Artificial Intelligence Research
Institutional-grade research on the AI cycle: model economics, hyperscaler capex exposure, labour displacement, and equity winners and losers. PRZC Research analyses the structural forces reshaping the AI sector.
The Analysis Bottleneck Is Over: LLMs, SIGINT, and the LoRaBug Architecture
How large language models eliminate the processing constraint that has limited signals intelligence since Bletchley Park. Introducing the LoRaBug — a novel expendable LoRa mesh collection architecture. PRZC Research open-source intelligence assessment.
Read ReportChina AI & DeepSeek: The Efficiency Shock and Its Investment Fallout
Full institutional analysis of DeepSeek V3/R1 — the $6M training claim, H800 architecture, benchmark performance vs. GPT-4o and Claude. China's AI ecosystem (Qwen, ERNIE, Kimi, Minimax). Export control efficacy. Nvidia's $600B selloff. Anthropic's defensibility. Geopolitical Taiwan risk. OVERWEIGHT/UNDERWEIGHT investment framework.
Read ReportThe Neurotypical Ceiling: Why AGI Cannot Be Reached Through Consensus-Convergent Training
A structural hypothesis: AGI is unreachable via training methodologies that converge on neurotypical consensus. Frontier models have hit the neurotypical performance ceiling — further scaling produces marginal gains. The highest human minds are structurally different processors, not faster neurotypicals. The ceiling is architectural, not computational, and the entire AGI timeline debate is asking the wrong question.
Read ReportThe Last Refuge: Optimus, Physical AI, and the End of the Jobs That Were Supposed to Be Safe
Every automation wave left physical trades standing. Humanoid robots combined with frontier AI reasoning remove that assumption simultaneously. Tesla Optimus, Figure AI, Physical Intelligence — the two components that made trades automation-proof (dexterous manipulation and spatial judgment in unstructured environments) are being solved at commercial scale. When the last refuge closes, there is no next one.
Read ReportThe Headless Enterprise: When AI Agents Replace Human Nodes
The corporate structure is already a network of humans processing information. Replacing human nodes with AI compute nodes transforms the org chart into a compute topology — a company running at infrastructure cost rather than payroll cost. The per-token pricing signal is already embedded. The $280B BPO sector is the clearest casualty: labour arbitrage eliminated before the product is deployed.
Read ReportWhen Copilot Becomes Pilot
Every major AI product is named "Copilot" for a reason — deliberate subordination framing. The context window has crossed the threshold where AI can fly without constant human correction. When AI stops assisting and starts executing, Microsoft named its own disruption — and the software-interface layer it defends is $6T in the blast radius.
Read ReportClaude OS: The End of the Personal Computer
Cowork already bypasses Microsoft Office at the interface layer. The next step is a Claude-native Linux distribution that replaces Windows and macOS entirely — not by competing on features, but by making the OS layer invisible. Combined with the OpenAI/Jony Ive device attacking Apple mobile, this is a coordinated assault on the two pillars of personal computing: a $6.6T blast radius, and we are only two integrations away.
Read ReportThe Last Inimitable Minds: Why Neurodivergent Cognition Becomes Irreplaceable as AI Approaches AGI
RLHF, Constitutional AI, and preference learning all optimize toward human consensus — which is neurotypical by definition. The closer AI gets to AGI, the more perfectly it replicates the majority cognitive mode, and the more neurodivergent thought becomes the one category AGI cannot reach by design. A structural thesis on cognitive scarcity in the post-AGI economy: the paradigm-breaking move, the anomaly fixation, the framework rupture — self-renewing forms of human thought that no training dial can reach.
Read ReportThe Anti-Bubble Thesis: How Anthropic's Mythos/Capybara Is Replacing the NASDAQ 100
Claude Mythos leaked. $20B ARR. $1T in triggered equity selloffs. PRZC Research argues the AI bubble narrative is structurally inverted — Anthropic is a displacement cycle, not a bubble, and a sector-by-sector map of which NASDAQ 100 companies face existential risk from a single private AI lab.
