Theses
The questions the AI economy turns on.
Each thesis is one question about the AI economy, argued on both sides by named voices, with dated tests that settle it in public. They are grouped by where the money moves along the AI value chain.
9 theses
Semiconductors & Compute Silicon
Who designs, makes and equips the logic chips that do AI compute.
Does the concentration of leading-edge chipmaking give TSMC pricing power over its customers?
TSMC's revenue for September 2026 was NT$511.9 billion, 54.6% above September 2025. The tests read its monthly revenue as the Taiwan Stock Exchange publishes it.
Next test: Nov 10, 2026 · 8 tests Call · 72% yes
Will a panel-level or substrate-less advanced packaging line (CoPoS, CoWoP, or OSAT 310x310mm panel) reach announced mass production for an AI accelerator product by the end of 2027?
The CoWoS shortfall forces panelization and substrate-less relief paths into real mass production by end-2027, moving part of the AI package bottleneck out of TSMC's wafer-based lines.
Next test: Nov 12, 2026 · 6 tests
Will HBM supply, rather than advanced packaging capacity, be named by Nvidia or TSMC as the primary accelerator supply constraint in any 2027 earnings call?
The binding constraint migrates from CoWoS interposer capacity to HBM stacking and DRAM supply, so the named bottleneck in 2027 is memory rather than packaging.
Next test: Nov 3, 2026 · 6 tests Call · 72% yes
Will a mainland Chinese foundry other than SMIC publicly confirm volume production at a 7nm-class (N+2) or better logic node before 2027-12-31?
Process learning diffuses from SMIC to Hua Hong/HLMC and other domestic fabs, widening the choke point so that entity-list controls aimed at single firms stop determining outcomes.
Next test: Oct 28, 2026 · 7 tests
Will large software firms' gross margins drop a point or more by Q3 2027 while Nvidia's stays at or above 70%?
The idea: AI inference costs are climbing faster than software revenue, squeezing software profitability while chipmakers keep their fat margins, moving profit upstream to silicon.
Next test: Nov 18, 2026 · 6 tests Call · 84% yes
Will the PHLX Semiconductor Index (SOX) total return exceed the iShares Expanded Tech-Software ETF (IGV) total return over the period 2026-01-01 to 2027-09-30?
Bottleneck rent at the silicon layer keeps semiconductor equity returns ahead of software equity returns through the AI capital cycle.
Next test: Nov 12, 2026 · 6 tests Call · 72% yes
3 theses
Memory & Storage
The memory and storage that feed the accelerators and hold the data.
Will memory makers keep earning scarcity profits from AI demand through 2027?
Micron's revenue for the quarter ended September 3, 2026 was $54.2 billion, against $11.3 billion a year earlier. The tests read Micron's next quarter and Korea's chip exports, most of which are memory.
Next test: Oct 29, 2026 · 8 tests Call · 68% yes
Will Korea's monthly memory semiconductor exports set a new all-time monthly record in any month between 2026-11-15 and 2027-12-31?
Inference workloads are read by some as memory-bound rather than compute-bound, pulling HBM, DRAM and CXL demand toward record levels, while this side tests whether Korean export data actually registers a new all-time monthly high.
Next test: Oct 27, 2026 · 8 tests Call · 76% yes
Will DRAM contract or spot prices fall at least 30% from their 2026 peak at any point before December 31, 2027, per a dated public price series?
Memory scarcity pricing reverses into a glut, with DRAM prices falling 30%+ from peak as greenfield capacity and AI efficiency gains close the gap.
Next test: Oct 29, 2026 · 7 tests Call · 80% yes
11 theses
Hyperscalers, Cloud & Compute Services
Who rents compute and runs models for others, and spends the capex.
Will NVIDIA state by Nov 30, 2027 that inference is over 50% of its data center compute demand?
As AI shifts from building models to running them in everyday products and agents, inference calls could outgrow training runs. The question is whether NVIDIA says so on the record.
Next test: Nov 9, 2026 · 6 tests Call · 72% yes
Will spot/on-demand hourly rental prices for NVIDIA H100 80GB SXM capacity on major public GPU marketplaces be at or above their November 2026 level as of the stated test dates in 2027?
Deployed accelerator capacity stays scarce enough that even prior-generation H100/A100 rental prices hold or rise rather than collapsing as Blackwell-class supply floods in, confirming there is no idle-GPU glut.
Next test: Nov 9, 2026 · 6 tests Call · 72% yes
Will three or more distinct operators publicly offer surplus internal AI compute for external resale (or disclose underutilized AI capacity) between 2026-11-15 and 2027-12-31?
The no-glut claim breaks down: firms that over-ordered ahead of site readiness or lost their internal AI programs begin marketing surplus accelerators externally, revealing idle capacity.
