Michael Eisenberg on AI Investing, Liquidity and Venture Discipline

Aleph co-founder Michael Eisenberg tells Harry Stebbings that AI can be both transformative and surrounded by a financial gold rush. He argues that investors must distinguish durable business value from the excitement of a technology cycle.

Eisenberg compares AI with the early internet and subsequent fiber-optic boom: speculative investments can fail while infrastructure built during the cycle remains useful. He expects substantial investor losses alongside the creation of important companies and technologies. That is a forecast, not a quantified outcome.

He worries that limited partners may believe they are diversified across venture funds while those funds hold the same fashionable companies. Other risks include chasing prominent logos and paying high prices to copy a business model into markets where it may not reach the same scale.

Eisenberg attributes to investor Gavin Baker the view that foundation models depreciate quickly as newer models emerge. He points to the difficulty of keeping financial information current in a fast-moving market. The transcript’s opening auto-caption says “appreciating,” but the later discussion and context refer to depreciation.

Stebbings relays the argument that AI company valuations reflect the teams capable of building models, not just the models themselves. Eisenberg accepts the importance of talent but asks how durable an asset is when people can leave. He uses past departures and a corporate talent deal as reasons for caution about the number of independent winners.

Stebbings suggests large cloud providers may acquire smaller model companies while larger labs compete independently. Eisenberg says that is possible but focuses on where users begin their day and on underlying models accessed through APIs. His account of the eventual market structure is speculative.

He rejects treating all foundation-model companies as a uniform venture asset class. In his view, a few may generate large returns for investors while others lose money. The cost of building models and the motives of corporate partners make company-by-company judgment essential.

Aleph invests in Israel, and Eisenberg says the firm chose applied AI rather than foundation models based on its assessment of local talent and institutions. He cites earlier investments in maritime analytics company Windward and insurer Lemonade to distinguish longstanding AI applications from the recent large-language-model surge.

He does not claim to know which industries are universally best suited to AI. He sees opportunities in paper-heavy processes that can be broken into short workflows, while predicting that incumbent companies with fragmented data may struggle to adapt. He cites Lemonade, a portfolio company, as an example of a company he believes has an advantage.

Rather than entering crowded software categories such as sales or customer service tools, Eisenberg prefers areas where specialized domain knowledge makes competition harder. He says execution still matters more than competition alone and acknowledges that an unconventional investment can fail.

He describes an Aleph-backed venture applying synthetic biology and chemistry to rocket fuel, with a hoped-for cost advantage. Market education and technical execution remain significant challenges. The claimed cost is the venture thesis, not a verified commercial result, and Eisenberg has an investment interest.

Stebbings suggests that such hard technology is inherently riskier than SaaS. Eisenberg disagrees, arguing that competition and price erosion can make familiar software models risky too. He is skeptical of treating predictable-looking software revenue as a bond-like venture strategy.

Based on research he says he did at Benchmark, Eisenberg argues public companies can earn premium multiples when they are the only way for investors to participate in an emerging trend. Growth and gross profit still matter. He offers this as an investment interpretation, not a complete valuation model.

He predicts horizontal SaaS processes will face disruption from AI and says only one of Aleph’s preceding seventeen investments was horizontal SaaS. Stebbings raises a competing argument that AI could make vertical software easier to reproduce as well. Neither outcome is established in the interview.

Eisenberg repeats an unverified rumor that Amazon had told employees not to buy more software, explicitly acknowledging he does not know whether it is true. He uses it to speculate that more organizations may build customized software with AI. It should not be treated as a confirmed Amazon policy.

Stebbings counters that many companies lack the expertise to integrate AI and therefore pay consultants. Eisenberg agrees that specialized knowledge has value and points to consulting firms that combine software, data and services. He expects customers to pay increasingly for measurable improvements, though the pace is uncertain.

He interprets slow growth at many public software companies as a sign that purchasing may be slowing and that earlier software has already captured some basic efficiency gains. Stebbings offers other possibilities, including market saturation and slow product expansion. These are competing explanations, not a settled diagnosis.

On adoption, Eisenberg holds two ideas together: AI capabilities and some uses are advancing quickly, while economy-wide deployment can take much longer. Regulated industries such as finance and agriculture face additional constraints, and regulators themselves must understand rapidly changing technology.

He argues that restrictive data rules could slow development and predicts Europe will be more affected than the US or Israel. He says European founders should leave; that is his policy and business opinion at the time, not a neutral assessment of every jurisdiction.

Stebbings raises the view that powerful AI systems should be closed to protect military advantage. Eisenberg agrees AI can matter in warfare, but doubts that nations can fully keep the technology closed. He argues that continued technological competitiveness is necessary; his geopolitical claims are opinions and forecasts.

He questions whether restrictions on advanced chips will decisively slow China, suggesting other hardware and designs could be combined. Asked if China was two years behind the US, he instead points to the sophistication of TikTok’s recommendation systems. That comparison does not establish relative frontier-model capability.

Stebbings expresses discomfort with investors treating defense solely as a margin opportunity when the products can make killing more efficient. Eisenberg, whose children served in the Israeli military, stresses the seriousness of war and says weapons entail responsibility. He also frames defense in terms of his own views on patriotism and security.

Eisenberg expects substantial investment in defense but warns that selling to governments involves regulation and expertise that outsiders can underestimate. He names Palantir and Anduril as examples of companies whose experience in the market matters. This is his assessment, not a projection of individual company returns.

