CrossWork · Late-stage private AI
CrossWork is a venture capital group for the late stage: large private companies in markets still being built, with access through an advisory board of former operators.
CrossWork backed Cerebras in 2023. In May 2026 it raised $5.6 billion, the second-largest US IPO of 2026 by a first-time issuer, behind only SpaceX.
Seven recent exits, valued at more than $50 billion in aggregate at the time of each event.
Our focus now is deep agentic AI. Gartner expects 40% of enterprise applications to embed task-specific AI agents by the end of 2026, up from under 5% in 2025.
By our reading of published forecasts, that pace outstrips the early growth of cloud and SaaS. (Our view.)
Thesis
Three observations, and the value curve that follows from them.
Companies stay private longer, so the steepest part of the value curve now sits before the IPO.
The rounds that matter are allocated through relationships. Public-market investors arrive after the repricing.
The winners are identifiable before the market names them, by people who operate in the field. That is what our advisory board is for.
Exits
Realized exits have already returned capital to investors. Early positions in companies that went on to define their categories. Detailed performance is available to verified investors in the CrossWork portal. See the outcomes
Acquired by Hg · May 2024
Audit, risk and compliance software; the largest VC-backed acquisition of 2024 at the time, per Crunchbase.
Acquired by IBM · closed May 28, 2025
IBM bought the Cassandra and Langflow company to put enterprise data within reach of AI.
Meta investment · June 2025
Meta paid about $14 billion for a 49% stake, valuing Scale at more than $29 billion; a partial realization for holders.
Growth buyout led by Vector Capital · closed Oct 16, 2025
A private-equity path to liquidity for a top AI-focused database; terms undisclosed.
IPO · May 2026
raised; the second-largest US IPO of 2026 by a first-time issuer, behind only SpaceX.
Acquired by SAP · closed May 7, 2026
SAP bought the master-data company to make SAP and non-SAP data AI-ready.
Agreed to be acquired by Cinven · July 2026 · pending
Product-experience software for consumer brands; closing subject to regulatory approvals.
Why now · Compute
Agents do work at machine speed, so demand for AI compute compounds with every task they take on.
Global AI token consumption is expected to grow about 40 times over the next four years, according to Qualcomm CEO Cristiano Amon (COMPUTEX keynote, June 1, 2026).
Qualcomm CEO Cristiano Amon, COMPUTEX keynote, June 1, 2026. Qualcomm's combined third-party sourcing and internal analysis.
The window
Each platform shift produced a small set of category leaders, and once they listed the private window on them closed. The pattern is repeating in agentic AI, faster.
ServiceNow, a cloud category leader, listed in 2012.
Facebook listed in May 2012 and Twitter in November 2013: the social-mobile leaders went public within eighteen months of each other.
Cerebras listed in May 2026 and SpaceX in June 2026. OpenAI and Anthropic have filed confidentially for IPOs, and Discord reportedly filed confidentially in January 2026.
The window thesis and the agentic stack
Discipline
What we buy, and what we do if the cycle turns.
Many investors worry about an AI bubble. So do we. That is why we back companies with real revenue, real customers and a visible path to liquidity, rather than pre-revenue stories.
Bubbles are a pricing problem; discipline on revenue and price is the answer.
Roughly $100M+ in annual revenue and a $1B+ valuation, with a visible path to liquidity.
Exposure
The fund also invests in companies that benefit from AI, not only AI-native ones, so the portfolio is not tied to the AI cycle alone.
a.
The companies building compute, models, data and agent platforms.
AI compute · AI models · AI data · data infrastructure
b.
Businesses whose margins and growth improve because of AI: cost-savings and productivity beneficiaries, and AI-enabled incumbents.
Commerce software · risk and compliance software · logistics · payments
Portfolio · Cerebras
Most AI chips are cut from a wafer; Cerebras keeps the whole wafer as one processor.
The wafer is about 57 times the area of a leading GPU die, drawn here to true relative scale.
Keeping a model's working data on one wafer avoids shuttling it between chips; Cerebras says its systems run inference up to 30 times faster than GPU systems. For buyers, that is cost and speed per token.
Illustrative geometry, sourced figures.
Portfolio · Stripe
Stripe's 2025 total volume, ranked against national GDP on one shared scale.
Equivalent to roughly 1.6% of global GDP, per Stripe.
Payments infrastructure compounds with the economy it serves, and agents will soon be paying for things too.
Nominal GDP, IMF WEO April 2026; Stripe total payment volume 2025, per Stripe. Volume processed is not revenue.
Illustrative geometry, sourced figures.
Case studies
Two positions, and what the market saw afterwards.
Case study · 2023 to 2026
Our operators saw wafer-scale compute as a platform, not a chip: the constraint on training frontier models was moving from algorithms to hardware, and Cerebras had built for that.
In May 2026 the market agreed. Its IPO raised $5.6B at $185 a share, the largest US tech IPO since Uber.
Case study · Midas I to 2024
A system of record for audit and risk, bought by operators before investors noticed it. The people who had run those functions knew what they would pay for.
In 2024 Hg acquired it for $3B+, the largest such purchase of 2024 at the time.
Portfolio
Company names are shown as text, for identification only.
Advisory board
Our advisory board is made of former operators: finance chiefs, engineering leaders, scientists and founders. They hear about companies before the market does, sit in diligence on the businesses they know best, and tell us when to walk away.
Meet the advisory boardPick a company. Three former operators frame it.
Is the revenue concentrated in a few buyers, and how does that change as inference demand broadens?
Wafer-scale is a systems bet, not a chip bet. The question is the software stack that keeps customers from porting away.
Who signs the purchase order: a lab, a cloud or an enterprise? Each buys for different reasons.
Illustrative commentary written by CrossWork; not advice and not a view on price.
Who invests
Every investor comes to CrossWork through a conversation.
Patient capital with a long horizon, looking for access that public channels do not provide.
Long-term capital adding late-stage private companies that public portfolios cannot reach.
They have run large companies and know how a category leader is built, and how one stalls.
They read a business the way its eventual buyers will, from the numbers up.
They have built and sold companies, and know what a durable one looks like from the inside.
They watched growth move from the public markets into private rounds, and prefer to be where it happens.
How it works
Tell us who you are and what you bring.
We talk about what you are looking for and whether we are the right fit.
If it fits on both sides, we share the memorandum.
