Sep 28, 2026 · 6 min

AI’s infrastructure bottleneck threatens to widen the value gap

Signal over Noise: The real-world constraints to scaling AI

AI’s technical progress is accelerating, but compute, power and business adoption may decide whether that progress becomes economic value or investment returns.

3 key takeaways

1AI capabilities have advanced rapidly since ChatGPT, yet economic value capture has lagged behind technical progress.
2Compute and power availability create real-world constraints that could determine which companies capture AI’s eventual value.
3The pace of business adoption may matter as much as model performance in translating AI advances into investment returns.

Don't miss

Ulrike Hoffman-Burchiadi reframes the AI investment question around compute, power and adoption constraints rather than capability alone.

The brief

Ulrike Hoffman-Burchiadi opens by framing a widening gap: AI capabilities have accelerated since ChatGPT, but economic value and investment returns have not kept pace.

The episode argues that real-world friction—not just model performance—helps explain why AI’s technical progress has yet to translate proportionally into economic value capture.

Compute and power availability emerge as central constraints, potentially shaping which companies can scale AI and ultimately capture its value.

Business adoption is another bottleneck: the pace at which organizations implement AI may determine when capability gains become measurable economic returns.

The episode closes with UBS’s standard disclaimer, emphasizing that its views are informational and not tailored to any recipient’s investment objectives or circumstances.

What was said on this episode

16 statements · 8 positive · 6 negative · 1 mixed · 1 neutral

Ulrike Hoffman-Burchiadion Artificial IntelligenceNegative0:21
AI capability growth has outpaced economic value capture.
“AI capabilities have surged, but value capture has not kept pace.”
Listen at 0:21
Ulrike Hoffman-Burchiadion AI capabilitiesPositive0:33
AI capabilities are improving at a super-exponential rate.
“AI capabilities are improving at a super exponential rate”
Listen at 0:33
Ulrike Hoffman-Burchiadion AI agentsPositive0:53
OpenClaw and Muse demonstrate that AI-agent adoption can scale quickly.
“The launch of OpenClaw in November of last year and the launch of Muse earlier this month are two data points that show how quickly. the adoption of AI agents can scale.”
Listen at 0:53
Ulrike Hoffman-Burchiadion OpenClawPositive1:05
OpenClaw gained 100,000 GitHub stars within three months.
“OpenClaw was the first open source project in history that gained 100,000 stars on software platform GitHub in just three months.”
Listen at 1:05
Ulrike Hoffman-Burchiadion MusePositive1:16
Muse reached 1.43 million US iOS downloads in its first 12 days.
“Muse was downloaded 1.43 million times on iOS in the US in the first 12 days, beating early mobile benchmarks for ChatGPT.”
Listen at 1:16
Ulrike Hoffman-Burchiadion AI demandPositive1:28
OpenClaw and Muse adoption signals support strong AI demand.
“All of this bodes well for AI demand.”
Listen at 1:28
Ulrike Hoffman-Burchiadion AI infrastructure and enterprise adoptionNegative1:35
AI infrastructure and enterprise adoption are progressing more slowly than capabilities.
“physical infrastructure to supply AI and enterprise adoption have been progressing at a more linear pace.”
Listen at 1:35
Ulrike Hoffman-Burchiadion Project JupiterNeutral2:39
Project Jupiter does not yet indicate broader data-center construction delays.
“At this stage, Project Jupiter appears to be an isolated case, not evidence of broader delays in data center build-outs.”
Listen at 2:39
Ulrike Hoffman-Burchiadion AI models and infrastructureNegative2:57
AI models can improve faster than physical infrastructure can scale.
“AI models can improve faster than pipelines, permits, and power systems can scale.”
Listen at 2:57
Ulrike Hoffman-Burchiadion AI infrastructure delivery riskNegative3:03
AI infrastructure delivery risk now warrants contractual protections.
“Delivery risk. has become a significant enough factor to require contractual protections.”
Listen at 3:03
Ulrike Hoffman-Burchiadion AI labsPositive3:24
AI labs could prioritize high-ROI inference and improve hardware efficiency.
“AI labs could direct scarce compute capacity towards the highest ROI inference use cases and encourage more efficient use of existing hardware.”
Listen at 3:24
Ulrike Hoffman-Burchiadion Cloud providersMixed3:36
Limited grid access could push cloud providers toward off-grid power.
“limited grid access could drive cloud providers to its off-grid power, creating opportunities for alternative suppliers.”
Listen at 3:36
Ulrike Hoffman-Burchiadion Artificial intelligence deploymentNegative3:45
Manufacturing constraints and developer shortages could slow AI deployment.
“Manufacturing constraints and a limited pool of proven developers, though, could slow deployment.”
Listen at 3:45
Ulrike Hoffman-Burchiadion AI infrastructure policyPositive3:53
Targeted policies could ease AI infrastructure bottlenecks.
“targeted policy measures could ease bottlenecks by streamlining permitting and grid connections, while support grid expansion and modernization.”
Listen at 3:53
Ulrike Hoffman-Burchiadion AI demandPositive4:06
The Jupiter incident indicates that AI demand remains strong.
“We read the Jupiter incidence as evidence that AI demand remains strong”
Listen at 4:06
Ulrike Hoffman-Burchiadion AI value chain bottlenecksNegative4:27
A single bottleneck can create excess capacity across the rest of the value chain.
“A single bottleneck can create a glut across the whole rest of the value chain. sets the overall pace.”
Listen at 4:27

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.