Priced for AI perfection - cracks are emerging

Update 27th Jan 2025

My original narrative was primarily based on Aswath Damodaran's Nvidia narrative and valuation.

The recent news from DeepSeek further supports this narrative.

Read details on the potential implications of the announcement from:

• Jeffrey Emanuel (60 mins, very in depth, worth reading)
• Ben Thompson/ Stratechery
• Casey Newton/ Platformer (also good counter points)

Key points:

• Nvidia is priced to perfection: Nvidia has established a near-monopoly in AI training and inference infrastructure, achieving gross margins exceeding 90% on high-end data center products. This dominance has led to a market valuation surpassing that of entire national stock markets.
• Emerging Competitive Threats:
  • Hardware Innovations: Companies like Cerebras and Groq are developing specialized hardware that offers significant performance improvements and cost efficiencies over Nvidia's GPUs, particularly in AI inference tasks.
  • In-House Chip Development: Tech giants such as Amazon are investing in custom silicon, aiming to reduce reliance on Nvidia by producing chips that, while potentially less advanced, offer better value by operating at lower margins.
• Software and Efficiency Challenges:
  • CUDA's Potential Decline: Nvidia's proprietary CUDA platform has been a significant competitive advantage. However, advancements in open-source software and improvements in competitors' drivers, notably from AMD, could erode this advantage.
  • Efficiency Innovations: Startups like DeepSeek are pioneering methods to train and run AI models more efficiently, potentially reducing the demand for Nvidia's high-end GPUs.
• Market Dynamics: The rapid influx of capital and talent into AI research and development is fostering an environment ripe for innovation. This surge increases the likelihood of breakthroughs that could challenge Nvidia's current market position.
• US Sanctions have reverse effect: DeepSeek were forced to innovate due to US sanctions on H100s but not H800s, demonstrating that the US/ SV approach of throwing more compute at the problem may have been a distraction.

Summary: While Nvidia currently enjoys a dominant position in the AI hardware market, cracks are emerging on all sides: a combination of emerging hardware competitors, software advancements, and market dynamics poses significant challenges to its ability to maintain its current valuation and market share.

Original Narrative (FV remains the same)

Catalysts:

• Nvidia's AI chips remain the central growth driver.
• Continued demand in auto chips presents potential, though small.
• Growth in gaming and crypto chips, though less impactful.
• Nvidia's lead in AI architecture supports rapid market expansion.

Assumptions:

• Revenue growth will decelerate but remain robust due to AI scaling.
• Margins may compress slightly due to competition but stay strong.
• Earnings could see slower growth as scaling pressures increase.

Risks:

• AI growth may plateau, challenging high expectations.
• Increasing competition and regulation in AI and chip markets could erode profitability.
• Dependent on TSMC and exposed to China risks

Valuation

Damodaran uses a discounted cash flow to arrive at a fair value of $87. I've used the same inputs from his model for Revenue and Earnings which means a future PE multiple of 18.4x in 10 years with a discount rate of 8.49%.

Revenue: $422 billion in 10 years

• Gaming/Other: $92bn
• AI: $300bn (60% market share of a $500bn market in 10 years)
• Auto: $30b

Total: $422bn

Gross Margins

• 70-75% in 2024 Q2
• 65% in 2034 due to rising costs from TSMC

Operating Margins

• 61.9% operating margin, 67.0% adjusted margin in 2024 Q2
• Projected to fall to 60% by 2034