Anthropic Tops $30 Billion Run Rate, Seals Broadcom Deal

Anthropic's revenue run rate has topped $30 billion, up from $9 billion at the end of 2025, with more than 1,000 business customers spending over $1 million annually (more than doubled since February). The company confirmed collaborations with Broadcom and Google — including a supply-assurance pact through 2031 and access to ~3.5 GW of compute beginning in 2027 — to scale capacity for rapid customer growth. Despite an ongoing legal dispute with the Pentagon that has caused more than 100 enterprise customers to express doubts and poses risks Anthropic says could total billions in lost revenue, the deals boosted market sentiment (Broadcom +6.2%, Alphabet +1.8%) and are sector-positive for AI infrastructure providers.

Analysis

The entry of a viable non-Nvidia accelerator option materially changes bargaining dynamics for hyperscalers and large enterprise AI buyers over a multi-year horizon. Procurement will shift from single-source dependency to a two-vendor negotiating posture, which compresses long-term price/margin assumptions for incumbent GPU suppliers even if immediate share gains are limited by software lock-in.

Second-order winners are firms that sell the ancillary stack — interconnect, power, and data‑center integration services — because customers pursuing multi-vendor resilience will pay upfront integration costs. Conversely, smaller accelerator vendors and niche middleware providers face a squeeze: hyperscalers will prefer standardized, high-volume stacks from established infra players, raising the bar for new entrants.

Key risks are execution and software portability. Real-world migration cycles (model retooling, retraining, validation) are measured in quarters-to-years and can blunt the commercial impact of alternative silicon for a long time; a single convincing benchmark or superior developer tools from the incumbent would reverse sentiment quickly. Legal or contracting headlines around major enterprise customers can produce abrupt short-term volatility even if the structural shift remains intact.

For portfolio construction, treat this as a convexity play: overweight infrastructure incumbents that can monetize non-GPU stacks while hedging the incumbent GPU exposure. The binary outcomes — successful multi-vendor adoption vs entrenched software moat — argue for optioned exposure rather than outright unilateral shorts on market leaders.

Key Decisions for Investors

• Long AVGO equity (size 3-5% net exposure, 9–18 month horizon). Rationale: direct lever to non-incumbent AI silicon adoption and data-center integration deals. Risk/reward: potential +20–35% rerating if execution and design wins materialize; principal risk is under-delivery on performance/volume — hedge with 12-month AVGO put protection sized at 20% notional.
• Long GOOGL (class A or C) cloud exposure vs NVDA: establish a 0.6x pair trade (buy $1 of GOOGL / sell $0.6 of NVDA) over 6–12 months. Rationale: cloud vendor monetization of alternative accelerators should outpace GPU ASP erosion. Risk/reward: asymmetric upside into cloud share gains; downside if NVDA cements software advantages — fund with short-dated call spreads on NVDA to limit financing cost.
• Buy NVDA protective put spread (3–6 month) instead of naked short. Rationale: protects against downside from multi-vendor adoption while capping cost of insurance. Structure: buy a ~10–20% OTM put and sell a further OTM put to lower premium. This preserves optionality for the dominant vendor’s continued strength.
• Event-driven idea: Buy call spreads on AVGO and GOOGL expiring into next major cloud/AI earnings and benchmark publication windows (3–9 months). Rationale: positive execution or benchmark announcements are binary catalysts; structured call spreads cap premium while capturing upside. Close on catalyst miss or once 50% of max profit is reached.