Is the AI infrastructure buildout still the highest-conviction trade for 2026?
The short answer is yes, with a critical caveat about digestion speed. NVIDIA’s fiscal Q2 2027 delivered $96.2B in revenue (+106% YoY, +18% QoQ) and $2.22 non-GAAP EPS, with Q3 guidance of $108B. CFO Colette Kress guided ~70% revenue growth in fiscal 2028, described as supply-constrained – well above the ~44% consensus. Vera Rubin shipments began in August 2026 and are expected to account for ~20% of data-center revenue in Q3, the fastest product ramp in company history. The $500B AI capital-investment partnership was announced. NVDA trades at 29x trailing earnings versus the Nasdaq-100’s 34x, with a forward multiple near 26x. So what does this mean for a portfolio manager? The data suggests that AI-driven demand is not just resilient but accelerating. However, the capex digestion could be closer than bulls assume. Hyperscalers are guiding $800B in 2026 to $1.3T in 2027 – a 62.5% increase in just 14 months. If the buildout hits a rough patch in late 2026 or early 2027, the entire AI stack could face a pause. However, the current data points suggest that the capex wave is already well underway and that the biggest beneficiaries – NVIDIA, Broadcom, and memory players – are already pricing in the bulk of that growth.
What does Broadcom’s September 2 print actually tell us?
Broadcom’s setup into the September 2 print is straightforward: Q2 FY2026 revenue of $22.2B (+48% YoY), adjusted EPS of $2.44 (+54%), with AI semiconductor revenue of $10.8B (+143%). Q3 guidance: total revenue $29.4B (+84% YoY), AI chip revenue $16B (>200% YoY), adjusted EBITDA ~$20B. The September 2 print matters because $16B in AI chip revenue implies Broadcom is capturing significant AI infrastructure spend. A beat-and-raise would signal that hyperscalers are pulling forward orders, while an in-line print would suggest that the growth is transitioning from acceleration to consolidation. The difference between a 200% YoY increase and a lower print is material for a stock that trades at a forward P/E near 19x and sits >23% below its all-time high. Key risk: Alphabet’s diversification to Marvell/MediaTek for custom AI silicon. If Alphabet shifts part of its custom AI silicon spend to competitors, Broadcom could miss its $16B AI guidance and the stock could re-rate lower. However, Broadcom still holds key contracts with other hyperscalers, and its position in the Ethernet and networking stack gives it secular tailwinds beyond just AI chips. In practice, a positive September 2 print would be a green light to add to positions. An in-line print would be a cautious wait-and-see moment. A miss would force a reassessment of the entire AI infrastructure thesis.
Is Micron the cheapest way to play the same thesis?
Micron’s late-September catalyst: consensus fiscal Q4 revenue of $50.8B (+348% YoY) and EPS >10x YoY to $31.28. P/E of 21 and forward multiple of 6 make Micron the cheapest of the major AI memory plays. HBM demand is structurally tight: Nvidia’s NVL72 holds >20TB of HBM, AMD’s Helios holds 31TB, and HBM production requires ~4x the wafer capacity of traditional memory. Top-five U.S. hyperscalers are raising combined capex from $800B this year to $1.3T by 2027, supporting memory consumption. The memory cycle has historically been more volatile than the GPU cycle, but the current data suggests that the AI-driven HBM shortage is structural, not cyclical. If you believe that Vera Rubin and NVL72 deployments will continue to outpace supply, Micron becomes a high-conviction, low-multiple play. The forward multiple of 6x is a discount to even the most pessimistic scenarios for memory pricing. The September 30 catalyst is the fiscal Q4 earnings report, which will provide the first hard data on how the memory shortage is playing out in real time. If Micron beats on both revenue and EPS, the stock could re-rate toward its historical average multiple, representing potential upside.
How does Bitcoin fit into a chip-led portfolio?
Strategy (MSTR) purchased $370M of Bitcoin on Aug 31 at an average of $80,318, resuming buys after selling roughly $430M between $59,000‑$64,000. The company holds ~4% of total Bitcoin supply. Hyperliquid’s fee-burn model used $141M of $169M Q2 revenue to repurchase and permanently destroy 4.8% of its max supply, driving an 84% price increase over 12 months. Solana’s SGP-0002 passed, doubling the rate at which new SOL issuance tapers off (long-run tokenomics improvement, H1 2027 implementation), while SGP-0003 – a transaction-fee-burn proposal – failed. Bitcoin rose 25% in August 2026, its best August since 2017. Bitcoin fits into a chip-led portfolio as a hedge against AI infrastructure concentration risk and as a speculative catalyst. The Hyperliquid model demonstrates that fee-burn mechanisms can create substantial token value accrual, and Solana’s SGP-0002 improvement suggests that tokenomics are evolving in favor of holders. A 5% portfolio cap with a 5-year hold horizon balances the speculative upside with portfolio risk. The $370M Strategy purchase at $80,318 suggests institutional interest at current levels, and the 25% August gain shows that macro tailwinds are still supportive.
What could break this trade?
1. Oil above $90/barrel on renewed Middle East tension (US strikes on Iran, Strait of Hormuz traffic down to ~5 ships/day from 130+ pre-war). 2. Analyst caution on Microsoft and other mega-cap AI spenders with $190B+ FY2026 capex run-rates. 3. University of Helsinki recession thesis tied to high corporate debt and a CAPE ratio near 41. 4. All five major oil majors have underperformed the S&P 500 since the war began despite elevated crude.
How should I actually size this?
Consider a balanced approach with heavy weighting in NVIDIA and Broadcom for quality and visibility, a smaller position in Micron for valuation and catalyst asymmetry, some Bitcoin exposure, and a cash reserve for post-earnings add-ons.
FAQ
What’s the biggest AI chip earnings catalyst in September 2026?
Broadcom’s September 2 print is the key catalyst. The $16B AI revenue guidance implies >200% YoY growth, and any beat-and-raise would signal that hyperscaler capex is accelerating beyond expectations. A miss would force a reassessment of the entire AI infrastructure thesis.
Is Micron cheaper than NVIDIA?
Yes. Micron trades at a forward multiple of 6x versus NVIDIA’s forward multiple near 26x. However, the memory cycle is historically more volatile than the GPU cycle, so the discount reflects that risk. If you believe the HBM shortage is structural, Micron’s cheapness becomes an advantage rather than a disadvantage.
Should I buy Bitcoin before or after the Fed’s next move?
The Fed meeting is scheduled for September 16, 2026. Strategy’s $370M purchase suggests institutional interest at current levels, and the 25% August gain shows that macro tailwinds are still supportive. A wait-and-see approach until after the Fed meeting could avoid potential volatility, but you might miss the upside if the Fed signals a dovish shift.