Why UBS remains bullish on the hyperscaler stocks despite capex concerns

UBS said the major cloud providers are benefiting from unprecedented AI demand, with combined 1Q26 revenue up 39% to $84.8 billion and sequential revenue gains of $6.5 billion, while it expects this to rise to $8.5 billion in 2Q26. Google Cloud led growth at 63% and Microsoft, Amazon, Oracle and CoreWeave now have a combined $2.10 trillion backlog, up 184% year over year. UBS also projected 2026 capex of $673.14 billion, up 76%, underscoring continued aggressive AI infrastructure spending.

Analysis

The key message is not that AI demand is strong, but that the demand curve is still outrunning the industry’s ability to monetize it cleanly. The multi-year capex wave is now so large that the next leg of upside shifts from model improvement to infrastructure bottlenecks, pricing power, and backlog conversion — which favors the hyperscalers with the best internal demand capture and the least execution friction. That creates a barbell where MSFT remains the most “quality” AI infrastructure owner, while AMZN has the most room for margin surprise if Trainium and Bedrock continue to substitute for third-party inference spend.

The second-order winner is Oracle, because it is the cleanest beneficiary of externalized AI capacity demand without the same degree of internal cannibalization risk. The market still treats Oracle as a laggard cloud name, but the mix of committed backlog and AI-specific commitments gives it a longer-duration re-rating setup than the consensus expects; the risk is not demand, it is whether delivery ramps fast enough to convert backlog into revenue without margin leakage. By contrast, Google’s AI monetization looks underappreciated because the TPU externalization story creates an entirely new revenue layer, but this also raises the probability of a near-term capex shock that can keep reported free cash flow under pressure for several quarters.

The main risk to the group is that the market begins discounting capex saturation before revenue recognition catches up. If enterprise buyers keep front-loading commitments, the narrative can stay bullish for months even if near-term FCF deteriorates; if booking growth slows, these names could de-rate quickly because current multiples already embed sustained hypergrowth. The contrarian read is that the best relative long is not necessarily the largest cloud by revenue, but the one with the cleanest mix shift into AI-specific spend and the highest optionality on monetization of backlog — that argues for being selective rather than buying the entire Mag 7 AI basket.

Key Decisions for Investors

• Long MSFT vs. QQQ for 3-6 months: MSFT has the cleanest enterprise AI distribution advantage and should trade with lower downside volatility than the broader mega-cap complex if AI spend remains durable.
• Long AMZN vs. MSFT into the next 1-2 quarters: if Bedrock and Trainium continue to take share, AMZN has greater operating leverage to AI infrastructure monetization and a better chance of upside surprise relative to current expectations.
• Initiate a medium-term long ORCL position on pullbacks: the risk/reward is favorable because backlog conversion can drive estimate revisions faster than the market is currently discounting; stop if delivery guidance slips or margin expansion stalls.
• Consider a pair trade long GOOGL / short a basket of high-multiple software names over 6-9 months: TPU externalization and AI infra monetization are underappreciated relative to software names that depend on slower seat-based AI adoption.
• Avoid chasing CRWV upside here; use it as a tactical trading vehicle only. The name benefits from the capex wave, but with lower quality visibility and higher financing sensitivity, it is better expressed as a momentum trade than a core hold.