-
Oracle ( ORCL ) has a backlog of $523B but shares fell 23.25% YTD as capex hit $20.54B. Nvidia ( NVDA ) posted 4Q revenue of $68.13B, up 73% YoY. Microsoft ( MSFT ) spent $29.88B in capex with $625B in RPO.
-
OpenAI pulled out of Oracle’s Stargate data center development because power infrastructure delays mean Nvidia’s current Blackwell chips will be outdated by the time the facility opens, with next-generation Vera Rubin available instead.
-
An analyst named NVIDIA just named his top 10 AI stocks in 2010. Get it for free here.
“The chip era is moving so fast that even the biggest infrastructure deals can’t keep up.” This line from CNBC’s Derrida Bossa captures the real tension now plaguing AI infrastructure investors. OpenAI reportedly backed away from expanding its flagship Stargate data center in Abilene, Texas, because the power won’t be ready for at least a year, and by then, OpenAI doesn’t want current-generation Blackwell chips because NVIDIA’s next-generation Vera Rubin will be available. The practical result: Oracle gave the loan, secured the site, ordered the hardware, and the customer said the chips would be dated before the building was ready.
For investors, this is the moment where the financial concept called capex-to-earnings dedication becomes very real. Understanding this is the difference between seeing Oracle’s $523 billion backlog as a treasure chest and seeing it as a liability waiting to be tested.
CNBC’s framework is correct, and investors who reject it are taking on more risk than they realize. Oracle, meanwhile, has made a huge bet on infrastructure that’s faster than the chip-era power grid and can keep up with construction schedules. In Q1 of fiscal 2026, Oracle’s capital expenditures reached just $8.5 billion, consuming more than 100% of cash flow and pushing free cash flow to negative $362 million. In Q2, capex for the first half of fiscal 2026 totaled $20.54 billion. It’s real cash that leaves the building to build infrastructure that will eventually have to be filled by paying customers running current generation hardware.
READ: The analyst named NVIDIA in 2010 Just naming his top 10 AI stocks
The financial mechanics at work here are straightforward. When a company signs a cloud infrastructure contract, it books a residual performance obligation (RPO), which is a revenue commitment that has not yet been recognized because the service has not been delivered. Oracle’s RPO rose 438% to $523 billion in Q2 of fiscal 2026. This number seems unusual. But Oracle’s quarterly revenue is about $16 billion, meaning the company is sitting on years’ worth of contract work. The question is whether these contracts hold up when the hardware it was built around goes back a generation.
Microsoft is also spending aggressively. Microsoft’s financial expenses are expected to reach $29.88 billion in fiscal 2026, nearly doubling year-over-year. But Microsoft has an important buffer that Oracle lacks. Microsoft’s quarterly revenue is $81.27 billion, giving it about five times its earnings base to absorb capex shocks. If a hyperscale customer delays or directs a contract, Microsoft can absorb it. Oracle can’t afford it. It doesn’t have the cushion that other megacaps have.
NVIDIA’s financial results illustrate why infrastructure developers are rushing to stay current. NVIDIA reported FY2026 Q4 revenue of $68.13 billion, a 73% year-over-year increase – a growth rate that reflects the chip market moving faster than any other infrastructure maker. The data center segment was the engine of this growth, with network revenue increasing 263% year over year as hyperscaler races to build AI capacity. NVIDIA reported FY2026 Q4 revenue of $68.13 billion, up 73% year-over-year, with data center networking revenue up just 263% YoY. CEO Jensen Huang described the moment clearly: “Grace Blackwell with NVLink is the king of speculation today, offering the lowest price command per token, and Vera Rubin will extend that leadership even further.” The message to infrastructure developers is clear: the window for any chip generation is shrinking.
The implication for infrastructure developers is that a data center designed around Blackwell GPUs may be commercially inferior when it opens. The electrical infrastructure, which is a binding constraint in Abilene, takes 12 to 18 months to prepare. Chip generations are changing faster than that. OpenAI’s decision to move capacity to new sites where Vera Rubin can be employed from day one is a reasonable response to this dissonance. For Oracle, this means that a site that has attracted billions of promises may need to be renegotiated or leased.
Oracle has a total delivery commitment of $95.2 billion, which reflects how committed the company is to the hardware cycle. This number is not behind revenue. It’s an upfront obligation to pay for chips and infrastructure regardless of whether the end customer shows up on schedule.
Stargate’s situation reveals a structural divide in AI infrastructure investment. Companies with diversified revenue streams and large cash flows can treat delaying or renegotiating contracts as a mistake. Companies whose growth thesis depends on a small number of anchor customers attracting large new capacity are in a completely different situation.
An investor who bought Oracle early in 2026 is sitting on a 23.25% year-over-year decline, with shares trading around $149.77. This decline reflects the market price of exactly what Bosa described as a risk: if the anchor tenant moves capacity to new sites with new chips, the revenue recognition timeline is stretched even further while capex obligations remain constant.
Despite the selloff, analysts maintain a consensus price target of $257.29 with 32 buy ratings, suggesting that the long-term thesis remains. Bell’s case assumes that Oracle successfully replenishes capacity and maintains its existing contracts — a meaningful implementation hurdle given the chip cycle dynamics now in play.
Microsoft, by contrast, is down 15.56% year-to-date but has a $625 billion business RPO spread across Azure, Microsoft 365, and dozens of enterprise products. Revising a single infrastructure contract doesn’t move the needle for Microsoft like it does for Oracle.
Stargate’s situation creates a clear framework for distinguishing AI infrastructure companies by their ability to absorb the timing gaps between capex commitments and revenue recognition.
Specifically for Oracle, the key metrics to track are not the RPO headline number, which would seem impressive, but the rate at which RPO turns into recognized revenue. IaaS revenue rose 68% year-over-year to $4.08 billion in Q2 of fiscal 2026, which is a real-time signal of whether promising deals are actually panning out. If growth slows while capex remains high, pressure on free cash flow will intensify.
NVIDIA reads this situation differently. OpenAI moving away from Blackwell-era capacity to wait for Vera Rubin means that demand for next-generation chips is moving forward. NVIDIA has guided Q1 FY2027 revenue to around $78 billion, and Vera Rubin’s tenure has yet to meaningfully contribute to that number. The risk for NVIDIA investors is different: The stock has returned 3.49% year-to-date despite earnings strength, reflecting broader market caution rather than any fundamental deterioration.
The single most important thing to understand from this story is that in this fast-moving chip era, the companies most at risk of implementing infrastructure are those whose revenue base is too small to absorb the time difference between when the deal is signed and when the hardware is ready to be deployed. Oracle builds the right infrastructure for the right market. Can it do that fast, and with the right chips at the right time, is a question the market is now pricing in.
Wall Street is pouring billions into AI, but many investors are buying the wrong stocks. The analyst who first identified NVIDIA as a buyback in 2010 — before its 28,000% run — has identified just 10 new AI companies that he believes can deliver returns beyond that point. One dominates the $100 billion equipment market. Bill addresses the single biggest obstacle to maintaining AI data centers. The third segment is a net play in the optical network market that is quadrupling. Most investors haven’t heard of half of these names. Get a free list of all 10 stocks here.
