Signal Over Noise: Amid neocloud noise, AI demand remains the signal

AI infrastructure & data centers

• Ulrike Hoffmann-Burchardi · UBS Global Wealth Management

  Higher rates raise financing risk for leveraged neocloud providers, possibly forcing restructuring or bankruptcy despite strong demand.

  Neoclouds spend billions upfront on AI chips, data centers, power and networking, often before the associated revenue is realized. They then try to recover the investment through multi-year customer contracts. Much of this expansion is debt-financed, including floating-rate debt, so higher borrowing costs hit immediately. As an illustration, a provider with 10 billion of floating-rate debt would pay about 200 million more in annual interest if rates rose two percentage points, with no added GPU or customer. Higher rates also raise hurdle rates, so projects that worked at cheap financing may no longer earn adequate returns. The result can be a mismatch: a provider with strong demand, high utilization and valuable assets may still lack cash flow after interest and capex. That is a financing problem, not a demand problem. If a provider cannot meet its obligations, its assets can be sold to another operator. This creates short-term noise while a new owner absorbs the capacity, so ownership may change while demand for the capacity remains. The speaker still calls highly leveraged infrastructure investments riskier with higher rates.

  “If the increased financing costs cannot be offset with higher compute prices, it could force highly leveraged providers into restructuring or even bankruptcy.”at 0:55

• Ulrike Hoffmann-Burchardi · UBS Global Wealth Management

  Rising AI GPU rental prices show compute demand outstrips supply, so compute providers keep pricing power.

  Rising rental prices are the best sign that demand exceeds supply. SemiAnalysis composite indices track the USD price per GPU hour to rent AI compute. One-year rental contracts for several generations of AI chips all show strong year-over-year pricing trends. H100 and A100 contracts are sharply higher year over year, and the most performant B200 instances are still at recent highs, up about 60% since last October. Cloud providers should therefore be able to raise prices when contracts come up for renewal. This pricing power means financing stress at individual neoclouds should not be treated as evidence that compute demand is weakening. Firm GPU pricing also suggests higher financing costs can be offset through higher prices on contract renewals. Even in a rate-driven distress, valuable GPU assets can be transferred to stronger balance sheets, leaving the structural semiconductor demand story intact.

  “One year rental contracts for different generations of AI chips all show strong pricing trends year over year.”at 4:03

• Ulrike Hoffmann-Burchardi · UBS Global Wealth Management

  Hyperscaler earnings next week should show continued cloud compute strength, given firm rental pricing.

  The speaker expects continued cloud compute strength to appear in upcoming hyperscaler earnings next week. The expectation follows from the firm rental pricing for AI compute, where GPU rental contracts show strong year-over-year gains and cloud providers should be able to raise prices at renewal. The call is that the strength in compute pricing will be visible in the reported results of the large cloud providers in the coming week.

  “We therefore expect continued cloud compute strength to show in upcoming hyperscaler earnings next week.”at 4:32