The shortage is the schedule: AI capacity relief is not a 2027 line item

Dave Vellante's latest Breaking Analysis argues the AI bubble will not burst soon, and his reason is not demand enthusiasm: it is that the supply chain cannot clear ^[1][2]. For anyone signing infrastructure contracts in the next two budget cycles, that is the operative forecast, because the shortages that slow deployment are the same ones that delay the moment buyers get choice ^[4].

The specific list matters. According to Vellante, "the AI supply chain remains constrained by high-bandwidth memory, advanced packaging, network fabric, power and site readiness" ^[3]. Not one of those five is software, and not one of them responds to a procurement decision made this quarter. He goes further on the consequence: the bottlenecks "delay price discovery (the point at which buyers have more choice); and they postpone the moment when the market discovers whether it has overbuilt" ^[4]. He does not expect easing before at least 2028, and possibly beyond ^[2]. Measured from the August 2026 publication of that analysis, that is at least two more annual planning cycles conducted under a supplier's terms rather than a buyer's ^[0][5].

The practical reading for operators is unglamorous. A budget that assumes falling unit costs for accelerated compute next year, or a second source appearing to discipline the first, is a budget assuming price discovery that Vellante says has not happened yet ^[4]. Constrained markets do not produce list-price erosion; they produce allocation, and allocation is negotiated on the seller's calendar.

The supply response that is visible sits at the wrong layer. Foxconn is ramping global AI server production to ride booming cloud investment ^[8], but server assembly does not appear on the constraint list ^[3]. Power does, and Tim O'Reilly's framing is the sharpest version of that point: "It may be a mistake to assume that the AI race is about who builds the best intelligence. It may turn out to be about who builds the electrical grid" ^[9]. SiliconANGLE notes Google increasingly sees the bigger opportunity in supplying AI infrastructure through its cloud business, which is booming ^[10], while SemiAnalysis contends Google has effectively conceded frontier model leadership following Demis Hassabis's change of job and Jeff Dean's departure, a call SiliconANGLE itself considers early ^[11].

Capital is behaving as if the constraint is real and durable. Databricks raised $5 billion while continuing to stay private ^[13]. Silicon Data raised a $30.5 million Series A to build an independent AI benchmark layer ^[12], which is the sort of thing that gets funded precisely when buyers cannot yet compare prices. Equity markets, meanwhile, are not reading the sector as one trade: SiliconANGLE described investor outlooks as mixed, with CoreWeave, Nebius and Supermicro outperforming and being rewarded, while Cisco, Cerebras and Applied Materials outperformed and were not ^[6][7].

What to watch: after a week's earnings breather, Nvidia and others report ^[14]. The useful signal is not the beat but the language around it, specifically whether guidance is framed by supply commitments and site readiness rather than order books. On the software side, the SaaSpocalypse still has not arrived despite AI's spread, though reports that Workday may be in play are an early sign that SaaS firms could get crunched ^[15]. Buyers should be planning for 2027 delivery slots, not 2027 discounts.