Next test: Nov 9, 2026 · 6 tests Call · 72% yes
Will cloud revenue growth at Microsoft, Google, and AWS outpace their combined depreciation growth near Q3 2027?
Record spending on AI data centers is pushing up depreciation costs. The question is whether cloud revenue is growing even faster.
Next test: Nov 9, 2026 · 6 tests Call · 72% yes
Will Alphabet, Microsoft and Amazon each post positive free cash flow for calendar 2027?
Cloud and AI spending is surging, but so is capex. This tracks whether all three hyperscalers still generate more cash than they spend over the trailing year through 2027.
Next test: Nov 9, 2026 · 6 tests
Will D&A as a share of revenue rise year-over-year for at least 3 of 4 big hyperscalers in Q3 2027 vs Q3 2026?
Years of heavy AI infrastructure spending are moving from construction into active use, and depreciation is catching up with that spending curve.
Next test: Nov 9, 2026 · 6 tests
Will any of Microsoft, Alphabet, Amazon, or Meta disclose a further extension of the estimated useful life of servers or network equipment in a 10-K or 10-Q filed between 2026-11-15 and 2027-12-31?
At least one major hyperscaler again extends server useful lives, offsetting depreciation-driven margin compression through accounting choice rather than economics.
Next test: Oct 28, 2026 · 6 tests Call · 66% yes
Will a listed AI neocloud report a GPU impairment or depreciation charge of $100M+ by end of 2027?
GPU leasing deals assume chips hold value over their contract term, but faster hardware cycles could force write-downs. This page tracks whether a major US-listed neocloud discloses a large impairment or accelerated depreciation tied to aging GPUs.
Next test: Nov 9, 2026 · 6 tests Call · 78% yes
Will the average trailing-twelve-month ROIC of Microsoft, Alphabet, Amazon and Meta fall by the quarter nearest Q3 2027 versus the quarter nearest Q4 2024?
Massive AI capital spending is piling into balance sheets and depreciation schedules faster than AI revenue shows up, which could drag down returns on invested capital across the biggest cloud players.
Next test: Oct 28, 2026 · 6 tests Call · 72% yes
Will Oracle report, in any fiscal quarter reported between 2026-11-15 and 2027-12-31, a year-over-year decline in total gross margin percentage of 300 basis points or more alongside a capex guidance increase?
Oracle, the leveraged builder of AI capacity against concentrated counterparties, will show the telecom-builder signature: gross margin compressing through the utilization floor while capex guidance rises and RPO converts slowly.
Next test: Nov 9, 2026 · 6 tests Call · 70% yes
Will combined disclosed AI-attributed annualized revenue across the major US AI providers reach $300bn by the end of 2027 as reported in public filings or earnings disclosures?
AI revenue scales fast enough that the revenue-to-depreciation ratio on AI infrastructure rises toward the level where returns clear an 8-10% cost of capital, contradicting the value-destruction case that argues capex outruns the revenue needed to earn it back.
Next test: Nov 9, 2026 · 7 tests Call · 66% yes
2 theses
AI Financing, Credit & Capital Structure
Who lends to, invests in and insures the buildout, and on what terms.
Will at least 5 US-listed IPOs of AI-linked companies each raise $500 million or more between Jan 2026 and Sept 2027?
Public markets could reopen as a major funding source for the AI buildout, producing a wave of large listings instead of companies staying private.
Next test: Nov 23, 2026 · 6 tests Call · 66% yes
Will a publicly reported payment default, covenant breach, restructuring, or rating downgrade to speculative-grade occur on a GPU-collateralized loan or securitization of at least $500 million before 2027-12-31?
The asset-life versus debt-tenor mismatch produces at least one public distress event on a large GPU-backed facility within the window.
Next test: Nov 24, 2026 · 6 tests
No thesis page yet in Networking, Interconnect & Optics, Data Centers & Physical Infrastructure, Power, Energy & Grid, Materials, Industrial Supply Chain & Construction, Foundation Models & AI Intelligence, AI Software, Applications & Agents, Physical AI, Robotics & Autonomy, AI Security, Trust & Governance, AI Adoption, Productivity & Industry Transformation.
A thesis page states one question and both sides of it; DeepStack takes no side in words, and its view is the number it gives each test before the date. Analysis and opinion, not investment advice. Voices are paraphrased from public writing and attributed by name; DeepStack shows no score, ranking or accuracy for any person. The Call’s number is the chance, from 0 to 100, that the Call gives the yes answer to a test’s question, recorded before the test’s date; it is DeepStack’s own reading, not a market price and not advice. The readers’ mark on the same line is the readers’ median, the middle of the numbers signed-in readers gave the test, shown from 25 readers.