Iren’s 1.4 GW data-center switch-on in April and Broadcom’s 74% AI-chip surge create a one-two punch that sidesteps Nvidia’s valuation ceiling while still capturing the fastest-growing slices of the AI stack.
The Setup: Power, Not GPUs, Is the New Bottleneck
Wall Street still prices AI upside as if Nvidia’s H100s are the only story. Reality is shifting underfoot: hyperscalers now throttle deployments not by GPU availability but by megawatt capacity. Iren’s 3 GW pipeline sits inside that bottleneck, while Broadcom designs the domain-specific silicon that replaces Nvidia’s general-purpose GPUs inside those same walls.
Iren’s April Catalyst: 1.4 GW That Could Triple Contract Value
The Sweetwater 1 site flips online in April, adding 1.4 GW of grid-tied power. Management’s base case—$3.4 billion annual recurring revenue by the end of 2026—assumes only 16% utilization of secured power. Simple math: every extra 200 MW slice can replicate the $1.94 billion Microsoft deal. Seven identical wins fully utilize Sweetwater and lift ARR above $16 billion, a 4× upside scenario the market has not modeled.
- Five-year, $9.7 billion Microsoft master lease already locked (averages $1.94 B ARR).
- Remaining 1.2 GW at Sweetwater equals six additional Microsoft-scale contracts.
- Stock trades at 2.3× 2026E sales under the conservative $3.4 B guide—cheap if contracts accelerate.
Broadcom’s Hidden Leverage: Custom Silicon Margins Approaching 50%
Fiscal Q4 numbers show AI revenue up 74% year-over-year and guidance for Q1 implies a 100% plus run-rate. Net margin hit 46% last quarter on the back of custom AI chips that sell for 2–3× Nvidia’s average selling price because they bake in customer IP. Alphabet, Meta and ByteDance have all disclosed in-house accelerator programs that rely on Broadcom’s design libraries. Each new 3-nm tape-out carries a $250–$400 million non-recurring engineering fee; once in production, Broadcom collects a 60% gross margin royalty stream for five to seven years.
Valuation Cross-Check: Where the Market Still Yawns
Nvidia’s 28× forward sales leaves zero room for hiccups. Iren, even under the bull-case $16 B ARR scenario, trades at 0.9× sales. Broadcom, at 12× earnings, is priced like a mature semiconductor play despite AI revenue that could double again in 2026. Spread between the two names offers a barbell: Iren for torque, Broadcom for cash-flow ballast.
Risk Stack: Execution, Regulation, and Demand Migration
Texas grid stability and EPA permitting could delay Iren’s 2026 energization schedule by three to six months, pushing first revenue into 2027. For Broadcom, customer concentration is rising—top three hyperscalers now represent 38% of AI sales. Any architectural pivot to internally designed silicon (rumored at Amazon) would shave 5–7% from consensus revenue. Hedge: own both; their risk profiles are negatively correlated—power delays hurt Iren but don’t impact Broadcom, while silicon transitions hit Broadcom yet leave Iren’s colocation demand intact.
Positioning: How the Desk Is Playing It
Portfolio weighting 60% Broadcom / 40% Iren captures margin-rich cash flow while preserving upside optionality on power contracts. Use Broadcom’s 1.9% dividend to fund Iren’s volatility. Rebalance trigger: if Iren secures two more gigawatt-scale leases before July, equal-weight the pair and let the thesis ride through 2027.
Stay ahead of the AI infrastructure curve—bookmark onlytrustedinfo.com for the fastest, most definitive analysis on every earnings release, permitting docket, and tape-out that moves these names.
