Greater China Technology Hardware: How Long and How Much Will AI Profit Beat?

Core Conclusion

AI server gross margins rose from ~12% in early 2023 to 17% by 2H25 and still have room to run, driven by higher-margin GB300/B300 platforms and growing in-house component content. This margin expansion—not shipment volume—will dominate 2027 profit dynamics, delivering an estimated 33% year-on-year AI earnings growth even if revenue growth moderates. Consensus forecasts, anchored to a fading top-line narrative, have yet to fully reflect the durability of this profit cycle.

What the Market Is Underpricing

The market fixates on the peak of AI revenue growth, missing that margins are now the primary earnings driver. Because analyst models extrapolate past revenue trajectories, they lag the structural margin uplift. As AI server margins continue to shift from 17% toward higher levels on next-generation racks, the earnings revision cycle can persist well into 2027, extending the window for positive surprises.

Evidence Chain

Supply-chain checks trace AI server gross margins from ~12% in 1H23 to 17% in 2H25. The GB300/B300 platform amplifies this trend: higher average selling prices, more complex designs, and a greater share of proprietary components compress incremental costs relative to revenue. These attributes allow margin expansion even without unit growth acceleration.

Direct beneficiaries—Foxconn Industrial Internet, Gigabyte, and Quanta—possess product mixes weighted toward these high-margin configurations. Their 2027 consensus earnings still embed conservative margin assumptions, implying upward revision potential. The 33% profit growth estimate is derived from supplier guidance and ASP analysis, with margin improvement contributing the majority of the uplift.

Key Risks

Non-AI segments (PCs, conventional servers) may take longer to recover, partly offsetting AI-driven gains and introducing earnings volatility. Hyperscaler capex cuts or order pushouts would reset valuation multiples. Escalating trade barriers and tariffs add input-cost uncertainty, while component bottlenecks could cap the pace of margin accretion. If AI margins plateau near 17% rather than rising further, the 33% profit scenario becomes harder to reach.

Valuation and Trade Implications

Overweight the names with the most margin expansion headroom: Foxconn Industrial Internet, Gigabyte, and Quanta. Their current valuations do not fully price in the compounding effect of higher-margin AI revenue. Hold Wistron and Inventec, where AI margin benefits are largely discounted and non-AI headwinds constrain earnings elasticity. The trade is not about AI revenue momentum—it is about positioning for an underappreciated margin super-cycle.