HOLD into May 20 earnings. Goldman expects $80B and $1.86 EPS (above consensus $78B/$1.74) but explicitly notes "bar for outperformance is relatively high." Wait for the print. On a clean beat with strong Q2 guide ≥$87B and Rubin H2 confirmation: deploy in $195–225 zone. On a miss or guide-down: $180-195 re-entry zone.

Strong Add. Best risk/reward in megacap AI outside of MU. Forward P/E 25× with 30%+ growth, 66% gross margin, AI capacity sold into 2027, and 2nm volume production H2 2026. Taiwan geopolitical risk is structural — price in a permanent discount, not a binary risk.

Strong Add. Best complement to MU/PLTR — custom silicon/networking vs memory and software platform. The ASIC moat is sticky: switching costs from a hyperscaler's own silicon programme are multi-year. Deploy after the May 20 NVDA print confirms hyperscaler capex is intact, then ahead of AVGO's own Q2 print (June 3). At $400-430, the risk/reward is asymmetric given the $73B backlog.

Strong Add. Accumulate at $370–400. Long-term compounder — the only company that owns chips (TPU), models (Gemini 3), infrastructure (GCP), and global distribution (Search, YouTube, Android). Cloud growth +63% vs AWS ~17% represents a serious competitive gap being established. The stock hit all-time high $398.37 May 8 — new highs often beget new highs in genuine compounders.

Strong Add on pullback. The 6.7% post-earnings drop on a 0.5% revenue miss vs $1.8B is a non-fundamental derate. Q4 guide acceleration to $1.91–2.05B (sequential +12%) is the more important signal. At $300-315 the risk/reward is asymmetric. PT range $170-455 reflects analyst uncertainty, not stock weakness.

Add (pullback) only. NEE belongs in a concentrated AI portfolio as a hedge against AI capex digestion — when NVDA/MU/TSM get hit, utilities hold. The AI power narrative is real (hyperscalers need 24/7 reliable power away from grid delays) but NEE's share of that wallet is still small and unproven at scale. Don't overpay: the sector premium at 22× is pricing in perfect execution.

Strong Add. This is the most mathematically compelling setup in the universe right now: a 14-year compounder that has never had a bad year trading at its lowest valuation relative to own history, while its AI revenue (Now Assist) is accelerating and management raised full-year guidance on the same print that crashed the stock. The Iran war FX headwind (75bps) is the mechanism, not a structural deterioration. The business is exactly what it was when it traded at 95-110× forward earnings.

Strong Add. CRM is the value anchor in an AI portfolio skewed toward expensive names. For a portfolio with MU (deeply undervalued chips) and PLTR (most expensive software), CRM provides the middle path: real AI monetisation at a value multiple. The 13.8× fwd P/E with $15B OCF and $50B buyback makes this a durable hold even if Agentforce takes longer to ramp. Best risk-adjusted software pick in the cohort. Accumulate ahead of the Q1 FY27 print (~May 28).

Add (smaller size) — but only on pullback. The Q1 print was excellent: +32% growth (re-accelerating), raised guide, GPU monitoring as concrete new revenue. But the stock surged 31.3% on the day. Initiating at $190 after a 31% spike is poor discipline. The thesis is intact at $165-185 — a 10-14% pullback from spike levels. Keep size modest given PEG 3.35: conviction add at a fair price, not a concentrated bet.

HOLD only — DO NOT ADD at $144. The combination of: (1) Burry still short with June 2027 puts and Dec 2026 $100 puts; (2) Peter Thiel selling $580M in 90 days; (3) Jefferies "Underweight" with $70 PT; (4) 110× forward P/E with institutional ownership at 56% (low) — means the asymmetry is against adding. The business is extraordinary. The valuation assumes perfection. Hold what you have. If it corrects to $100-115, that is the reload zone.

HOLD — do not add at $757. The thesis is intact: HBM supercycle, locked pricing, 74.9% gross margin, FQ3 guide at $33.5B. But the stock is up 800% in 12 months and just had its best week since 2008. Adding here is chasing a parabola. The risk/reward asymmetry is poor: upside to $900-1000 if cycle holds; downside to $450-550 if HBM4 Vera Rubin design win fails or cycle turns. Better to add the thesis at $600-650 where the margin of safety returns.

Add on pullback. PEG 0.91 is the most disciplined valuation entry in the IO Fund — better risk-adjusted than ARM at similar allocation size. The thesis is real: MI300X ramping, MI400 roadmap intact, and AMD is the only credible alternative when hyperscalers want to diversify away from NVDA single-source dependency. At $408 the stock is 21% above IO Fund's entry — wait for a pullback to $300-340 before initiating. Do not chase above $420 (already in "What NOT to Do").

Watch — wait for pullback to $150-170. The thesis is genuine: every Blackwell rack needs ALAB's connectivity chips, and this is not replicable quickly by competitors. But at $195 the average analyst target has been met, 8.43% short interest signals valuation risk, and PEG 2.34 provides no margin of safety on a slip. IO Fund's 65% gain from $118 gives them margin of safety that new entrants do not have. A 15-22% pullback restores the risk/reward.

Watch. IO Fund's 8% allocation at $518.60 is a strong conviction call — and at $1,340 they have nearly tripled. For new entry, the stock is at all-time high territory and the 251% YoY growth makes 2027 comparisons increasingly difficult. The AI inference NAND thesis is real, but this is structurally more cyclical than MU's locked-price HBM business. Wait for a meaningful pullback to $1,050-1,150 (20-22% from current) before initiating. Do not add a second memory name on top of an existing MU position without trimming first.

Watch — the gate is the Q2 print on August 6. For IO Fund, this is one of their best trades: $29.05 entry, 442% gain, and the thesis (800G ramp) is still unfolding. For new entry at $157 the math is entirely different: you have no cost basis advantage, single-customer concentration risk, an unproven Texas fab ramp, consistent insider selling, and a stock up 1,027% in 52 weeks. The only responsible entry is after Q2 confirms the H2 ramp is on track — and at a price that reflects that information, not the current optimism. Watch $120-140 as a post-confirmation entry zone.

Watch — wait for post-spike consolidation. The NVIDIA $2B investment is a genuine structural signal for co-packaged optics as an AI infrastructure category. Q3 +90% YoY revenue is real. But at $893 the RSI was 84 on the announcement day, and the stock has moved from $434 in a matter of months. IO Fund's trade from $434 was excellent — new entry at $893 is a different risk profile entirely. Let the spike consolidate, then look for $800-880 as the entry zone. COHR is a cleaner risk/reward in the same optics layer at current prices.

Strong Add. This is the variant perception thesis: the market sold META -10% on a capex raise that is identical in character to the capex raises by AWS and Google Cloud that ultimately drove those businesses to higher multiples. At 22× fwd P/E for 33% growth with a 21% discount to intrinsic value, META is priced as if the AI ad flywheel is already breaking down — it is not. IO Fund is building into this weakness. The $580-630 zone is the right accumulation range. Reality Labs drag is the known overhead — price it in and move on.

Add on pullback. GEV is the highest-quality AI power infrastructure name — better earnings visibility than Bloom Energy (BE) and a cleaner AI power story than NextEra (NEE). IO Fund entered at $686; at $1,045 the stock is +52% from that entry and is now trading near DCF fair value ($845). The narrative fair value of $1,175 requires perfect execution on the FY27 earnings recovery and continued hyperscaler order momentum. At $900-970 the margin of safety returns. The $163B backlog makes this a genuinely durable compounder — but price discipline matters at these levels.