JOSHUA KUSHNER'S Thrive Capital has spent the past several years becoming one of the most consequential investors in frontier AI, leading OpenAI's $6.6 billion round, backing Cursor, Databricks, and Anduril, and closing its largest venture fund ever at more than $10 billion in February. The firm now manages upwards of $50 billion. But its latest move is not another bet on a foundation model. Thrive Holdings, the firm's operating arm, is in talks to raise at least $2 billion, according to Bloomberg, doubling the war chest of a company that launched barely a year ago with $1 billion and a thesis that would have sounded strange in Silicon Valley five years ago: that the biggest returns in AI will come not from building the intelligence, but from owning the businesses it transforms.
The company has already secured $1 billion in capital commitments, with existing backers (pension funds and endowments among them) pushing for additional capacity. Thrive Holdings is structured as a permanent capital vehicle, a design borrowed from private equity that allows the firm to own and operate businesses indefinitely rather than cycling them through a five-to-seven-year fund life. "The idea is to build these businesses over decades," Kareem Zaki, a Thrive Capital partner and founding member of Thrive Holdings, told Bloomberg earlier this year. Its two core bets so far are Crete Professionals Alliance, a roll-up of more than 20 accounting firms now generating over $300 million in annual revenue, and Shield Technology Partners, an IT services consolidator that crossed $100 million in annual revenue by the end of 2025 after a $100 million infusion in February. OpenAI has taken an ownership stake in Thrive Holdings and is embedding its own engineers directly inside portfolio companies to redesign workflows. The arrangement turns Thrive into something like a full-stack AI play: funding the supply of intelligence on one side, and owning the demand for it on the other.
The ketchup principle
To its admirers, Thrive Holdings is not a pivot but an inevitability. Kushner founded Thrive in 2010 with a set of claims that sounded eccentric at the time: that the firm would be stage-, geography-, and sector-agnostic; that it would incubate companies as well as invest in them; and that it would function as an "embedded operational commando unit" for founders. The firm demonstrated the model early. When GitHub lost its CFO, head of product, and head of marketing in 2015, Thrive dispatched Nabil Mallick, a former 3G Capital operator whose prior experience was running Heinz's North American finance and operations, to serve as the company's de facto interim CFO. "It also helps to know ketchup," Mallick has said, "which is just like software." Zaki, the son of Coptic Christian immigrants from Cairo, has argued since joining the firm that "tech is a horizontal sector, not an isolated vertical," meaning every category-leading company in every industry will eventually be defined by its adoption of technology. Thrive Holdings is that thesis taken to its logical extreme: if you believe AI will remake accounting and IT services, why merely sell tools to those industries when you can own the firms outright and rebuild them from within?
Yet Thrive is not alone in this logic. Silicon Valley, historically fixated on building the next software platform, is increasingly placing parallel bets on buying existing businesses in industries that have barely been touched by technology, then grafting AI onto their operations. Bezos is reported to be assembling a $100 billion fund through Project Prometheus to acquire and transform manufacturing companies in chipmaking, defense, and aerospace. Sequoia partner Ravi Gupta and former Meta chief revenue officer John Hegeman are raising at least $1 billion for Ithaca Holdings, a Berkshire-style entity that will acquire companies and inject them with new technology. General Catalyst, in talks to raise $10 billion of its own, has committed more than $1 billion to applying AI in low-tech categories, including its own accounting roll-up called Accrual.
The aggregate capital being marshaled for this thesis (call it the AI roll-up) is staggering. Between Thrive Holdings, Bezos, Ithaca, and General Catalyst, something north of $110 billion is being raised or discussed to pursue variants of a single idea: that legacy industries sitting on fragmented workforces, aging proprietors, and routine-heavy workflows are the ideal proving ground for generative AI. The logic runs that selling software to these businesses has never worked well (accountants are not early adopters), but owning them outright gives investors the authority to rebuild operations from within. For Thrive, the vertical integration is the point: you do not need to convince a CPA firm to adopt OpenAI's tools if you already own the CPA firm and OpenAI is your partner.
The trouble is that the logic of acquisition and the logic of AI transformation are not the same thing, and conflating them carries real risk. Roll-ups in professional services are nothing new; private equity has been consolidating accounting, dental, and veterinary practices for decades (Crete co-founder Jake Sloane's prior roll-ups spanned veterinary clinics and plumbing companies). Those deals generate returns primarily through financial engineering, buying at lower multiples, layering debt, extracting procurement efficiencies, not through operational reinvention. Venture capitalists pursuing the same strategy under the banner of AI must deliver something demonstrably different: not incremental margin improvement, but the kind of step-change in productivity that justifies VC-scale return expectations. Analysts remain uncertain whether AI-enhanced services businesses can deliver the higher returns that venture investors require compared with traditional PE.
Crete offers a useful early test case. Its in-house tech team, working alongside OpenAI, has built custom tools for data mapping, memo drafting, and audit testing, reportedly saving staff hundreds of hours monthly at acquired firms. That is real, if modest. But transforming $300 million in accounting revenue into a venture-scale outcome requires either extraordinary margin expansion or an acceleration in acquisition pace, both of which tend to collide with the unglamorous realities of integrating dozens of small professional practices, each with its own culture, clients, and quirks. Thrive's operational embedding playbook, honed over a decade at GitHub, Stripe, and elsewhere, is a genuine differentiator here; most PE roll-ups do not arrive with an in-house tech team and a direct line to OpenAI's engineering org. Whether that advantage proves decisive or merely decorative remains the open question.
Nitin Nohria, the former dean of Harvard Business School who became Thrive's executive chairman in 2021, has compared the firm's ambitions to those of J.P. Morgan, who positioned himself as the trusted party anchoring the movement of capital across the railroads and steel industries that defined his era. "With AI, data centers, the associated energy and chips, these are our equivalent of Morgan's railroads and steel," Nohria has said. Thrive Holdings is a bet that the comparison extends further: that the firm sitting at the intersection of frontier AI investment and analog-economy ownership will capture value that a pure model builder or a pure acquirer cannot. If that thesis holds, the most important AI company of the next decade might not be a lab in San Francisco but a holding company that owns your accountant, your IT help desk, and the models running both. If it does not, a great deal of pension-fund money will have bought some very expensive CPA firms. ■
For more, join 75,000 subscribers getting tech's favorite brief here
