AI Infrastructure Is Moving From an Adoption Cycle Into a Utilization Cycle
Why agentic AI suggests we may still be underestimating hyperscaler demand, cloud growth and the duration of the data-center buildout
General research, not investment or trading advice or trade signals. Capital is at risk. Read the risks & conditions ↗
Opening the original PDF…
Original document · PDF · Scroll to read, or choose a page above.
General research, not investment or trading advice or trade signals. Capital is at risk. Read the paper’s risks and conditions before relying on the view. Holdings are those disclosed at the source date and may have changed.
The source states October 2026 without a publication day. This report is excluded from calculations requiring an exact start date.
Read this paper's risks & conditions ↓Report summary
What to watch ahead of hyperscaler earnings: agentic usage, cloud demand and whether new productive compute capacity can turn that demand into revenue and returns.
Cloud demand
The case is that useful agentic inference could grow faster than productive compute capacity. That could support cloud growth and extend the infrastructure buildout. Higher demand still needs to justify the price paid for an AI-linked asset.
The earnings checks
The paper looks at cloud demand and what new capacity earns. Over the next 12–24 months, the tests are whether incremental gross profit covers depreciation and financing costs, and whether infrastructure returns exceed the cost of capital.
Holdings at the source date
The paper discusses companies the author owns. It gives no new orders, position weights or entry dates, so the ownership disclosures cannot establish changes to the September portfolio. Read the original PDF for the full indicator discussion and company analysis.
Summary sources: p. 1, p. 4, p. 15, p. 16, p. 17, p. 18. Page references use the physical PDF page numbers.
Risks & conditions
Read any potential return alongside the paper's risks, conditions and limitations.
Useful demand growth must translate into acceptable investment returns. Depreciation, financing, power procurement, utilization, price deflation and counterparty concentration can absorb the benefit. Source: p. 4, p. 16, p. 17, p. 18.
Power-user concentration and frontier-lab financing create demand and counterparty risk. Capex can compress free cash flow, require additional funding or dilution, and increase sensitivity to long-duration yields. Source: p. 5, p. 12, p. 16, p. 17.
Private-lab revenue estimates are unaudited alternative-data estimates. Consumer panels with different definitions must not be mechanically combined; the vulnerability chart is not a causal AI-demand series. Source: p. 4, p. 6, p. 8, p. 9.
The paper contains no dated trade, allocation or price target. Its ownership table is research context and cannot establish an October portfolio transaction. Source: p. 15.
Source record
The cover states October 2026. No exact publication day or time is stated. Source: p. 1.
Original filename: Stoic_Research (We have underestimated everything).pdf
Imported:
PDF SHA-256: b30490961bc87995be11d790663830606a6f6c647a841c77e3ad29d45b6ee085
Current assessment
No later review published.
This is the view at the paper’s source date, subject to its original conditions. It has not been reassessed here for today’s market.
Thesis history
AI Infrastructure Is Moving From an Adoption Cycle Into a Utilization Cycle
What to watch ahead of hyperscaler earnings: agentic usage, cloud demand and whether new productive compute capacity can turn that demand into revenue and returns.
Adding the paper to this library does not change its source date. Later reviews use their actual dates and link to the supporting research. Neither records a portfolio trade.
Equities research timeline
Published equities reports appear here in date order. A change in this thesis is recorded in a separate, dated assessment with the evidence behind it.
The week uncertainty clears
The case for a tactical Nasdaq bottom during the 27–31 July earnings and macro window. Friday’s close is the confirmation test; the date alone is not enough.
Framework · Conditional bottomAI Compute to 2030: Scarcity Migrates, It Does Not Disappear
Where AI compute supply could get stuck next: integration, energized sites and firm power. The question is whether usable capacity produces high-margin revenue before fixed commitments overwhelm cash generation.
Framework · Selective thematic frameworkMain Full-Risk Portfolio
A full-risk, long portfolio based on earnings conversion and valuation normalization, with position weights, valuation scenarios and the conditions that could undermine the thesis.
Deep dive · Long portfolioAI Infrastructure Is Moving From an Adoption Cycle Into a Utilization Cycle
What to watch ahead of hyperscaler earnings: agentic usage, cloud demand and whether new productive compute capacity can turn that demand into revenue and returns.
Deep dive · Positive demand thesis; conditional returnsThe original PDFs and dates are unchanged. Where a source gives only a month, reports with overlapping dates cannot be put in a definite order. A new report in this category does not confirm the thesis or record a trade.
Explore equities researchConnections between the papers
These reading links point to shared drivers or later developments, with page references to both original PDFs.
Agentic intensity extends the demand question
October explicitly raises the author’s prior assumption about compute consumption as agentic work increases usage intensity. The usable-capacity and capital-return tests from August remain relevant: rising demand alone does not demonstrate adequate returns after depreciation, financing and price pressure.
AI Compute to 2030: Scarcity Migrates, It Does Not DisappearThis paper: p. 4, p. 16, p. 17, p. 18 · Linked paper: p. 27, p. 28, p. 29
Stronger demand; the cash-return test remains
September’s portfolio depends on investment converting into free cash flow per share. October adds a more positive agentic-demand view while retaining the distinction between useful demand and investment returns. It supplies no new allocation or dated trade for the September book.
Main Full-Risk PortfolioThis paper: p. 4, p. 15, p. 16, p. 17, p. 18 · Linked paper: p. 2, p. 19
A related paper can add context without confirming a forecast. These links and reviews do not record model-portfolio trades.