Top of the Morning: The AI Economy 4Q26 update - Part 1

Paul Hsiao · UBS Chief Investment Office, Global Wealth Management

AI capex is supply-driven and far less sensitive to higher interest rates than in past cycles.

Rates have risen at both the short and long end in 2026, which normally restricts economic activity by raising borrowing costs and required returns. Paul argues this cycle differs because AI spending is supply-driven, so much less of it depends on interest rates. That is why hyperscalers can announce hundreds of billions of dollars of capex toward $1 trillion a year despite higher rates.

“The difference between this cycle and last cycle with all everything talking about AI is that it's really supply driven. So much less of it is dependent on interest rates than before.”at 3:03

Paul Hsiao · UBS Chief Investment Office, Global Wealth Management

Hyperscaler capex to be revised sharply higher: about $700-800B in 2026, rising toward $1T a year by 2028-29.

Market consensus for 2026 hyperscaler capex was about $500-600B, but year-to-date spending is tracking a $700-800B pace. Paul expects spending to accelerate to about $1 trillion per year in 2028 and 2029. He frames this as a supply-driven build-out that announced plans already point to, so he expects upward revisions to continue.

“hyperscaler CapEx intention will revise significantly higher this year... year to date it's really looking more on track to a 700 to $800,000,000,000 pace in 2026 and then going into 2028 and 2029 we expect that to accelerate to 1,000,000,000,000 per year.”at 3:27

Paul Hsiao · UBS Chief Investment Office, Global Wealth Management

Hyperscaler free cash flow turns negative through 2027, forcing reliance on equity and bond markets to fund AI build-out.

Paul projects total free cash flow for some major hyperscalers to dip into negative territory in 2026, stay negative in 2027, and turn positive only in 2028. Cash on hand and free cash flow were expected to fund the data center build-out, but they can no longer cover rising investment needs, so hyperscalers must tap equity or bond markets. The key risk he flags is that if long rates stay high, the interest embedded in hyperscaler bond issuance rises, making funding harder for some companies through 2027 and 2028.

“we do see free cash flow in total for some of the major hyperscalers dipping into negative territory this year, staying in negative territory in 2027 before going back into positive territory in 2028. So what that means is they can't just rely on the free cash flow to finance more and more investment needs.”at 4:23

Paul Hsiao · UBS Chief Investment Office, Global Wealth Management

Semis were overcrowded; money rotated in Q3 into hyperscalers and software as a safer AI trade.

Semiconductors were already a crowded trade going into 2026. Power stocks sold off in August after regulatory concerns, such as Texas pausing data center demand. Investors then rotated into hyperscalers, which Paul sees as a safe haven within AI given their revenues and high cash balances, and into software, which had been unlocked by equity markets early in the year. Paul describes this as a shift from a crowded part of the AI trade to a more defensive part rather than a fade of the AI trade.

“investors still remain concentrated in AI, but we did see a pretty significant rotation from semis into hyperscalers and software... semis was already a crowded trade”at 5:38

Paul Hsiao · UBS Chief Investment Office, Global Wealth Management

Credit spreads stay range-bound and AI-linked bond demand robust; no repricing of hyperscaler credit risk.

Bond spreads remained range-bound in Q3, and demand for AI-linked debt stayed strong. Bond investors have not repriced risk, so the market has not changed its view on hyperscaler creditworthiness from Q2 to Q3. Paul treats this as the credit backdrop to watch as hyperscalers keep issuing debt.

“bond spreads remain [range] bound and demand from AI linked paper remains robust. So we're not seeing that sort of repricing of risk... The bond market hasn't changed its opinions on the credibility of these hyperscalers in Q3 compared to Q2.”at 6:54

Paul Hsiao · UBS Chief Investment Office, Global Wealth Management

Cybersecurity is underpriced given AI-enabled attacks on large institutions and the need for defenses.

Over the summer, concerns emerged that models from OpenAI and Google's Gemini could exploit vulnerabilities at large institutions. Paul argues that this security risk is underpriced in the market. He points to two questions for 2027: whether a more robust federal AI framework will address security, and whether cybersecurity companies can design defenses that other companies can use to mitigate AI-driven exposure.

“I think the underpriced in the market really is security... I think over the summer, we've heard some concerns that OpenAI and Gemini, we've had some of these models being able to take advantage of some vulnerabilities in some large institutions.”at 8:52