[오픈채팅방 Archive] jasonlee snippets

1. Where will the opportunities arise in Gen AI?

• So far in Gen AI, value has accrued towards bottom of the value chain (e.g. Nvidia; about 80%+ so far), as seen during cloud transition (refer to pic 1)

• As cost of compute stabilize, many expect Gen AI value stack to shift in to a V-shape, awarding those higher up the stack (application, dev tools which are more consumer-facing - refer to pic 1)

• The billion dollar question is, will there be a renaissance of individual AI applications worth billions of dollars on the backs of OpenAI, Anthropic’s API, or will it be foundation model companies gobbling everything up?

• My thoughts —> it is ok for application companies to PARTIALLY use larger model player API to power some amount of critical workflow (synthetic data generation, reasoning, etc.), but it is not enough to just build fancy UI around an API. So far, application companies that have found success (generating recurring, quality revenue) such as HeyGen, PhotoRoom, ElevenLabs, Runway, MidJourney, Perplexity, Cognition Labs no doubt leverage larger models but all own a critical component of the workflow

• The more you can build in-house to offload critical workloads to internal assets AND drive better customer value vs. off-the-shelf solutions (e.g. OpenAI API), greater moat you will be building

• Another question is, will SaaS incumbents with established distribution, be able to reap benefits from AI upsell? So far, answer is No (e.g. Notion, Adobe, Figma, etc.). Time will tell as applications begin to charge customers for AI features (e.g. ServiceNow)

2. Is AI a hype?

• Some very smart investors believe investing in foundation models + hardware layer is ridiculous.
• “Foundation models are the fastest depreciating assets in history.”

• Despite recent VC fundraising difficulties, investors have piled into AI, rewarding the sector with hefty valuation and check sizes

• AI is only 3% of total deals, 15% of total capital, but 5x the valuation and 6x the round size of non-AI companies. Excitement is real.

• Pessimists worry about the capital-intensive nature of the business, and no hard evidence to back up an ROI claim —> it is true there is zero, tangible evidence of AI showing ROI in production at scale as of right now

• Question over ROI is central to the debate over whether AI is a bubble or not

• Napkin Math (refer to pic 2): Assuming 25% ROIC on $2.4T Net PP&E (investment over 2 years) equals $600B; You need $1.8T in return ($1.2T D&A + $600B ROIC). That equates to getting rid of 5% of payroll in OpEx, or 3% increase in rev at 50% EBIT margin. Personally, I think this is doable.

• Another evidence point is comparing earnings multiple of Nvidia with Cisco at internet hype

• Cisco saw its earnings multiple skyrocket from 37x to 132x at its internet bubble peak, while Nvidia’s 5 year average earnings multiple is 40x, and is currently trading at 39x

3. Where should we invest?

Platform Plays

Country-specific Plays

Modality / Vertical Specific Plays

New Architectures?