Sequoia Partner Pat Grady Releases 15-Minute Video on AI Prepared for Boston College Investment Committee
Pat Grady has released a 15-minute video detailing the AI situation prepared for the Boston College Investment Committee. He stated that this is not a roadshow, but merely the facts as seen by Sequoia, which were not intended for external dissemination.
He illustrated a trend of dual-round financing with seven internal cases from the past 12 months: Sequoia entered as a partner to help companies grow, with an average post-investment valuation of $110 million; about a month later, the next round valuation from the sole funding party averaged $3.4 billion. He noted that he had never seen this before, calling it a bubble. Some in the market have referred to this dual-pricing model as separating the partnership relationship from pure capital.
He described the growth rate as being on a different scale from history. He mentioned that JEV achieved $100 million in revenue within seven days, with The Information recently reporting its valuation at around $10 billion; a certain Instinct maintained a daily growth of about 10% after its release; and a high-value knowledge work company’s revenue is projected to grow from about $200 million at the end of last year to $700 million this year. He chose to skip over two other steeper figures.
In terms of structural judgment, he separated it into several layers. Companies want to have their own intelligence because foundational models only provide a limited few tiers on the Pareto frontier, and real workloads often do not fall within these tiers. Therefore, self-training, self-deployment, and specialization for workloads lead to dedicated architecture labs outperforming general labs. The application layer needs to be redone approximately every four months because the underlying models are changing and the baseline is rising. Organizations are shifting from hierarchical commands to intelligent networks, with AI managing information flow, allowing for greater autonomy among people; the direction is clear, but not as exaggerated as Jack Dorsey’s post a few months ago.
Internally, models have mostly been self-generating for much of last year, and alignment was only recently taken seriously; there is a focus on custom silicon, new architectures, and continuous learning, with computing power being in short supply for everyone. To secure API tokens, prices are being lowered while consultants are sent to Fortune 500 companies to create custom tools for token consumption. There exists a diffusion gap between model capabilities and implementation, which is the window for the application layer. This year, large-scale cloud vendors have started borrowing to fund capital expenditures, no longer relying solely on free cash flow.
In terms of market mechanisms, buyers are divided into two categories: early-stage institutions involved in partnership building require board access and information, while later-stage funders seek hot markers and multiples on paper. The event driving this is a pricing competition after revenue curves are rewritten daily. Funding jumped from a $110 million mark to a $3.4 billion mark within 30 days, with almost no new operating cycle in between, only new bidders. Beneficiaries are those who can enter at lower tiers and then enjoy high-tier narratives; those under pressure are employees and co-investors who calculate their shares based on headline valuations but acquire them at lower-tier costs. The bubble was named on the spot by the parties involved using their own seven samples.
Source: Public Information
ABAB AI Insight
Grady joined Sequoia in 2007 and has managed growth rounds since 2015, with public tags including Snowflake, OpenAI, Harvey, and OpenEvidence; after becoming co-leader with Alfred Lin in November 2025, Sequoia broke its old rule against investing in direct competitors, appearing simultaneously in OpenAI, Anthropic, and xAI, and ramping up its involvement in so-called re-industrialization. This video for LPs serves as a sandbox for the leaders to present internal pricing structures as industry facts that can be shared externally. The dual-pricing model has been criticized by Mercor's founder; some within Sequoia explain it as separating partnership relationships from pure checks when others are willing to offer higher prices. Grady now upgrades the explanation using the average of seven samples to a diagnosis: a 30-fold increase in one month is not a partnership premium, but a bubble dashboard.
The capital path is a layered liquidation after time compression. Partnership rounds buy governance and information rights, while bidding rounds buy externally referable markers. $110 million is the price willing to sit on the board, while $3.4 billion is the price willing to pay for narrative; the 30 days in between are insufficient to validate the second pitch deck, only enough to sell the first pitch deck to the next train. Self-training and dedicated labs are a counter to the limited tiers sold by general models: workloads do not fall on those few points, so intelligence must be re-owned. The application layer reconstructs every four months, transferring the model iteration tax from labs to product teams. Large-scale vendors borrowing for capital expenditures indicates that free cash flow can no longer cover the costs of training and inference factories.
This parallels the rapid price increases in growth rounds around 2011 in social and cloud, the SPAC era separating "relationship checks" from "public checks," and each time multiplying daily revenue curves directly into ten-thousand-fold valuations at the cycle's end. In terms of industry positioning, venture capital has shifted from "investing in a wave of platforms" to a control period where they act as both building partners and marker underwriters in the same week. The diffusion gap is the only non-bubble window left for the application layer: capabilities are already in the lab, but Fortune 500 companies have yet to implement them.
Structural judgment belongs to the transfer of pricing power through capital concentration. The mechanism is: when revenue can double daily, valuation negotiations no longer wait for the next quarterly report, but only for the next account willing to offer a higher marker. After separating partnership building from funding, low-tier costs and high-tier narratives can coexist; employee options are priced based on narratives, while partner accounts are priced based on costs. The bubble is not an emotional term, but a price difference calculated using seven internal samples.