Key Takeaways:
Andrew Graham says the market is misreading the AI trade — the problem isn't demand fatigue but a widening performance gap between hyperscalers and the semiconductor companies that supply them.
The AI investment narrative has shifted from euphoria to skepticism over the past year, with investors questioning whether the billions flowing into data centers will ever generate adequate returns. Andrew Graham, managing partner at a hedge fund focused on technology, argues the concern is misplaced. The issue, he said, is not that AI demand is fading but that the market has failed to distinguish between companies that own the infrastructure and those that merely rent it.
"Investors are conflating a performance gap with AI fatigue," Graham said. "The hyperscalers that control their own compute and the chip companies that design the most advanced silicon are pulling away from everyone else. GOOGL and AVGO are the strongest expressions of that divergence."
Graham's thesis centers on two structural advantages. Alphabet Inc. (GOOGL) operates one of the world's largest private cloud networks and designs its own tensor processing units through Google Cloud's TPU program, giving it control over both the hardware and the software layer. Broadcom Inc. (AVGO), meanwhile, has become the dominant supplier of custom AI chips for hyperscalers through its ASIC (application-specific integrated circuit) business, with reported design wins at Google, Meta and ByteDance.
The performance gap Graham describes is visible in the numbers. Broadcom's AI-related revenue reached $12.2 billion in fiscal 2025, more than triple the prior year, according to company filings. The company's custom chip division, which designs processors for specific hyperscaler workloads, has become the fastest-growing segment in its semiconductor business. Alphabet's Google Cloud revenue rose 28% year over year to $12.6 billion in the first quarter of 2026, with operating margins expanding to 19.2% from 14% a year earlier.
The divergence extends beyond financial performance. Nvidia Corp. (NVDA) remains the dominant supplier of general-purpose AI accelerators, with an estimated 80% share of the data center GPU market. But hyperscalers are increasingly designing their own chips to reduce dependence on Nvidia's premium pricing and to optimize for specific inference workloads. Broadcom's custom chip business has emerged as the primary beneficiary of that trend, with analysts at Morgan Stanley estimating the company's AI-addressable market at $60 billion by 2028.
"The market is treating all AI exposure as equal, and that's a mistake," Graham said. "The companies that control the silicon and the cloud infrastructure have pricing power and margin visibility. The ones that just consume compute don't."
The implications for investors are straightforward. Alphabet trades at 22 times forward earnings, a discount to its five-year average of 26 times, according to Bloomberg data. Broadcom trades at 28 times forward earnings, below its peak of 35 times in mid-2025. Both valuations reflect what Graham calls a "conglomerate discount" — the market pricing their AI businesses at the same multiple as their legacy operations.
"If the market starts separating AI revenue from legacy revenue in its valuation models, both stocks could re-rate significantly," Graham said. "That's the opportunity the consensus is missing."
The counterargument is that hyperscaler capital expenditure, which totaled $230 billion across the five largest cloud providers in 2025, must eventually translate into revenue growth for the end customers. If enterprise AI adoption slows or if the return on invested capital fails to materialize, the entire thesis breaks. Graham acknowledges the risk but argues the structural winners — companies that own both the chip design and the cloud platform — are best positioned to weather any slowdown.
"The next catalyst is the second half of 2026, when Google Cloud's TPU v6 ramp and Broadcom's next-generation ASIC production both hit volume," Graham said. "That's when the performance gap becomes visible in the financial statements, not just the product roadmaps."
This article is for informational purposes only and does not constitute investment advice.
