Bessemer Venture Partners closes $5.75B fund, with $1.75B for early stage and $4B for growth

Bessemer closes $5.75B fund, eyes personal AI as next major category

Bessemer Venture Partners has closed a $5.75 billion fund, split between $1.75 billion for early-stage investments and $4 billion for growth. Talia Goldberg, a partner at the firm, says roughly 70% of Bessemer's investments have historically started before companies have revenue or even a name, and that posture isn't changing. What is changing is the growth strategy: rather than spreading capital across many late-stage positions, Bessemer plans to lead growth rounds with large, concentrated checks in a smaller set of companies it believes will dominate their categories.

---

“We just raised $5,750,000,000. $1,750,000,000 of that is dedicated to early stage companies... And then $4,000,000,000 for growth, which lets us keep backing companies as they're at their inflection points... Returns are concentrating in fewer larger winners. And so the right move for us is to be a meaningful investor.”

Token market fit

Goldberg frames the current AI investment landscape around a concept she credits to a portfolio company: token market fit, defined as categories where the average end user can productively consume roughly $10,000 a month of tokens. By her count, only three categories have genuinely reached that threshold so far: coding, video and media production, and high-frequency trading. Legal, customer support, and sales have not, despite the attention they've received.

Her investment in Legora, a legal AI company, illustrates why she sees that gap as opportunity rather than failure. Most legal AI deployments are still copilot tools, not autonomous agents. Large law firms haven't materially changed headcount yet. As agents get better at doing actual legal work rather than assisting it, she expects token spend in those categories to climb significantly.

There's a reasonable counterargument here: some categories may find genuine product-market fit while remaining structurally low-spend. Unlike software, where products are never quite finished and iteration is continuous, legal work has natural completion points. A deal closes. A brief is filed. That's a constraint on how much token spend even a fully automated legal stack can generate. Goldberg acknowledges the distinction but argues legal is too early to write off.

---

Aggregation risk

On whether model labs and front-door AI platforms will extract an app-store-style cut from inference providers, Goldberg's view is that the market won't resolve cleanly in either direction. She points to Fall as a case where the aggregator also controls model access, including third-party models, letting it capture both sides. But she sees the Amazon/Shopify dynamic, where large platform aggregators and independent tools coexist under tension, as the more likely long-run structure. She flags Instinct and the Meta/Amazon/Shopify ecosystem as live examples of that aggregation-disaggregation tension playing out.

---

Personal AI agents

Goldberg calls personal AI agents arguably the most important next category in AI. Bessemer is a seed investor in Instinct Inc., though Goldberg concedes the position is smaller than she'd like given how the product has developed. She describes the consumer reaction as the second "oh my God" moment she's witnessed since ChatGPT launched, specifically because it's reaching people with no technical background. Unlike the coding-agent wave, which required familiarity with terminals and developer workflows, personal AI agents are landing with non-technical users immediately.

Her view is that the interaction model, proactive, simple, and genuinely delightful, will migrate into enterprise workflows. She doesn't see this as winner-take-all: multiple agents will coexist, with labs and large technology companies all competing hard to own the relationship.