Investors should consider Broadcom (AVGO) as a primary play for custom AI silicon following its successful partnership with OpenAI to develop the Jalapeno chip. Micron Technology (MU) remains a high-conviction growth opportunity as long-term contracts and high-bandwidth memory demand are projected to drive gross margins toward 86%. Despite new competition in custom chips, NVIDIA (NVDA) remains a core holding as major AI labs reaffirm they cannot acquire enough compute to meet demand through 2028. For enterprise software exposure, prioritize companies where the CEO directly leads the AI strategy, as these firms are three times more likely to see a tangible return on investment. Be cautious with Google (GOOGL) in the short term due to model delays and a "talent bleed" of senior researchers to competitors.

OpenAI

• Announced Jalapeno, their first custom-designed in-house chip produced in collaboration with Broadcom (AVGO). • The chip is an ASIC (Application-Specific Integrated Circuit) designed specifically for LLM inference, similar to Google’s TPUs. • Development cycle was exceptionally fast, moving from design to manufacturing tape-out in just nine months. • Despite the new chip, OpenAI reaffirmed they will not reduce orders from NVIDIA (NVDA), stating they "cannot get compute fast enough."

Takeaways

• Vertical Integration: OpenAI is following the path of Apple and Google by designing its own silicon to increase efficiency and reduce long-term reliance on third-party hardware. • Compute Scarcity: The "insatiable" demand for compute suggests that hardware providers will remain in a supply-constrained, high-pricing environment through at least 2028.

Broadcom (AVGO)

• Partnered with OpenAI to develop the Jalapeno chip. • CEO Hawk Tan reported that demand for AI compute is "much more than we can address" and expects this elevated demand to persist through 2028.

Takeaways

• Strategic Partner Status: Broadcom is positioning itself as the primary "enabler" for companies wanting to build custom AI silicon, diversifying its revenue beyond standard networking chips.

Micron Technology (MU)

• Reported "blowout" earnings with 445% year-over-year revenue growth. • Disclosed four long-term contracts with "very large customers" that lock in historically high memory prices. • Management expects the memory market to remain undersupplied for at least the next year, with gross margins projected to expand to 86% in Q4.

Takeaways

• Structural Shift: The AI boom is driving a fundamental change in memory demand (HBM - High Bandwidth Memory), moving the sector away from its traditional "boom and bust" cyclical nature toward sustained growth. • Valuation Gap: Despite massive gains, some analysts suggest these stocks remain "cheap" (trading below 10x projected earnings) because the market has not yet priced in the longevity of the AI buildup.

Anthropic

• Rumors suggest the imminent return of their high-end model, Claude 3.5 Sonnet (Fable), to subscription tiers. • Released Claude Team, an organizational-level tool that integrates AI into company workflows. • Accused Alibaba (BABA) of a massive "distillation attack," allegedly using 25,000 fraudulent accounts to harvest Claude’s capabilities to train Chinese models.

Takeaways

• Vendor Lock-in Risk: As AI tools like Claude become more deeply embedded in organizational context (permissions, memory, and workflows), the "switching costs" for companies to move to a competitor will rise significantly. • Geopolitical Risk: Anthropic is lobbying Washington to treat "model distillation" (copying AI behavior) as a national security threat, which could lead to sanctions against Chinese tech firms.

Google (GOOGL)

• Experiencing a "talent bleed" with several senior researchers from DeepMind (including Noam Shazeer and John Jumper) leaving for rivals OpenAI and Anthropic. • Reports indicate that the next-generation model, Gemini 3.5 Pro, has been delayed from a June release to a July launch.

Takeaways

• Execution Hurdles: The departure of key researchers and model delays suggest internal friction as Google merges its London-based DeepMind team with its Mountain View-based Google Brain team.

Investment Themes & Sector Insights

Enterprise AI ROI

• A KPMG study found that companies where the CEO actively owns the AI strategy are 3x more likely to report a return on investment (ROI). • Actionable Insight: When evaluating companies for investment, look for those where AI is a C-suite strategic priority rather than just an IT department initiative.

The "AI Subsidy" is Ending

• Organizations are starting to see a shift from "free experimentation" to "cost management." • There is a massive jump in enterprise interest for lower-cost LLMs and token-budgeting tools. • Actionable Insight: Software companies that provide "AI Orchestration" and cost-monitoring tools (like OutSystems or Mission Cloud) are likely to see increased demand as enterprises try to rein in spending.

Semiconductor Market Sentiment

• The market is currently in a "gut-check" phase. Recent volatility in stocks like ARM, ASML, and Cerebras shows investor nervousness about a potential bubble. • However, the Goldman Sachs view suggests that consensus forecasts may still be underestimating the total size of the AI infrastructure buildup by 50%.