Anthropic brings Claude Cowork to mobile and web as usage data shows most users aren’t coding

Anthropic launched Claude Cowork on mobile and web, moving from desktop-only to cross-device execution where tasks can continue in the background and be reviewed from a phone. In beta, Cowork usage from 1.2M sampled sessions (May 11–May 31) across 600k+ organizations shows business process/operations at 33.4% and content/copywriting at 16.4%, while software development is only 8.7%—a shift toward “work around the work.” The rollout is supportive for enterprise adoption (with double usage limits extended through Aug 5), but comes amid reported security concerns and heightened geopolitical/export constraints affecting enterprise customers.

Analysis

The incremental winner is the infrastructure layer that monetizes autonomous work, not the app layer itself. If enterprise users are comfortable letting agents run unattended in the background, the bottleneck shifts to compute, storage, and reliability—good for capacity providers like WULF, but only if contracted demand converts faster than financing and buildout costs. The stock market usually overpays for the first proof of usage and underprices the risk that utilization lags the narrative by 2-4 quarters.

BABA is a cleaner loser on the geopolitical signaling than on direct economics. The real second-order effect is not the ban itself; it is that Chinese enterprises will accelerate migration toward domestic models and local stack control, which fragments the global enterprise AI market and reduces the odds of a single winner-take-most platform. For MSFT, this is more a margin and bundling issue than a share-loss story: if standalone agents are good enough, Copilot must justify its price through distribution and workflow integration, not model quality alone.

The contrarian point is that "work around the work" is attractive product language but a weak monetization proxy. These tasks are episodic, highly substitutable, and easy for customers to multi-home across vendors, so usage growth does not automatically translate into durable ARPU expansion. The thesis breaks if security incidents, procurement friction, or slower-than-expected conversion from trial to paid seats show up over the next 1-3 quarters.

Key Decisions for Investors

• Long WULF on pullbacks, but only as a 6-18 month infrastructure call option; size small until financing terms and customer credit quality are confirmed. Falsifier: if the lease is followed by dilution or capex delays, the equity can retrace sharply despite the headline.
• Buy BABA 1-3 month put spreads on strength to express geopolitical and access-risk overhang. Risk/reward is best if the market fades the headline; thesis is invalidated if there is no follow-through on restrictions and management signals improved China-enterprise AI access.
• Do not short MSFT on this print; instead, treat it as a monitoring item for Copilot pricing power and M365 attach. If next quarter shows seat growth without ARPU lift, re-evaluate a relative-value short against higher-multiple AI software exposure.
• No trade on TILE or TISI from this catalyst set; there is no clear revenue or margin transmission mechanism.