Back to News
Market Impact: 0.25

Anthropic says Claude can now use your computer to finish tasks for you in AI agent push

NVDA
Artificial IntelligenceTechnology & InnovationProduct LaunchesAntitrust & CompetitionCybersecurity & Data PrivacyManagement & Governance
Anthropic says Claude can now use your computer to finish tasks for you in AI agent push

Anthropic enabled Claude to control a user's computer (open apps, navigate browsers, edit spreadsheets and attach files), demonstrating a workflow that exports a pitch deck to PDF and attaches it to a meeting invite. The feature positions Claude as a direct competitor to viral agent OpenClaw and follows industry moves (Nvidia's NemoClaw, OpenAI hiring OpenClaw's creator), heightening competitive pressure in personal AI agents. Anthropic warns the capability is early, implements permission prompts and other safeguards, and notes Claude can still make mistakes.

Analysis

This move accelerates a bifurcation: agent utility pushes compute to two places at once — on-device for low-latency, file/system-level automation and cloud for large-model reasoning and retrieval. Near term (0–6 months) expect enterprise pilots to prioritize vendors that can stitch local device access with auditable cloud logs; that amplifies demand for endpoint security and orchestration software while creating incremental spend for GPU/accelerator hardware in notebooks and workstations. Over 6–24 months the second-order effect is likely a surge in demand for secure enclave/TEEs and attestation services (hardware + software) as firms attempt to certify that an agent’s local actions meet compliance — a non-linear revenue opportunity for security vendors with device management footprints. Tail risks are concentrated in two fast-moving vectors: high-profile data-exfiltration or unauthorized actions that force enterprises to pause deployments (days–weeks reaction), and swift regulatory guardrails (months–quarters) that limit agent capabilities or require costly compliance controls. Infrastructure constraints (GPU supply, enterprise-grade edge accelerators) could either throttle adoption or push prices higher, creating a short supply / pricing tail that magnifies hardware vendor earnings. Conversely, rapid commoditization of agent orchestration (open-source toolkits + cloud bundles) is the main secular downside over years — it would compress software ASPs and push monetization toward services and hosting. For positioning, the straightforward hardware beneficiary is NVDA given its dominant accelerator stack and enterprise relationships, but valuation already prices significant upside; preferred exposure is time-limited, convex option structures or a pair to hedge software cyclicality. Security and orchestration names with device management footprints (endpoint protection, IAM, RPA orchestration) sit in the sweet spot for durable revenue across pilot-to-production transitions and should be prioritized for multi-quarter holds. Monitor for a high-visibility breach or regulatory guidance as a binary catalyst that can reset adoption curves within 1–3 months.