One-Line Summary

AI’s stack advances on two fronts: a $200B Samsung–Broadcom chip pact to expand compute supply, and fresh capital in the works for computer-use agents that automate office tasks.

Industry & Biz

Samsung signs $200B Broadcom AI chip pact through 2030

Samsung Electronics signs a long-term pact with U.S. chip designer Broadcom to expand cooperation across memory, contract chip manufacturing, and advanced packaging, with total value envisaged to exceed $200 billion by 2030, according to Samsung as reported by CNBC. The deal is aimed at supplying AI and high-performance computing customers end to end. ¹

CNBC notes that long-term production commitments from Broadcom, a leading custom AI chip designer, could lift utilization at Samsung’s advanced fabs as more tech companies build their own accelerators instead of relying only on general-purpose GPUs. This aligns with rising demand for specialized chip design and manufacturing partnerships. ¹

Over a five-year collaboration period, the companies plan to combine Broadcom’s ASIC design strengths with Samsung’s manufacturing; Samsung says Broadcom’s next-generation communications chips will be built on its sub-2-nanometer process, and the partners will also collaborate on high-bandwidth memory (HBM) products. Advanced packaging is part of the scope. ¹

The tie-up supports Samsung’s effort to narrow the gap with TSMC: in June 2026, co-CEO Jun Young-hyun said he discussed next-gen foundry work with Nvidia’s Jensen Huang, including future HBM4E and HBM5, and Samsung notes 2-nanometer orders are in discussion after winning a $16.5 billion Tesla logic-chip deal in 2025. Together, these moves signal Samsung’s push to anchor more AI silicon business. ¹

Prentis seeks $100M to build computer-use agents

Prentis is a new AI research lab, launched in April 2026 by Ritankar Das with Reid Hoffman and Marc Pincus, that trains models to operate a computer across everyday software; TechCrunch reports it is in talks to raise $100 million at a $1 billion valuation. The goal is to build agents that can control desktops to automate routine workflows. ²

According to TechCrunch, investor materials and sources say Prentis has signed contracts worth up to $50 million with early customers and targets a $75 million annualized run rate by Q3 2026; the deck notes these figures reflect a fee equal to 20% of savings, are performance-dependent, and are not recognized revenue. Early customers span healthcare management, manufacturing, and goods/apparel. ²

In its pitch deck, Prentis claims its Hive-32B model beats OpenAI’s GPT-5.4 and Anthropic’s Claude Opus 4.6 on WindowsAgentArena and ScreenSpot‑v2, and that a smaller model yields roughly 10× lower cost per task than frontier APIs; TechCrunch says it has not independently verified these results and flags crowded competition from Anthropic, OpenAI, and Mira Murati’s Thinking Machines, including Anthropic’s acquisition of Vercept in 2026. ²

What This Means for You

For teams budgeting AI features, Samsung’s Broadcom pact signals more bespoke accelerators, advanced packaging, and HBM capacity coming into play, which can diversify hardware beyond GPUs and influence availability and pricing in cloud offerings over time. If you buy AI inside SaaS, this may show up as improved capacity or new instance types in vendor roadmaps. ¹

On the software side, computer-use agents are moving from demos to enterprise contracts: if your work involves claims, refunds, or back-office steps across multiple apps and documents, these agents aim to execute the click-by-click steps you document. That puts process clarity and high-quality screen-level instructions at a premium. ²

Non-developer skills grow in importance: mapping workflows, writing unambiguous step lists, and defining exception handling become inputs these agents need to succeed. Operations, product, and CX managers can lead small, scoped pilots and quantify impact in hours saved and error reduction. ²

Treat early vendor claims cautiously: establish baselines and guardrails (least-privilege access, virtual desktops, and audit logs) before trials, and align success metrics to any “percent-of-savings” pricing so outcomes can be validated. Document what counts as “savings” and who approves the measurement. ²

Action Items

1. Ask your vendors about hardware roadmaps: Request specifics on custom accelerators, sub-2nm nodes, HBM-backed instances, and how these could affect pricing, quotas, and SLAs for AI features you rely on.
2. Pick one desktop workflow to scope: Write a 10-step, screen-by-screen guide for a repetitive task (e.g., claims follow-up, invoice coding) and share it with IT to evaluate a limited agent pilot.
3. Set guardrails for agent trials: Draft a one-page checklist covering least-privilege access, use of virtual desktops, data retention, and audit logging; review it with security and legal before any pilot.
4. Build a savings-based ROI template: Create a simple calculator to baseline time/cost today and attribute “20% of savings” pricing fairly, including how to measure errors avoided and rework.