Interview
Eric Vishria: Where is the Value in AI - Chips, Models or Apps? | E1206
- Foundational models are predicted to be the fastest depreciating asset in history, yet their development is expected to drive a major shift in AI potentially larger than previous transformations, with activity levels anticipated to remain higher than the 2010–2011 period and exceed 2024 activity.
- NVIDIA is not expected to remain the sole dominant player in the AI infrastructure layer, while the foundational model war is forecast to benefit consumers by advancing the state of the art despite Benchmark holding no such investments.
- Investment evaluation criteria are shifting away from spreadsheet-driven unit economics, which are viewed as offering false precision, toward assessing whether an entrepreneur is extraordinary, possesses unique insights, and operates in a market capable of sustaining a large company.
- Future compensation models for ventures may discard market norms to secure critical talent, and investors are advised to prioritize founders who constantly learn and apply new mental models rather than those with all answers upfront.
- Career investors are projected to improve as investors due to accumulated data and mental models from 15 years and hundreds of pitches, but may become less effective board members due to a lack of empathy and an over-reliance on deterministic outcome views.
- The speaker plans to allocate 80% to 85% of time to portfolio companies given the high conviction nature of the firm, while retaining flexibility to write checks of $50 million or $150 million regardless of fund size constraints to seize exceptional opportunities.
- Internal investment voting relies on a "one to ten" scale where a score of four or below discourages an deal, though investments may proceed without unanimity if the sponsoring partner has strong backing and firm trust exists.
- Partners are expected to bring specific superpowers to the firm, including Bill Gurley's focus on fundamentals, Peter Thiel's unmatched insights into people and motivations, and Matt Kohler's ability to assess the depth of entrepreneurial insight.
- AI monetization is expected to evolve from $20 monthly subscriptions toward bundled API models, with products perceived as "magic" driving demand while companies balance higher per-seat pricing against potential margin denigration from implementation costs.
- In the AI medical scribe space, Microsoft and Nuance are predicted to leverage bundled packaging to dominate the market, while incumbents are expected to be highly active in monitoring their positions rather than ignoring the threat.
- Vertical SaaS companies using commoditized foundation models may face an innovator's dilemma requiring them to "burn down" existing businesses to succeed, with product quality potentially secondary to the distribution moats held by established players.
- If AI captures a significant portion of the $200,000 value attributed to a software engineer, it could generate giant outcomes and massive market value creation, justifying high valuations despite current tension as the prize is considered large enough to support fundamental pricing.
- Large acquisitions, potentially reaching $100 billion and exceeding previous $30 billion precedents, are viewed as possible due to current capital scales, though antitrust regulations remain a potential hurdle for such transactions.
- Venture success is not deterministic as variables evolve, making early certainty impossible; success probability increases significantly when learning entrepreneurs rethink their navigation during platform shifts.
- Time management is expected to invert with children, making the work week easier but weekends harder and leaving the speaker exhausted by Monday morning, alongside a reflection on a missed early acquisition decision regarding Rockmelt.
- Benchmark partners are expected to keep the speaker out of bad investments while helping identify opportunities they might miss, with the firm operating on high trust where a partner can proceed with an investment even without a unanimous vote.
- The current period is described as the most exciting and disorienting time in a 25-year technology career, with the firm expected to be more active in 2024 and potentially participate in the field to capture extraordinary opportunities in the current AI shift.