Based on Blackwell Ultra, GR00T N1, Dynamo, and NVQLink announced at NVIDIA GTC 2025 keynote, we analyze the structural growth and investment opportunities in 5 industries: AI accelerators, data center infrastructure, Physical AI/robotics, autonomous driving, and quantum computing.
The GTC 2025 keynote was not just a product announcement, but a blueprint for the cross-industry flywheel built by NVIDIA. CEO Jensen Huang reaffirmed the annual GPU roadmap leading to Blackwell Ultra (H2 2025), Vera Rubin (H2 2026, 336B transistors, HBM4), and Feynman (2028), and unveiled GR00T N1 (humanoid robot foundation model), Dynamo (30x inference acceleration), NVQLink (quantum-GPU hybrid), and Halos (autonomous driving safety platform). FACT
The combined TAM of the 5 key industries reaches over $1.5 trillion by 2030, all of which are in a phase of structural growth. In particular, with hyperscalers committing to invest $660-690B in AI infrastructure in 2026 alone, AI accelerators and data center infrastructure offer high-confidence investment opportunities based on cash flow. FACT
A strong Positive Bias was recorded with a signal weighted sum score of +45/100, and NVDA's Forward P/E of 25.2x is at a 53% discount compared to the sector average of 53.2x. However, the surge in oil prices ($103-106/bbl) due to the Iran war (2/28~), concerns about CPI reacceleration, and uncertainties in semiconductor export controls are acting as short-term volatility factors. INFERENCE
The US economic fundamentals are still in an expansion phase (ISM Manufacturing PMI 52.4, stable unemployment rate), but the outbreak of the Iran war (2/28) has caused Brent oil prices to surge by 50%+ from $70 to $103-106, and the S&P 500 has entered a -12% correction from its peak. A large-scale rotation to energy, industrials, and defensive stocks is underway, and technology stocks are virtually flat or negative. FACT
Key Judgment: The AI infrastructure CapEx cycle is structural and will not be꺾by short-term macro headwinds. There is no change in the hyperscaler $660B+ investment plan, and rather, AI competition is so fierce that they are continuing to invest even by mobilizing debt. However, the triple burden of energy costs, inflation, and interest rates will put downward pressure on profitability and valuation. INFERENCE
| Indicator | Value | Direction | GTC Industry Impact | Source |
|---|---|---|---|---|
| Fed Funds Rate | 3.50-3.75% | HOLD | Continued freeze → AI CapEx financing costs can be managed | Bloomberg, 2026-03 |
| CPI YoY | 2.4% | STABLE | Before reflecting oil prices — possibility of breaking through 3%+ in March CPI | BLS, 2026-03-11 |
| ISM Mfg PMI | 52.4 | EXPANDING | Manufacturing expansion → supports demand for robotics and industrial AI | ISM, 2026-03-02 |
| 10Y Treasury | 4.28% | RISING | Rising long-term interest rates → burden on growth stock valuations | FRED, 2026-03-13 |
| DXY | 99.8 | WEAK | Weak dollar → advantageous for non-US companies such as TSMC/Samsung | TradingEconomics, 2026-03 |
| Brent Crude | $106/bbl | SURGING | Risk of soaring data center electricity costs (30→90GW needed) | OilPrice, 2026-03-17 |
| China PMI | 52.1 | EXPANDING | China's AI/robotics demand remains strong | Caixin, 2026-03 |
| 10Y-2Y Spread | +55bp | NORMAL | Inversion resolved → no signal of imminent recession | FRED, 2026-03-13 |
The AI accelerator market is growing at $120.2B (2025) → $375B (2030), CAGR 23.6%, entering the Mid Cycle. NVIDIA dominates 86% market share based on AI accelerator sales, and the CUDA ecosystem (4.7M+ developers, 18,658 patents) is the key lock-in factor. FACT
Blackwell backlog of 3.6M units is sold out by mid-2026, and CoWoS packaging is fully operational with TSMC allocating 60% to NVIDIA. The supply-constrained situation where demand greatly exceeds supply continues. FACT
Competition Risk: Hyperscaler custom chips (Google TPU, AWS Trainium, Meta MTIA) are expected to expand to 45% of the market by 2028, but this is mainly in the inference area, and training still has a high dependence on NVIDIA. INFERENCE
Data center infrastructure market is $236B (2025) → $934B (2030), CAGR 24-32%. In the Early-to-Mid Cycle, hyperscalers invest $700B in 2026 alone. The key bottleneck is power — transformer lead time 18-36 months, grid interconnection wait 3-5 years. FACT
CPO (Co-Packaged Optics) enters full volume ramp in 2026, and it is a transition period in networking architecture where Ethernet overtakes InfiniBand. Power infrastructure (nuclear SMR, gas turbines), cooling (liquid cooling), and networking (CPO, 800G/1.6T) form a structural growth story with a longer lifespan than the GPU cycle. NARRATIVE
The humanoid robot market shows the highest growth rate among the 5 industries at $4.9B (2025) → $165B (2034), CAGR 40-50%. GR00T N1, unveiled at GTC 2025, is the world's first open humanoid foundation model and is the core of NVIDIA's strategy to build the 'CUDA of Robots'. FACT
China dominates ~60% of the market based on units (AgiBot 31%, Unitree 27%), and Unitree R1 ($5,900) is a signal of price destruction. Tesla Optimus 100K production target (2026) and BYD 20K deployment plan are catalysts for industrial scaling. FACT
The autonomous driving market is $203.5B (2025) → $2.2T (2036), CAGR ~25%. 2026 is called the 'Year of Autonomous Driving', with Waymo recording over 150,000 rides per week and expanding to 20 cities. NVIDIA's automotive sales are expected to grow approximately 3x from $1.7B → $5B. FACT
Drive AGX Hyperion 10 and Halos safety platform announced at GTC are establishing themselves as the industry standard AV compute, and Uber aims to install 100,000 units by 2027. Regulatory approval is still the biggest variable. INFERENCE
The quantum computing market is $3.5B (202