Anthropic Coreweave Percent
AI Will Do Three-Quarters of Anthropic Employees' Work by 2030 — and the Capital Markets Are Betting on It
Employees at Anthropic — the company behind the Claude large language model — were surveyed about their expectations for AI in their own workflows, and the median response was that artificial intelligence will handle 75 percent of their work by 2030. A comparable survey of American workers broadly put that figure at 28 percent. The gap reflects a stark information asymmetry: the people building the tools have a fundamentally different model of the near future than the people who will eventually use them.
The Anthropic figure is most meaningful when considered alongside the nature of the work involved. Anthropic employees perform research, coding, writing, analysis, and synthesis — precisely the cognitive tasks on which AI is currently advancing fastest. The 28 percent expectation among average American workers likely reflects a more grounded assessment of physical, relational, and domain-specific work not yet subject to meaningful automation. Both numbers, analysts note, can be accurate simultaneously for different categories of workers.
On the financing side, the Financial Times reported that Anthropic is offering Blackstone a 10 percent yield on chip-financing debt — significantly above investment-grade corporate rates, and a figure that reflects the risk profile of AI infrastructure secured by rapidly depreciating Nvidia GPU assets. The offer illustrates the gap between AI infrastructure's revenue-generating potential and the uncertainty around whether that revenue will materialize before the hardware cycle turns.
That uncertainty was on full display in the CoreWeave story. The Nvidia-backed AI cloud company lost roughly 25 percent of its market value this week as investors focused on its heavy debt load accumulated during a period of extremely bullish demand projections. Jensen Huang went on record saying the market turmoil 'is going to pass' — the kind of statement a major strategic partner makes to halt a sentiment spiral — but acknowledged the underlying concern: debt service math being tested against actual utilization rates. Huang's reassurance is a sentiment argument, not a balance sheet one, and the bond market, where the real pressure sits, does not trade on sentiment.
Disney's acquisition of 66 Character.AI employees out of that company's bankruptcy — at a reported cost of $60 million a year, roughly $909,000 per employee annually including a $30 million retention pool detailed in the bankruptcy filing — signals how scarce conversational AI expertise remains even in a market crowded with AI startups. Character.AI raised over $150 million and reached a roughly $1 billion valuation before bankruptcy, a trajectory that illustrates a pattern across first-generation AI consumer companies: extraordinary technical capability paired with difficult monetization. Disney's specific interest in character-consistent conversational AI points toward theme park and streaming applications built around its Marvel, Star Wars, and Pixar intellectual property.
Disney's Character.AI Bet and the Compression of Creative Work
Disney's decision to bring 66 Character.AI employees in-house — at a reported cost of $60 million annually out of that company's bankruptcy proceedings — is less a talent acquisition than a technology strategy made explicit. Character.AI's specific expertise was building AI personas that maintain character consistency over extended, open-ended conversations. Set against Disney's ownership of arguably the world's most valuable character library — Marvel, Star Wars, Pixar, and Disney Animation properties dating to 1928 — the combination points toward a near-term application in theme parks and a longer-term bet on personalized streaming narrative experiences.
The bankruptcy context reframes the deal's financial logic. Character.AI raised over $150 million and reached a valuation of roughly $1 billion at its peak before filing. The acqui-hire at $180 million over three years — including a $30 million retention pool detailed in the bankruptcy filing — represents a steep discount to the peak valuation. It illustrates a pattern visible across the first generation of AI-native consumer companies: the distance between impressive technical capability and a sustainable business model has proven longer than initial venture valuations implied.
The labor dynamics playing out inside creative industries map onto this story in ways that deserve attention beyond the financial headline. Skoda's design chief gave an interview this week describing being 'shocked' by what AI generative design tools have already done to his studio's workflow — producing in hours the volume of design variants that previously required weeks from a full team. His warning about fewer jobs in automotive design is not speculative; it describes a compression already underway in his own department. The pattern is not full elimination of designer roles but a radical reduction in the number of designers needed to produce a given volume of work.
The doctors and lawyers taking AI training gigs at up to $400 a session represent a further dimension of this compression. For a senior physician, such sessions are supplemental income. For a recent law school graduate who cannot yet command associate-level salaries, the AI training economy is a genuine alternative income source that in some cases competes directly with entry-level legal work. That sequence — disruption hitting entry-level knowledge workers first, while senior practitioners remain insulated — inverts the historical pattern of automation, which typically targeted lower-skilled routine tasks before moving up the skill ladder. The current wave appears to be moving in the opposite direction.