Meta’s AI Talent Drive, Microsoft Layoffs and the Economics of Startup Growth

Harry Stebbings, venture investors Jason Lemkin and Rory O’Driscoll discuss Meta's recruitment of Nat Friedman and Daniel Gross, competition for AI talent, startup finance and changing workforces.

Stebbings introduces Friedman and Gross's move to Meta and cites reported figures for their NFDG fund. The guests examine why successful investors might choose to return to building products at a consequential moment in AI.

Lemkin cites a report that the $1.1 billion fund was roughly half deployed and had a fourfold paper return. The figures are reported estimates, not distributions or independently verified realized returns.

The panel says large technology companies can now pay extraordinary sums for scarce AI talent. Lemkin wonders whether people attracted by the excitement will stay at a large company for years.

The investors run a simplified calculation of potential carried interest from the reported fund performance. O’Driscoll stresses that paper gains are not cash currently received by the general partners.

O’Driscoll describes his understanding of a possible liquidity option for limited partners and estimates what future investing opportunities the partners might give up. His numbers depend on assumptions about future returns and terms, not confirmed personal compensation.

He argues the outcome could offer limited partners early liquidity, while they would lose the stewardship of Friedman and Gross on future investments. He describes the possible fund purchase and allocation as his understanding, not verified final terms.

Stebbings says some limited partners may be disappointed to lose the partnership even if they receive cash and attractive economics. O’Driscoll thinks liquidity may soften that disappointment; neither claims to speak for every investor.

O’Driscoll says he competed with the pair for a deal and later worked with them. He sees their operational abilities as a reason they might prefer building to managing a venture fund.

The panel reflects on what venture investors uniquely contribute. Lemkin values finding founders who would otherwise struggle to raise money; Stebbings describes Project Europe backing an early robotics founder.

Stebbings questions the implications for investors in Safe Superintelligence after Gross's departure. The guests distinguish a prior operating role from fund management and offer opinions rather than claims about investors' private agreements.

Lemkin compares the move with other investors returning to operating roles. O’Driscoll says investors should consider people's incentives and motivations, while accepting that plans and personnel can change.

Stebbings asks whether Meta's concentration of talent will work. Lemkin predicts talented people will attract more talented people, though successful recruiting does not itself establish a successful product.

The panel discusses the value of reputation in recruiting. O’Driscoll expects Meta's hires to build a capable product but says the harder business question is whether another competitive model will create sufficient value for Meta.

Stebbings reports a founder's concern about competition for engineers from Cursor. The panel predicts that heavily funded AI companies could make recruiting difficult for other startups; the quoted compensation is anecdotal.

Lemkin predicts recruiting will be a major challenge for business-to-business AI in 2026. He argues that appointing an AI executive alone is not a substitute for building the right technical capability.

O’Driscoll distinguishes building foundation models from applying them, suggesting strong teams can be built in other locations. Lemkin counters that applications also need exceptional teams to stand out.

Lemkin relays an unnamed executive's claim that some AI engineers now require unusually large equity grants. The account is anecdotal, and the impact on dilution depends on actual grants and company valuations.

Stebbings raises the possible effect of employee stock compensation on venture returns and cites The Information's report that OpenAI's stock-based compensation exceeded revenue. The panel treats the report as an input to debate, not independently verified financial data.

O’Driscoll argues that private-company stock compensation should be considered partly as ownership dilution, while liquidity can make it resemble cash compensation. He says high dilution can still be worthwhile if the company creates enough value.

The panel cites a reported $4.4 billion in stock compensation and 119% of revenue for a previous year. O’Driscoll cautions that accounting charges alone do not reveal the percentage of ownership diluted, especially as valuation changes.

They illustrate the difference with a hypothetical stock issue relative to a company's valuation. Future value to employees depends on the share price and opportunities to sell, neither of which is guaranteed.

Stebbings turns to CoreWeave's reported $9 billion deal for Core Scientific and asks whether the company is using appreciated stock strategically.

O’Driscoll favors the transaction in principle. He recalls the collapse of hosting company Exodus and argues that replacing some fixed obligations with equity could make a data-center business more resilient if demand slows.

Stebbings questions whether data-center demand could decline. O’Driscoll considers a slowdown plausible over several years, while describing his view as a scenario rather than a prediction that demand must fall.

Lemkin says companies with valuable public equity can use it to address structural challenges. O’Driscoll adds that CoreWeave remains exposed to data-center demand and that its long-term contracts also matter.

The panel asks whether Circle might use its appreciated stock for acquisitions. They discuss different incentives for a profitable company and a loss-making business; their proposed future deals are speculation.

