The governance gap
Pension funds lose allocations when they overlook structural risks within a governance hierarchy. While most institutions expect a dual-class share structure where founders hold voting control, Anthropic inverted this model. An independent Long-Term Benefit Trust, consisting of five AI safety and national security experts with no equity, holds the authority to elect or remove a majority of the board. Dario and Daniela Amodei hold Class B super-voting shares, which gives them day-to-day operational control. This separation of economic interest from governance authority means a non-shareholder body manages the mission. You already know that most LPs prioritize voting rights to protect their economic interests. Amazon and Google hold zero voting rights or board seats despite committing tens of billions in capital. I would flag any fund that misses the concentration risk of a team where carry remains top-loaded to a single founder. In 2025, 18% of funds saw LPs exercise governance rights, such as key-man provisions, because of team disruption. Pension funds running 6 to 18 month due diligence processes must use the 2021 ILPA DDQ 2.0 to investigate how investment decisions require consensus or if they rely on a single individual. Current board members like Yasmin Razavi and Chris Liddell signal IPO preparation, but the LTBT remains the primary governance layer.
The scaling fallacy
LPs fail to secure allocations when they misjudge how a strategy scales or how a track record translates to new market realities. Anthropic saw its valuation increase 16x in just 14 months. This growth stems from usage-driven enterprise adoption, specifically Claude Code, which reached a $2.5 billion run-rate revenue in February 2026. The revenue trajectory appears unprecedented.
| Metric | 2025 Figure | 2026 Projection |
|---|---|---|
| Annualized Revenue | $9 billion | $45 billion |
| Valuation | $61.5 billion (Series E) | $965 billion (Series H) |
A GP that generated 25% net IRR deploying $200 million lacks a proven ability to manage a fund deploying $800 million because deal size increases and competitive dynamics change as fund sizes grow. Market conditions change as fund sizes grow. LPs must ask how a strategy changes at larger scales. Many funds rely on one exceptional investment to carry their track record, which creates a lack of consistency. This lack of consistency often causes LPs to discount interim marks, which show optimistic valuations while a fund remains in its holding phase. In May 2026, Anthropic announced a $65 billion Series H round at a $965 billion post-money valuation. The revenue grew from $9 billion to $45 billion in six months. The company reported $4.6 billion in revenue for 2025, up from $400 million in 2024. Claude Sonnet 4.6 launched in February 2026, and Claude Opus 4.7 followed in April 2026.
Operational and regulatory blindspots
Operational due diligence fails when investigators ignore related-party commercial ties and regulatory scrutiny. Anthropic maintains $80 billion in compute commitments through 2029, with much of that flowing to its two largest equity investors. Amazon serves as the primary cloud provider, while Google provides TPU access. The SEC flagged valuation methodology as a priority for 2025 and 2026, particularly for managers with complex asset exposures. I find the lack of transparency regarding whether cloud pricing reflects market rates or strategic investor pricing to be the most significant risk. If the pricing favors the hyperscalers, the economic reality for other users changes. I also note the complication of the FTX liquidation, where $884 million was paid for an 8% stake. This left institutional buyers like Mubadala on the cap table. The historical relationship with a company convicted of massive fraud requires scrutiny. CalPERS board members recently discussed the risks of AI, specifically after researcher Jacob Coxon resigned and claimed companies are gambling with human lives. Will the LTBT prioritize mission over profit when the IPO goes live?
