Technology thesis · Semiconductors & Chips
high conviction growthGPU acceleration
GPU acceleration is the compute foundation of the AI revolution; Nvidia's CUDA ecosystem creates a software moat as durable as the hardware advantage.
Position maintained continuously · last reviewed May 7, 2026
The thesis
Core thesis
NVIDIA controls roughly 90% of merchant AI training GPUs and the bulk of inference. The moat is not just hardware — it is CUDA, the software ecosystem NVIDIA put at ~6 million registered developers at GTC 2026. AMD's MI355X and MI450/Helios are technically competitive, but the CUDA switching cost remains the decisive lock-in. Demand has stayed supply-constrained through the Blackwell ramp, sustaining NVIDIA pricing power; the durable question is whether ROCm and hyperscaler custom silicon erode the ecosystem advantage, not the silicon lead.
State of the art (2026)
As of mid-2026, GPU acceleration is mid-generational-transition. NVIDIA shipped record Q1-FY27 data-centre revenue of $75.2B (up 92% year-on-year) on Blackwell, which Jensen Huang calls sold out. At GTC 2026 NVIDIA unveiled Vera Rubin (336B-transistor R100, HBM4); it entered full production by GTC Taipei in June 2026 with the first rack live at Microsoft Azure and hyperscaler availability in H2 2026. AMD is the only credible hardware challenger: MI350/MI355X shipped from mid-2025, and the MI400-series Helios rack is on track for H2 2026, anchored by Oracle deploying 50,000 MI450 GPUs from Q3 2026 plus OpenAI commitments. The decisive moat remains CUDA software, not silicon; ROCm has not yet closed it.
Everything below is live inside CanaryIQ
The full analysis behind the verdict — the structure is real; the content unlocks when you log in.
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what — and who depends on whom
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers — each with a remove-by condition
Decision frameworks
The same call, framed for your desk
Thesis changelog
When our view changed, and why
Change our mind
3 disconfirming conditions
The rest is inside
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and the changelog of every time our call on GPU acceleration has changed — all live inside CanaryIQ.