O’Driscoll distinguishes taking on dilution to build a viable business from distributing value after the business is profitable. Both investors say dilution needs to be considered in context.

Stebbings raises Thoma Bravo's proposed acquisition of Olo, citing roughly $320 million of annual recurring revenue and a $2 billion price. The figures are presented as the panel's discussion of the deal.

Lemkin regards the take-private as encouraging for vertical software, while noting that a roughly six-times revenue valuation does not offer the same outcome as a much higher multiple.

O’Driscoll describes the deal as a conventional purchase of a profitable, growing business. He says such transactions may create exits but are unlikely on their own to rescue every struggling unicorn.

The guests speculate that a private-equity buyer might add complementary products and sell more to existing restaurant customers. They do not claim to know the buyer's actual plan.

Asked about Owner, a company in which Lemkin has invested, he says its smaller customers illustrate why the long tail can matter in some consumer-facing markets. He cautions against assuming every software company should move upmarket.

Stebbings contrasts expensive headline deals with data suggesting venture deal volume has fallen. O’Driscoll sees a flight toward consensus and a small number of companies attracting disproportionate attention.

He argues that early momentum can compound in AI markets when a company continues to execute. Others may struggle for attention even if their products are close in quality, in his view.

Lemkin describes passing on a startup with solid growth because its differentiation and long-term advantage were unclear to him. He says investors have limited opportunities and are choosing to pursue more exceptional outcomes.

The panel discusses traditional SaaS growth benchmarks and whether a business can finance its path to scale. O’Driscoll notes that early companies dependent on future rounds must persuade later investors as well.

Lemkin emphasizes durable differentiation and market leadership, not just early growth. O’Driscoll warns that extrapolating current AI growth indefinitely could repeat a mistake investors made in earlier software cycles.

The discussion shifts to private equity in retail investment products. O’Driscoll raises concerns about fees, private valuations and liquidity, especially for investors less able to hold through a long cycle.

Lemkin explains target-date funds as products that adjust the mix of assets as retirement approaches. Both guests question how illiquid private assets fit that structure; their comments are opinions rather than individualized financial advice.

The panel discusses university funding pressures and their possible effects on research and venture investment. O’Driscoll worries that cuts can affect scientific programs and researchers.

The guests discuss a contemporaneous US tax bill and changes they say affect qualified small-business stock. They acknowledge that eligibility and tax outcomes vary, and speak from the perspective of investors who may benefit.

Lemkin describes his understanding of potential federal exclusions and says he uses trusts in his own planning. His remarks are personal, time-specific claims, not a determination that any investor or transaction qualifies.

Stebbings raises Microsoft's announced layoffs of 9,000 people and asks about replacing general sales roles with more technical customer-facing roles. The panel's discussion does not establish that all layoffs were caused by AI.

Lemkin predicts AI could replace a portion of simpler sales work and argues that enterprise customers will increasingly value people who understand the product deeply. These are forecasts, not measured results from Microsoft's layoffs.

The panel notes that the reported shift is framed as hiring people with different skills rather than simply replacing staff with AI. O’Driscoll reflects on companies seeking further efficiency even with substantial margins.

Lemkin describes Canva's AI discovery week, where employees could spend time learning tools. He sees it as one way to signal the importance of learning new technology.

Stebbings argues strongly that employees unwilling to embrace AI should be dismissed. Lemkin and O’Driscoll consider a more gradual training approach, and the discussion reflects their management opinions rather than evidence about individual workers.

Stebbings describes a recurring 20VC practice: a weekly hour to try tools followed by a show-and-tell. The panel sees continued experimentation as useful even while differing on how employers should respond to resistance.

The guests begin a prediction-market quick-fire round, including whether a Sequoia partner will leave. They offer speculative bets about an individual's future, not evidence of any planned departure.

On the 2025 economy, O’Driscoll compares the market's suggested probabilities with historical recession risk. Lemkin argues that the definition and timing of a recession matter to the wager.

O’Driscoll initially cautions that an expansion does not end simply because it is old. Later he revises his view, suggesting that risky behavior accumulated during good times can increase vulnerability.

The panel makes contemporaneous predictions about X's leadership and discusses Elon Musk's political activity. These are individual judgments at the time, not confirmed personnel decisions.

Asked whether CoreWeave and Circle can sustain high share prices, O’Driscoll says it depends on whether the market has reassessed their value or prices have detached from near-term fundamentals.

Returning to recession risk, O’Driscoll updates his earlier reasoning: long expansions may encourage debt and aggressive decisions that leave the economy more exposed. The guests close the discussion.