Jeremie Ontiveros of SemiAnalysis: Obtaining GPUs is Easier than Getting Power
Jeremie Ontiveros, head of data center and energy research at semiconductor research firm SemiAnalysis, stated that there is currently a certain degree of chip oversupply in the market, making it easier to obtain GPUs than to secure power for data centers.
He pointed out that predicting data center needs is relatively straightforward, as it can be tracked in advance through infrastructure signals, while the chip supply chain is more volatile.
Looking ahead to 2026 and 2027, the market is showing strong energy responsiveness.
Ontiveros warned that by 2028, supply chain risks could reverse, leading to potential oversupply risks for data centers.
Currently, the industry is more concerned about an oversupply of data services than chip oversupply, and the pace of expansion in the chip supply chain will be a key indicator to monitor in the future.
This expansion is a strong positive signal for capital expenditure, including investments in new semiconductor fabs and purchasing new equipment.
Driven by chip oversupply and power constraints, capital is flowing towards energy and data center infrastructure, benefiting power suppliers and related support, while pure chip suppliers and data service oversupply segments are under pressure.
Source: Public Information
ABAB AI Insight
SemiAnalysis has previously tracked data centers and chip supply chains, releasing multiple predictive reports based on infrastructure signals, forming a research path that anticipates supply-demand inflection points using public signals, including cross-analysis of GPU and power bottlenecks.
Ontiveros has shifted his focus from chip availability to power constraints, warning of potential oversupply in data centers by 2028, motivated by the traceability of infrastructure signals to reduce forecasting errors, while viewing investments in fabs and equipment as positive indicators of capital expenditure.
Similar cases include early warnings of telecom capacity oversupply in the 1990s and current tracking by multiple institutions of AI computing power and energy matching; the industry is currently transitioning from chip shortages to a rebalancing of power and data center supply and demand.
This essentially represents a restructuring of the supply chain, with constraints shifting from GPU supply to power and data center capacity, where the mechanism is that infrastructure signals can be observed in advance while chip fluctuations are greater, shifting pricing power from chip manufacturers to energy and data center holders.
ABAB News · Cognitive Law
- Power bottlenecks precede chip oversupply
- Traceable signals reduce forecasting errors
- Expansion speed determines capital expenditure direction