The AI Capital Absorption Test: When the Compute Boom Must Prove Itself
AI may be one of the most important technologies in modern history.
That does not mean every dollar being spent to build it will earn an acceptable return.
Big Tech is now committing roughly $700B+ annually to infrastructure across Amazon, Microsoft, Alphabet and Meta.
Yet the operating evidence still looks remarkably strong:
Microsoft commercial RPO: $678B
Google Cloud growth: +82%
AWS growth: +37%
Nvidia Data Center: +117%
AMD Data Center: +107%
Broadcom AI semiconductor revenue: +221%
So is AI already overbuilt?
Our conclusion is:
Probably not yet.
The stronger risk appears later.
The huge 2025–2027 infrastructure wave still has to enter service.
Then comes the real test:
Can AI generate enough gross profit before the hardware economically ages?
Our new Decentralised News research introduces the:
AI Capital Absorption Test 2027–2028
We analyze:
• hyperscaler CapEx
• cloud backlogs
• GPU demand
• data-center utilization
• inference-price deflation
• depreciation
• free cash flow
• power constraints
• interest rates
• agentic AI workload growth
And introduce several new DN concepts:
AI Capital Absorption Ratio
Does the gross profit generated by new AI capacity exceed its depreciation, capital cost and operating burden?
Compute Duration Mismatch
Will the hardware remain economically competitive long enough to earn back its investment?
Inference Absorption Elasticity
Can AI workload growth outrun the rapid decline in the price of intelligence?
Depreciation Wall
When the enormous CapEx wave begins showing up at scale as operating expense.
Four-Gate Overbuild Test
We do not call a systemic AI overbuild until multiple conditions deteriorate together:
Capacity
Monetization
Capital burden
Macro/credit
The most important hypothesis:
2026 still looks primarily like scarcity and installation.
2027–2028 may be the real Capital Absorption Window.
That is when today's infrastructure has to start proving its economics.
And agentic AI may decide the outcome.
Persistent agents consume vastly more inference than one-shot chatbot queries.
If agent workloads multiply faster than inference prices fall, the infrastructure may be absorbed.
If price compression wins, AI usage could explode while infrastructure returns disappoint.
The wrong question is:
“Is AI a bubble?”
The better question is:
“Can the machines earn back the capital before they economically age?”
Read the complete insights and use the free AI Capital Absorption Stress Engine: https://decentralised.news/ai-capital-absorption-test-2027-2028
#ArtificialIntelligence #AI #AIBubble #DataCenters #Nvidia #Microsoft #Amazon #Google #Meta #AgenticAI #Macro #Investing #Technology #DecentralisedNews