Index Ventures’ Shah Weighs Risks as Anthropic and OpenAI Eye IPOs

Investors weigh prices for perfection

Venture capitalists have prioritised AI startups for the four years since the public release of ChatGPT, motivated by the companies' high revenue and valuations.

Phil Town notes that while AI giants have profits, the main danger for investors is the "prices for perfection".

The Market Hang panel also discussed the influence of the midterm elections on the outlook for a bear market.

Index Ventures differentiates through founder belief

Index Ventures partner Shardul Shah argues that the venture capital industry has become overly industrialised through statistical screening and pattern-matching.

His firm differentiates itself by prioritising "taste", founder belief, and adaptability.

Index Ventures has grown from three employees at its founding roughly 30 years ago to 93 employees today, a figure Shah identifies as the firm's most important metric.

The firm’s equal-partnership structure is designed to prevent the concentration of returns among a few star partners, which Shah identifies as a primary cause of succession difficulties in the industry.

Index Ventures avoids a "spray-and-pray" model, maintaining a portfolio of 30 to 35 investments per fund because the firm lacks the capacity to provide impactful support for thousands of small checks.

Investment selection and portfolio management

Shah cites the investment in Wiz as a proof case for his "people-over-themes" method, noting that Index invested before a business plan existed.

The firm’s "hall of fame" investments include Revolut, Datadog, Wiz, Adyen, Robinhood, Figma, ClickHouse, and Roblox.

The AI-era portfolio includes companies such as Flapping Airplanes, Simile, Parallel, Cartesia, and 7AI.

At Index, partner meetings consist of a 30–45 minute founder presentation followed by a 30–45 minute deliberation period.

Investment quorum requirements are scaled by check size, requiring approximately three people for a $1 million investment and six or more people for investments of $100 million or more.

To maintain mentorship, Index conducts pre-briefs and post-briefs with the broader team, as not all firm members participate in investment committees.

Index does not use a thematic overlay for AI investing, specifically noting they do not have a "26 and under screen".

Shah joined the seed board for Parag Agrawal at Parallel based on his "character and his competence" rather than his celebrity status.

Shah invested in June at Simile despite not specifically seeking a simulation company, citing the team as the "greatest army of under-23-year-olds" building a vertically integrated business.

Performance metrics and security sector outlook

The 2012 Index Ventures fund achieved an 11x DPI compared to an industry average of 1.5x, while the 2015 $780 million growth fund reached a 5.1x DPI. Shardul Shah stated that Index Ventures does not prioritise valuation sensitivity within a 25% band, arguing that in a market with one-directional capital flows, it is rational for prices to rise.

Shah noted that higher prices and larger funding rounds can act as a luxury product effect that aids in talent recruitment by signalling scarcity. In the enterprise niche, Shah observed companies achieving £5 million to £15 million in annual contracts within their first year of commercialisation over the last six months.

Shah characterised public security companies as a resilient and permanent category that is approaching a trillion dollars of value. Regarding the sale of Wiz, Shah noted that he and Gili Raanan took different approaches to the investment, but both agreed with Raanan's belief that the company sold too early.

Shah’s current security investment is 7AI, which he co-invested in with Asheem Chandna at Greylock. Shah believes 7AI has the opportunity to rewrite the foundations of security and converge budgets in a way that would not have been possible five to 10 years ago.

Defining research taste in the AI development cycle

Benaich identified Anthropic's Opus 4 as a system he particularly likes for tasks such as legal, finance, and investment work.

In an Anthropic dual-arm experiment comparing human and model research paths, human graders preferred the research direction of frontier-class models 64% of the time, compared to 10% for older model generations. Research from OpenAI on Codex usage among its AI staff found that the vast majority of tokens were used for research writing and explaining, with very few tokens used for deciding which research questions or paths to explore.

Benaich described research taste as the ability to pick the right problem, which he considers a hallmark of intelligence and a critical component for recursive self-improvement to function effectively. AI systems should be designed to use tokens and compute budgets efficiently while demonstrating intelligence, rather than just boiling the ocean. Measuring taste in science and engineering is difficult because there are various paths to reach a correct result, unlike in art or wine where taste is more observable or explainable.

The mathematics community experienced an uproar regarding the model AlphaEvolve solving open questions, which highlighted a debate over whether the process of reaching an answer is more important than the result itself. Quantifying what constitutes a good process is considered important for determining if an AI's output is a result of true taste.

Operational pressures and rising costs for AI startups

The merger and acquisition environment has expanded beyond traditional incumbents acquiring private companies to include various private-to-private transactions. Index Ventures manages its capital by avoiding optimization for time duration, opting instead to maintain diversified exposure across different vintages. The firm treats its Limited Partners as rational allocators who prioritise consistent performance.

Monthly token expenditure for companies building agent-based and artificial intelligence tools is rising as employees increase their reliance on these systems. To offset these rising costs, some companies are opting to use less frontier models. Data providers are shifting their services from supplying reinforcement learning environments for laboratories to treating entire companies as environments.

There has been a revival of "doomerism" regarding artificial intelligence as of late 2026, a shift from the mood two years prior. Redwood Research appeared on a podcast to discuss various "terrible scenarios" concerning the future of the technology. Nathan Benaich asserts that cyber risk now outweighs extinction risk and suggests that frontier laboratories may begin selling AI-powered defence systems. With 80,000 H100-equivalent GPUs currently in Europe, Benaich believes inference chips have become a more plausible source of leverage.

Market Hang panel examines tech productivity and Apple stock

Apple stock experienced a decline of 2.56% on 3 November 2026. The Market Hang panel examined how memory prices are affecting the financial performance of the company.

The panel also discussed whether artificial intelligence is being used to perform "dirty work" in office settings. They further examined whether the technology is being used to evaluate job productivity.