Oracle FQ4 26 – B+

Oracle gets a provisional B+ for AI Infrastructure

• The market has been spooked by how much Oracle is spending on AI infrastructure, but I think that it may be doing better than most of its peers when it comes to earning an economic return on the money it is investing.
• FQ4 26 revenues / Adj-EPS were $19.2bn / $2.11, ahead of expectations of $19.1bn / $1.96, as AI and cloud migration caused growth in the cloud business to accelerate to 47% YoY.
• Unfortunately, to support the revenue growth, Oracle is going to spend more money than ever, and this is what made the market nervous, sending the shares down 10.1% in after-hours trading.
• Oracle said that it would invest $70bn this fiscal year, up from $55.7 in FY 26, which will be paid for through equity issuance and debt.
• The increase in spending went hand in hand with no change to guidance, which left the market wondering where the extra money is going.
• FQ1 27 guidance was broadly in line with YoY revenue growth of 27% – 29% and 57% – 63% cloud revenue growth, with Adj-EPS expected at $1.71 – $1.75, which is ahead of expectations of $1.69 but broadly in line with revenue expectations of $19.1bn.
• The debt market is pondering the same question as the credit default swap on Oracle’s debt (the cost to insure oneself against a default) has increased again to 198bp or 1.98%.
• For reference, in September 2025, this cost was 43bp or 0.43%, which I have long seen as a sign that the business model of compute is questionable.
• This is a major red flag as the bond market is usually a very good indicator of financial distress (e.g. Lehman Brothers).
• Against this, Oracle says that it has remaining performance obligations (firm orders) of $638bn, up $85bn from FQ3 26.
• The real issue is not the size of the order book but whether Oracle will be able to make a good return from those orders, meaning that the economics of compute is going to determine whether the management is right or the debt market is right.
• Oracle has not disclosed any good data in this regard, but a back-of-an-envelope calculation leads me to give Oracle a B+, which is better than most.
• In the last 12 months, Oracle added 1.2GW of capacity and its cloud revenues grew by $9.5bn.
• If I assume a steady build throughout the year, then the average incremental capacity that was online during the year would be 0.6GW.
• This gives me a revenue per GW of $15.8bn, which is way better than OpenAI and the other neo-clouds are managing to produce.
• Assuming the capacity was Grace Blackwell, this would give me a 5-year IRR of 18%, which increases if Oracle can keep the capacity running for more than 5 years.
• This is pretty good, and above the cost of capital for Oracle, representing a good deal for equity shareholders even with more shares going to be issued.
• However, this calculation needs verification as it ignores the degree to which GPUs have been purchased by the client and then given to Oracle to run, which could affect the numbers materially.
• It is also based on far from perfect estimates of capacity and the revenues that the capacity is able to generate.
• However, it is a sign that a more detailed analysis is warranted because if Oracle is going to post good returns from its AI investments, then there could be upside to its medium-term EPS estimates.
• Ignoring this, the company is on 22x FY2027 PER and 16.5x FY2028 PER, which is pretty reasonable considering Oracle is forecasting 18% YoY growth in underlying EPS in FY2027 and consensus has 35% growth in FY2028.
• I think it might be time to have a closer look at Oracle.