As investors turn from software toward physical and scientific technologies, he questions why a founder would choose a backer whose experience does not fit the problem. He also sees a middle ground between conventional SaaS and the most capital-intensive defense businesses, including semiconductors and synthetic biology.

Eisenberg proposes a “labless” model in which AI work is done internally and experiments or manufacturing are outsourced to existing facilities. He thinks infrastructure built in earlier funding cycles makes this possible for some companies. The practical economics depend on the specific work.

He says he returned to learning chemistry with ChatGPT and learned from a scientist he brought to Israel while building a synthetic biology company together. He also cites portfolio company Empathy and says an investor need not pretend to be an expert in every domain; knowing whom to introduce can help.

Stebbings asks where venture liquidity will come from, including acquisitions, IPOs and private equity. Eisenberg sharply criticizes then-US antitrust regulator Lina Khan and argues her approach restricts acquisitions. That is his political and regulatory judgment, not a settled legal conclusion.

He argues that the IPO market is open at a price a company may not like, citing Lemonade’s public listing with approximately $60 million in revenue and other historical examples. He prefers some companies to list earlier and build publicly. His examples and price references reflect the interview period, and listing suitability varies.

He doubts private equity will buy most software companies at prices their investors expect. In his view, higher borrowing costs and price sensitivity constrain those buyers. He also discusses liquidation preferences: in a disappointing exit, investors with senior rights may be paid before other shareholders.

Eisenberg expects more stressed software holdings beyond the example raised by Stebbings, though he does not identify or substantiate a complete set of affected companies. He says recent write-downs could also reduce private-equity appetite. These are forecasts and inferences, not verified portfolio-wide findings.

He describes selling all of Aleph’s stake in one unnamed company at an approximately elevenfold return because the firm’s view differed from management’s. He says he would still be pleased if the founder and remaining investors achieved more. The example reflects a historical decision, not a recommendation for every investor.

On secondary-market inquiries, Eisenberg says his team contacted people advertising potential buyers and found that some were seeking due-diligence information rather than presenting committed buyers. His experience does not prove that all secondary buyers are absent.

Stebbings prefers selling incrementally and retaining some upside. Eisenberg says he generally likes to ride winners but sometimes chooses a complete sale when he has a strong view of the risk and return. He treats venture positions as unusually binary and distinguishes that from averaging in and out of public stocks.

Aleph rarely exercises pro rata rights, he says, preferring to buy meaningful ownership in its first round. He cites a large initial investment in Dream Security, around 6.5–7% of the fund, as an example. His comments are specific to the firm’s strategy and portfolio.

He recalls another investor’s insight that follow-on capital can be useful in middle-performing companies: obvious winners may be inaccessible, while the weakest may not justify more money. Some less celebrated companies can improve with time and investment. Stebbings shares a similar observation from his own small fund.

Both see consumer businesses as potentially more transient than enterprise businesses, although Eisenberg notes that durable consumer moats do exist. He says it remains unclear how many AI products will develop defensible advantages when consumers perceive model outputs as similar.

Eisenberg does not typically invest by projecting a specific exit. Companies pivot, he says, and an imagined sale can create false conviction. He expects some venture funds to close, including partnerships where one strong investor eventually leaves and funds that fail to return capital.

On distributed capital versus total value, he says distributions matter but unrealized value also matters if the underlying business is compounding sustainably. The question beneath the headline multiple is how real the valuation is and how durable the company’s competitive advantage and growth are.

Growth of 30–40% can still describe a strong business, he says, while 10–20% growth can be awkward for a venture-backed company with a large liquidation preference. Some may remain profitable and operate for years, sell in secondary transactions or buy back shares. Outcomes depend on their finances and rights.

He describes a profitable portfolio company buying back some early investors’ shares, while Aleph retained its stake because he thought the price too low. He also recalls a past investment that distributed substantial dividends. Both are examples from his experience rather than forecasts of returns.

Asked where he needs to improve, Eisenberg says his directness can take founders time to appreciate and that he could deliver hard messages better. He also thinks well-prepared board meetings can often be shorter, with real governance and a few strategic questions prioritized.

His preferred preparation is a data-rich written letter identifying one to three strategic issues, supplemented by calls in advance. He dislikes board members turning every product plan into an opportunity to offer product advice. He prefers in-person meetings while recognizing that geographically dispersed boards make them difficult.

He says he sometimes becomes too absorbed in portfolio companies and needs to step back, particularly with first-time founders who may find his intensity unsettling. He recalls spending months debating an idea with a founder during the pandemic and believes a less aggressive approach might have helped the founder start sooner.

Eisenberg prefers founders who are new to a market, citing Lemonade’s founders as people willing to challenge insurance conventions. That differs from a preference for industry insiders and is his own investing style. Aleph has a direct financial interest in Lemonade.

In quick-fire comments, he says the response of young Israelis to war changed his view of their generation. He reports foreign investment interest in Aleph companies after October 7, while warning that founders from a small country may need help thinking at global scale. These are his observations at the time.

He names direct board product advice and insufficient selling during the low-interest-rate era as mistakes or frustrations. He values hard work and says he avoided investing in China because of what he views as political risk. His characterization of China investing is an opinion.

Eisenberg says he no longer actively sources deals, relying on referrals built over decades and making only a few investments a year. He sees his network as a strength and management advice as a weakness. He also acknowledges that people close to him may feel he gives them too little time.

At the close, Eisenberg thanks Stebbings and discloses that he is a limited partner in Stebbings’s fund, a financial relationship relevant to the interview.