The numbers arriving this week from OpenAI's latest fundraising round tell a familiar story: $40 billion raised, valuation doubled to $300 billion in five months, and SoftBank betting heavily that it can force the company toward profitability by December 2025. What's less discussed is what that scale actually requires. Of those billions, $18 billion will flow into Project Stargate—the sprawling infrastructure play that will house the GPUs training the next generation of models. This is the bet: that despite warnings from the Coreweave IPO (which landed at a fraction of expected value), the demand for capacity remains insatiable.
Yet the week also showcased the other side of AI's expansion: the moment when capability bumps into culture. OpenAI's new image generation using the GPT-4o model instead of DALL-E 3 suddenly made style transfer trivial. "Transform this photo in Studio Ghibli style" became a viral prompt, flooding social feeds with anime-filtered self-portraits. One Ghibli co-founder called it "an atrocity." The tension here is real but predictable: the model saw millions of images from Ghibli films during training, learned to encode that visual style into mathematical space, and now distributes it on demand. No permission was sought. No royalty exists. The question of whose intellectual property lives inside these embeddings remains legally murky—a gray area that extends beyond animation to software, writing, and code.
Anthropic's Economic Index offers a different lens on adoption: they parsed a million Claude conversations against the U.S. Department of Labor's O*NET database to map which tasks, roles, and industries are actually using AI. Software development dominates (37% of conversations from 3% of the workforce), while physical labor—cooking, healthcare support, transportation—barely registers. More striking: 55% of usage is augmentation (making people better at their work) rather than automation. That's not the disruption narrative most venture capitalists sold. It suggests the market is already correcting toward what AI is actually good for: scaling human capability, not replacing it. Whether that recognition spreads faster than the funding does remains the open question.