAI is spending like they're building memory fabs. Anthropic is spending like they're building SAP.