AI capex shift puts HBM memory and optical networks in focus

Original Insights Roundup

Holding fewer than 5 stocks concentrates idiosyncratic risk too much, while holding more than 15 dilutes edge to the point where an index fund is likely the cleaner choice. A workable framework is 5 to 10 stocks for ordinary investors, paired with a three-layer position structure and a VIX-based cash allocation method. 1
The “small account must all-in 1–2 names to double” logic misses non-fundamental blow-up risk: regulation, geopolitics, industry policy shocks, management scandals, and disruptive competitors. Diversification is not weak conviction; it is respect for market uncertainty and gap risk. 2

The real moat in compounding is patience. Thinking in seven-year timeframes instead of three creates edge because most participants are unwilling to sit through the slow grind of a “get-rich-slowly scheme.” 6

U.S. equity sentiment is not crowded even with the S&P and Nasdaq closing the week just shy of ATHs. Goldman’s U.S. Equity Sentiment Indicator moving closer to neutral suggests positioning still is not stretched and remains below prior peaks, which weakens the usual “everything is euphoric” bear case. 8
Wall Street consensus is becoming aggressively bullish on $SPX, with 8,000 increasingly seen as achievable within the next 12 months, or even sooner. The notable takeaway is not just the target, but how crowded that upside call is becoming. 11

$MSFT looks like a classic multiple/compression mismatch versus fundamentals. Two years ago it traded around ~$420/share with Azure growing ~30%, $235B commercial RPO, and AI revenue barely measurable. Today it is still around ~$420/share, but Azure re-accelerated to 40% growth, Commercial RPO reached $627B (+99% YoY), and AI scaled to $37B AI ARR (+123% YoY) with 20M+ Copilot seats (+250%). Price flat, business massively stronger. 9

Memory is being repriced from a cyclical commodity bucket into a core AI infrastructure profit pool. $MU, Samsung, and SK Hynix are expected to generate $646B in operating income in 2027, driven by HBM displacing standard DRAM at roughly a 3:1 wafer ratio while inference, agents, and robotics expand demand. 17
The Micron setup points to a deeper technology transition, not just a volume cycle. Micron’s HBM4E is expected to ramp in CY27, starting with JEDEC standard and then moving into custom HBM4E later. It will be Micron’s first 1-gamma-based HBM generation, and both versions will use TSMC for the base die, marking a break from Micron’s prior use of internal base die through HBM4. That signals tighter ecosystem integration and a more customized memory roadmap for AI workloads. 14

The choke point in AI infra is shifting upward from compute to interconnect. Goldman sees the AI optical networking market growing nearly 10x to $150B+ by 2028 as clusters evolve into multi-rack systems such as $NVDA Rubin Ultra NVL576. The key read-through: the next leg of AI capex is increasingly about the fabric that makes hundreds of GPUs operate like one system. 15

Nokia is increasingly being rerated as an “AI-era connectivity infrastructure company.” The near- to mid-term earnings hook is clearest in optical interconnect / AI data-center interconnect, and the point is that this is already showing up in recent quarterly results rather than being a pure narrative trade. 19

The GRVT campaign with Binance Wallet looks like a likely TGE prelude rather than a routine promo. The logic is straightforward: users depositing funds via Binance Wallet receive GRVT Points +40%, and the existing Points system already maps cleanly to a TGE-oriented airdrop mechanism. That makes the partnership read like a deliberate pre-launch incentive structure. 18

#SPX #MU #MSFT #NVDA