We research the AI buildout end-to-end — from nuclear fuel and grid equipment through wafers, optics, hyperscale compute, models, and agentic execution — and we translate architectural reality into unit economics, gross-margin trajectory, and long-duration valuation.
The AI buildout is not a single trade. It is a chain — and value migrates along it as bottlenecks shift. Segments 01 to 08 are the infrastructure supply chain, and the Bottleneck Map traces each of them down to individual nodes. Segments 09 to 16 sit downstream, from data software to consumer applications. They depend on that supply chain, so their Bottleneck Map link opens the map's summary. Rule of Value ranks the stocks in all sixteen.
01–08 · Bottleneck Map and Rule of Value
Chip architectures, IP, and foundry services powering AI training and inference workloads.
High-bandwidth memory, server DRAM, and memory interface / CXL controllers.
CoWoS and advanced packaging services, IC substrates, and packaging materials.
Data center switching, optical interconnects, cables, and SerDes.
Grid equipment, on-site generation, power distribution, and alternative energy.
Liquid cooling infrastructure, advanced cooling solutions, and services.
PCBs, laminates, copper foil, and passive components.
Server ODM, enterprise storage, and AI cloud/neocloud services.
09–16 · Bottleneck Map summary and Rule of Value
Databases and data platforms, observability, and data protection.
AI and agentic platforms, engagement and commerce platforms, and activity-indexed SaaS.
Network, endpoint, and identity security.
Seat-licensed HCM, productivity, and vertical SaaS sit on the wrong side of agentic decoupling — disrupted by the names in 10. Watchlist available in Coverage.
Autonomous vehicles, surgical and industrial robotics, defense autonomy, and the sensors and chips behind them.
Satellite constellations, launch and space infrastructure, and orbital and lunar compute.
E-commerce, ride-sharing and delivery, streaming and social, travel, and consumer fintech.
Large-cap defense contractors and other companies outside segments 01 to 15.
| Rank | Company | VAR-Resistance |
|---|---|---|
| 1 | ||
| 2 | ||
| 3 | ||
| 4 | ||
| 5 |
Preview — names and scores are locked. The full live ranking is available to Convequity members.
Most research is finance-first — and misses the architectural moats — or tech-first, and waves away unit economics. We do both, and we translate one into the other.
"I started Convequity because I kept seeing the same disconnect — Wall Street pricing companies off trailing metrics while the real story lived in the architecture, the market dynamics, and the founder's roadmap. We go deeper into the technical stack and wider into the structural forces to understand where growth and profitability are actually heading. Then we marry that to the financials with rigor. The asymmetry between what the market sees and what the technology enables — that's our entire edge."
We dissect tech stacks layer by layer — wafer chemistry, packaging, optics, custom silicon, data planes, cloud abstractions. Architectural bets compound across capex cycles and dictate where margin actually accrues.
Architectural choices and market structure (top-down GTM vs. bottom-up adoption, seat-based vs. consumption) determine entry barriers, gross-margin trajectory, and operating leverage. We model it explicitly — full assumptions, scenarios, and sensitivities, in transparent Google Sheets clients can stress-test themselves.
Wall Street's short-termism dismisses vision as intangible, focusing instead on minor details for quarterly models. Yet, a founder's vision is fundamental—it drives the product, shapes the culture, and guides the long-term roadmap. We explicitly weight this, backing visionaries who have a proven track record of executing their ambition.
The top currently open long positions by unrealized return, mapped to the AI Value Chain. Returns are unrealized, since position open. Subscribers get detailed quarterly rebalancing reports with holdings theses. Live portfolio verifiable on request.
Customers are redesigning around tight chips, memory, optics and power. We increased Cerebras to 1% and SPCX to 4%, and we trimmed Soitec to 3.5% and POET to 0.75%.
As of week ending 2026.05.01
Returns shown are unrealized P&L on the top open long positions by return from a 70+ name technology portfolio co-managed by Convequity, expressed as percentage gain from cost basis at position open through the most recent available portfolio mark. Losing positions exist alongside winners, and the full portfolio (open and closed) is available to qualified subscribers on request. Past performance is not indicative of future results. This is not investment advice or a solicitation.
A new fund — separate from the live portfolio above — running the same long-biased AI value-chain strategy and research process. We are currently raising capital from HNW and institutional allocators. Minimums: $250,000 for individuals; $1,000,000 for institutional entities.
No clicks, no dropdowns. The full coverage list, organized by AI Value Chain segment. The disrupted-incumbent watchlist is included as names to avoid, short-bias, or exit on rallies.
Coverage list is illustrative and updates as the chain evolves. Other tickers are actively being added to our coverage. Full report archive available to subscribers.
Shorter, denser AI systems could change which optical parts customers need. In the final part of our 4Q26 rebalance, we add ams OSRAM, prioritize Veeco and Aeva, and take smaller, speculative positions in MicroVision and Quantum Computing Inc., funded partly by trimming correlated exposure.
Huawei’s new phone chip is more than a sanctions workaround: it points to another way to improve chips when shrinking transistors gets harder. The bigger test is AI, where the same ideas could cut data-movement costs and make Huawei’s NPUs a serious challenge to NVIDIA.
Customers are redesigning around tight chips, memory, optics and power. We increased Cerebras to 1% and SPCX to 4%, and we trimmed Soitec to 3.5% and POET to 0.75%.
Nvidia trades at its lowest multiple since 2015. We compare our view with the Market’s and the Street’s, and explain why a shrinking moat makes Nvidia a possible value trap and Broadcom, at similar multiples, the stronger case.
Single-thesis arguments, framework updates, and cross-segment observations. Published when the thinking warrants it — not on a schedule.
Request access
Nvidia trades at its lowest multiple since 2015. We compare our view with the Market’s and the Street’s, and explain why a shrinking moat makes Nvidia a possible value trap and Broadcom, at similar multiples, the stronger case.
Cut off from EUV, Huawei is scaling a different way: folding two chips face to face so signals travel down rather than across. Part 2 explains why that raises clock speed without shrinking transistors, how the design repairs its own defects, and where the real wall lies.
Cerebras can generate tokens at extraordinary speed for a handful of users, but NVIDIA's GPU architecture delivers superior economics at scale. This report explains why the fastest AI inference system does not necessarily win at hyperscale.
SPCX is now the ultimate AI data-center and energy play. At ~$2.4T and ~80x forward sales, its valuation rests almost entirely on Space Data Center execution — where Starship, solar, and custom silicon could unlock a path that terrestrial buildouts cannot match.
Two ways to engage. The research itself, or the research plus the analysts behind it.
Full access to premium content for professional investors.
Premium Research, plus the analysts behind it. Direct dialogue, custom work, and bespoke value-chain mandates scoped to your portfolio.
Subscriptions and engagements are offered to qualified institutional investors and serious allocators only. All research is independent equity research and is not personalised investment advice or a solicitation. Multi-year and custom-scope terms available on request.
