According to the latest minutes from UBS's Global Tech and AI Conference, industry leaders believe the AI sector is in the second year of a decade-long supercycle, with demand described as "insatiable and continuously growing." The primary challenge lies in supply-side bottlenecks rather than insufficient demand.
The UBS conference minutes, released on December 4, highlight that for investors, this signals robust growth in AI is far from over. The entire supply chain—from chip design and cloud infrastructure to data center construction—is entering a multi-year boom. Leading players unanimously dismiss concerns of a "bubble," emphasizing their focus on fulfilling long-term, locked-in orders.
Key participants like Advanced Micro Devices (AMD), CoreWeave, and Nebius reinforced this outlook, showcasing staggering growth projections, multi-billion-dollar partnerships, and unwavering confidence in sustained demand expansion.
Emerging cloud providers (Neocloud) and infrastructure experts noted that AI demand is surging due to expanding model sizes, widespread post-training applications, and compute-intensive inference workloads. Data center connectivity visibility has extended from one quarter to 1–3 years, underscoring long-term industry planning. Traditional server demand is also accelerating with the rise of AI agent workloads. Market consensus indicates the real challenge is addressing supply constraints—from power and labor to advanced components—to deliver computing power faster.
**AMD: "Year Two of a Decade-Long Supercycle"** AMD CEO Lisa Su expressed strong optimism about the AI cycle, stating the company views this as "year two of a ten-year computing supercycle," not a bubble.
- **Ambitious Targets**: AMD raised its CAGR forecast from 50%+ to 60%+, targeting a $1 trillion total addressable market (TAM) by 2030 and over 10% market share. Its server CPU revenue share has grown to 40%. - **OpenAI Validation**: A 6GW multi-generation partnership with OpenAI—equivalent to "hundreds of billions in sales"—validates AMD’s competitiveness and attracts other AI-native clients. - **Full-Stack Solutions**: Through acquisitions like ZT, AMD is building system-level capabilities, with its MI450 rack solution launching in 2026. - **Supply Confidence**: Deep collaborations with TSMC and memory/package partners ensure AMD can meet demand.
**CoreWeave: "Insatiable" Demand, Delivery the Challenge** The AI cloud provider described demand as "insatiable and relentless."
- **Drivers**: Model scaling, post-training apps, and inference workloads fuel growth. CoreWeave’s backlog totals $55 billion. - **No Overbuilding**: Contracts reflect five-year commitments, with clients renewing H100 GPU clusters at near-original prices, proving compute’s enduring value. - **Nvidia Dominance**: Demand remains "overwhelmingly" for Nvidia GPUs, though alternatives may be considered if needed. - **Financing**: Asset-backed loans and 60% investment-grade clients bolster financial stability.
**Nebius: AI Hyperscaler, Not Just GPU Leaser** The Yandex spin-off aims to be a full-stack AI hyperscaler.
- **Explosive Growth**: Some clients double demand every 6–8 weeks; Q3 2025 pipeline grew 70% YoY, adding $4 billion in new business. - **Microsoft Deal**: A $19 billion, five-year agreement supports Copilot and new models, not competing with Azure. - **Profit Path**: Targets 20–30% EBIT margins via cost savings (e.g., in-house racks), premium services, and operational leverage. Hopper GPUs use a conservative 4-year depreciation. - **Scalability**: $8.5 billion in funding enables parallel data center projects to address delivery bottlenecks.
Comments