Dell’Oro Group reports challenges in assessing AI data center buildout announcements
Recent analysis of Artificial Intelligence (AI) data center announcements highlights the challenges in accurately assessing real capacity growth due to overlapping claims, inflated capacities, and power supply constraints. This understanding is relevant for data center operators, vendors, and investors planning or evaluating investments.
Market overview
AI has spurred numerous announcements regarding multi-gigawatt data center expansions and strategic collaborations. However, many of these declarations overlap or pertain to the same underlying infrastructure, complicating the assessment of genuine capacity additions in the industry.
Key findings
Multiple companies, including NVIDIA, AMD, and OpenAI, have announced capacity commitments, often involving shared investments and partnerships. Overlapping stakes and shared infrastructure lead to repeated counting of capacity in public announcements, obscuring actual built or planned expansions.
Power availability emerges as the primary bottleneck for data center expansions. Reports from regions like Santa Clara County demonstrate that even when physical facilities are ready, lack of sufficient power prevents them from becoming operational. This introduces uncertainty on when announced capacity will be usable.
Segment or supplier performance
OpenAI illustrates the complexity of capacity announcements: while it has publicly committed to deploying up to 26 gigawatts of semiconductor capacity from multiple suppliers, only 10 gigawatts are allocated to specific U.S. data center projects under its Stargate initiative. The remaining capacity remains unallocated or part of international expansions.
Microsoft, a key partner and investor in OpenAI, operates shared AI infrastructure that supports both its own and OpenAI's workloads. This further complicates attributing capacity figures solely to individual entities as investments and infrastructure usage overlap.
Technology or trend analysis
The prevalence of “braggerwatts”—bold yet aspirational capacity announcements lacking concrete power commitments—adds further distortion to industry capacity assessments. Without secured power or financing, many of these declared capacities may not materialize.
Timelines for data center buildouts are often compressed in public announcements but realistically span multiple years. Verification based on semiconductor shipments, construction progress, and power commitments is necessary to ground expectations amid optimistic declarations.
Forecast or analyst outlook
An aggregate count of all announced gigawatts exceeds industry models’ predictions of what will be built by 2029. Disentangling the announcements into unique, shovel-ready projects is crucial to understand realistic data center growth in the AI Edge Resource Allocator (ERA). This requires focusing on actual power availability and phased capacity deployment over time.
Overall, the industry faces challenges in reconciling headline capacity figures with deliverable infrastructure, with power supply remaining the key constraint on deployment timelines and usable AI compute resources.
Conclusion
Distinguishing true incremental capacity in AI data center expansions demands analysis beyond headline announcements, incorporating power availability, detailed construction progress, and overlapping stakeholder roles. For decision-makers, understanding these nuances is essential to form an accurate view of AI infrastructure growth. This Analyst Signals brief reflects a neutral, fact-based summary of the original research note.