Micron's Multi-Decade AI Bet, the Memory Cycle Peak, and Why We're Holding 35% Cash

The stock market staged a sharp reversal this week after Micron's blowout earnings reignited enthusiasm for AI infrastructure stocks, but is the rally on solid footing, or are investors getting ahead of themselves again?

Puru Saxena, founder of AlphaTarget and growth investor, is joined by AlphaTarget's analyst team to unpack the recent stabilisation in equity markets, the easing of oil prices and bond yields following progress on the Iran-US deal, and why the team remains cautious despite the improving macro economy backdrop. With the Federal Reserve signalling a possible rate hike later this year, the panel discusses why they continue to hold roughly 35% in cash and have shifted away from index hedging towards buying puts on individual overextended names, a key shift in investing approach.

The discussion then turns to Micron's bold earnings call, in which the CEO claimed AI and robotics demand could drive a multi-decade upcycle in memory requiring ten times current memory usage. Puru Saxena and Irnest Kaplan, CFA Charterholder and former Societe Generale analyst, weigh in on why such claims deserve scrutiny, and examine new research from Bernstein suggesting memory earnings could peak in the second half of 2027 before declining sharply, raising questions about how much further the semiconductor rally can run and what it means for disruptive stocks.

Anirban, PhD in Computer Science and former Director of Research and Portfolio Manager at Motley Fool Australia, breaks down Apple's newly unveiled on-device AI strategy, revealing how the company has built its foundation models on top of Google's Gemini rather than developing its own from scratch. The team explores how Apple's approach to personalisation, semantic indexing, and orchestration between on-device and cloud computing models could reshape how enterprises think about deploying AI securely at scale.

Irnest Kaplan also discusses the early signs of enterprises pulling back on AI spending after a period of unrestrained "token maxing," as companies grapple with the cost and security challenges of deploying AI agents across large organisations, a growing theme across business trends today.

Finally, the panel examines the fast-emerging premium inference market, with Oyvind detailing how companies like Groq and Cerebras are using SRAM-based architectures to dramatically accelerate inference speeds, and why Nvidia's own guidance points to a market that could be worth hundreds of billions of dollars