Read ReportThe Infrastructure Bubble: TurboQuant, $690B Capex, and the AI Hardware Reckoning
Google TurboQuant cuts inference memory 6x — the latest in a compounding efficiency trend that makes $690B in annual hyperscaler capex structurally fragile. The AI application bubble may be wrong. The infrastructure bubble is a different and more dangerous animal: 1-year chip life against 15-year debt structures, HBM overcapacity risk post-2026, and a power grid that cannot sustain the trajectory.
Read ReportState of AI: March 2026 — Comprehensive Intelligence Briefing
PRZC Research intelligence briefing covering Anthropic macOS Computer Use, NVIDIA GTC Vera Rubin ($1T orders), GPT-5.4, Llama 4, Gemini 3, DeepSeek V4, Xiaomi MiMo-V2-Pro, and the March 2026 investment watch list across AI winners and losers.
Read ReportFour AI-Era Winners: Six-Month Performance Scorecard
Six-month scorecard of PRZC's September 2025 AI picks: GOOGL +78% (strong win), TSLA +31% (mixed), TDOC -37% (thesis broken), CHGG -74% (analytical failure). Portfolio alpha +2.5pp. Disconfirmation protocol and category error framework.
Read ReportNo-AGI Thesis Update: Six-Month Scorecard and Verdict
Six-month verdict on the No-AGI thesis: AGI timeline 6/10, bubble deflation 3/10, China pragmatism 5/10. Overall score 4.7/10. Revised forward thesis with extended bubble deflation timeline and updated position guidance.
Read ReportSemrush: Adobe Acquisition Follow-Up — Thesis Superseded
Adobe acquires Semrush for $1.9B at $12/share (+74%). AI search disruption confirmed: CTR -47%, zero-click 60%. AIO pivot faster than expected. Acquisition probability lesson for bearish SaaS theses and AI-disrupted companies.
Read ReportAlphabet (GOOGL): The Commodity AI Layer — Default Winner at Scale
Strong Buy. Price $302, Target $420 (+39%). TPU cost moat ($0.39 vs $5.50/hr for competitors), Google Cloud +48% to $17.7B with $155B backlog, Apple Gemini deal ($5B), MAG7 valuation comparison at 15% discount. Five compounding structural advantages.
Read ReportAI Underperformance Report 2025
Enterprise AI deployments are underdelivering. Productivity claims are not materialising at the company level. This report maps the gap between AI hype and measurable business outcomes — and identifies the categories most at risk of multiple compression when the gap becomes undeniable.
Read ReportNo-AGI: China's Victory, America's Arrogance
While America bets $3 trillion on AGI arriving by 2030, China builds pragmatic AI infrastructure. If AGI remains science fiction, the S&P 500 faces 20-35% correction while Hang Seng absorbs just 5-10%. Analysis of how Western hubris created asymmetric risk.
Read ReportThe No-AGI Burst: The Withering Black Swan
The AGI bubble will not burst from a single event — it will wither. $3 trillion in market cap is contingent on AGI timelines that have no empirical support. This report maps where the exposure is concentrated and why the deflation will be slower, and more painful, than 2000.
Read ReportSemrush: High-Risk Bet on Reverse-Engineering AI Answers
Investment analysis of Semrush (SEMR) pivoting from SEO tools to AI optimization services. Core thesis: AI Answer Engines are proprietary and probabilistic—no stable algorithm to reverse-engineer. High existential risk as zero-click AI search erodes traditional SEO revenue.
Read ReportFour AI-Era Winners: Strategic Investment Picks for 2025 and Beyond
In September 2025, PRZC published four AI-era conviction picks with explicit price targets and thesis statements. Six months later, GOOGL returned +78%, TSLA +31%, TDOC -37%, CHGG -74%. This is the original report — read the scorecard alongside it.
Read Report