OpenAI just killed its sexiest consumer product to feed the enterprise beast, and the reasons why tell you everything about where the AI economy is actually headed.
The Summary
- OpenAI is launching a new AI model codenamed "Spud" focused on "high-value professional work" and killing consumer products like Sora to free up compute for business customers.
- 95% of ChatGPT's 900 million weekly users don't pay anything, and that math doesn't work when you're valued at $852 billion and bleeding money.
- This is a direct response to Anthropic, which has already planted its flag in enterprise AI and is winning corporate adoption battles.
The Signal
OpenAI CFO Sarah Friar just said the quiet part loud: consumer AI is a loss leader they can't afford anymore. The company is shutting down Sora, its AI video generator, to redirect compute resources toward "Spud," a new model designed for enterprise workflows. Friar called it "a little heartbreaking" but necessary because "it's not the main event right now."
Translation: making cool demos for Twitter doesn't pay the bills. Making Slack summarizers for Fortune 500 companies does.
"We need to make sure that our new model that's coming has enough compute."
The unit economics here are brutal. OpenAI has 900 million weekly ChatGPT users. Only 5% pay for Plus subscriptions at $20/month. That's roughly 45 million paying users generating maybe $900 million monthly, or $10.8 billion annually. Sounds massive until you realize they're valued at $852 billion and burning cash on compute faster than they can raise it. Anthropic, their main rival, is valued at $380 billion and losing money too, but they built for enterprise from the start.
Here's what changed: Anthropic's Claude has become the agent platform of choice for companies actually deploying AI at scale. They're not chasing consumer virality. They're embedding into Salesforce workflows, legal document review systems, and customer service stacks. Boring, profitable, sticky.
Key competitive dynamics:
- Anthropic already owns enterprise mindshare for "serious work" AI
- OpenAI's consumer dominance doesn't translate to B2B contracts
- Both companies are pre-IPO and need revenue growth stories that work on Wall Street
Spud represents OpenAI's admission that they misread the market. Consumer AI is like social media in 2010: massive engagement, unclear monetization. Enterprise AI is like SaaS in 2010: smaller TAM, clear ROI, actual revenue per user. The company that wins enterprises wins the compute war, because they can fund the infrastructure to keep improving models.
The Sora shutdown is the tell. Video generation is technologically impressive and consumer-sexy. It's also compute-intensive with no clear path to profitability. Every second of AI-generated video someone creates for free is compute OpenAI can't use to summarize a VP's email inbox for $500/month per seat.
"You'll see a new model coming from us in short order. We feel very excited about it."
This isn't just a product pivot. It's OpenAI rewiring itself from a research lab that went viral into an enterprise software company that happens to do research. The agent economy needs infrastructure, and infrastructure companies need margins. Consumer products are marketing now, not the business model.
The Implication
If you're building on OpenAI APIs, watch what they optimize for. Spud will be tuned for workplace tasks: structured outputs, reliability over creativity, integration over isolation. Consumer developers will get pushed to the cheaper, older models. Enterprise developers will get the new hotness.
If you're Anthropic, this validates your strategy. Let OpenAI chase consumer attention while you lock in enterprise contracts that pay 50x per user. If you're a startup trying to decide which foundation model to build on, the answer increasingly depends on whether you're B2C or B2B. And if you're a knowledge worker watching this unfold, understand: the agents coming for your job aren't being trained on fun stuff. They're being trained on your inbox.