Baseten Deep Dive: $13B Valuation, $600M ARR, and Talks at $26B

Baseten went from a $5 billion valuation to $13 billion in five months. Three months after that, investors reportedly approached the company about doubling it again, to roughly $26 billion. That is four capital raises in a year and a half for a company that bills customers by the GPU-minute instead of by the seat.

Most coverage of Baseten repeats the valuation ladder without explaining what actually produces the revenue underneath it, or what the company is doing with the capital besides buying more GPU capacity. This is a look at the business model, the funding history, the September 2026 moves into AI agent infrastructure and open-model safety research, and whether the valuation trajectory is outrunning the fundamentals.

Company Snapshot

Figures compiled from Dealroom, Bloomberg, Axios, and Revenuememo funding-tracker reporting, as of October 2026. Baseten is private and does not publish audited financials.

What Baseten Actually Does

Baseten sells the unglamorous layer that sits between a trained AI model and a live product feature: production inference. When a customer's app sends a prompt to a fine-tuned or open-source model, that request gets routed through Baseten's orchestration layer, which handles autoscaling, latency, and compliance so the customer's own engineers don't have to run GPU infrastructure themselves. The company is asset-light — it rents capacity across more than 15 cloud providers rather than owning data centers, which is why it can quote per-minute pricing instead of requiring long-term capacity commitments.

Named customers include Clay, Cursor, Lovable, Mercor, OpenEvidence, and Abridge — a roster weighted toward AI-native application companies whose own usage, and therefore their Baseten bill, scales directly with their product growth. That dependency cuts both ways, a point worth remembering in the bull/bear section below.

How Baseten Makes Money

The common thread: every line item bills against compute or tokens actually consumed, not seats provisioned, so revenue compounds automatically as a customer's own product usage grows — without a renewal conversation. That mechanic is what turned a $200 million annualized run-rate in December 2025 into roughly $600 million three months later, according to the company's own reporting around its Series F.

Funding History

Baseten has raised approximately $2.1 billion across seven rounds since its 2019 founding, per funding tracker Revenuememo. Early-round figures vary by source and are reported with less certainty than the last three raises, which are well corroborated across multiple outlets.

Sources: Dealroom, Series F coverage, Jun 23, 2026, and Axios Pro, Sep 22, 2026. Seed and Series A figures are omitted here due to conflicting third-party reporting; see FAQ for the ~$2.1B total.

Bloomberg reported on September 23, 2026 that Baseten was in talks for a round that could roughly double its valuation to about $26 billion — the same report said rival inference company Modal was separately in talks at around $15 billion. Axios had reported the talks a day earlier. Neither outlet confirmed a closed deal as of this writing, and no new round has been announced.

Product Portfolio

Betting Beyond Inference: Blaxel and Base Labs

Every story about Baseten leads with the valuation ladder. The more interesting story might be what the company did in September 2026, in the eleven days between a product launch and a funding leak.

On September 10, 2026, Baseten announced it had acquired Blaxel, a startup building microVM-based sandboxes that let AI agents execute code safely in isolated, fast-resuming environments. The stated logic: Baseten already serves and trains the models; Blaxel adds the execution, storage, and networking layer an autonomous agent needs to actually act on a task, not just generate a response. That pairs inference with agent runtime under one platform, which is a different bet than simply adding more GPU capacity.

A week later, Baseten launched Base Labs, a research arm publishing open work on continual learning and reinforcement-learning science, with Hugging Face and Goodfire named as partners on an open-weight AI safety initiative. Unlike a frontier lab, Base Labs is explicitly not training its own frontier model — it is a research and safety credibility play layered on top of a commercial inference business, the kind of move a company makes when it wants enterprise and government customers to trust it with model-level decisions, not just hosting.

Revenue and Key Metrics

Revenue nearly tripled in the single quarter between December 2025 and March 2026, which is the growth rate investors have been pricing into each successive round. Baseten does not disclose gross margin, operating loss, or profitability, and as a private company it is under no obligation to. What is public is that inference volume reportedly grew roughly 40x year over year against the 20x revenue growth figure — a gap that likely reflects falling per-call pricing as GPU costs and competitive pressure push rates down even as usage climbs.

Headcount is one of the few figures with no reliable public number: third-party workforce trackers range from roughly 165 to 415 employees depending on methodology and date, and Baseten itself said only that it planned to triple headcount in 2026 to meet demand, without giving a base number.

Competitive Landscape

Baseten competes in a crowded AI inference infrastructure field where every major player uses roughly the same usage-based pricing mechanic, because inference cost scales directly with compute consumed. The difference is scale and who is willing to pay the highest multiple for it right now.

Fireworks AI's $17.5B valuation (July 2026 Series D) already exceeds Baseten's confirmed $13B on higher estimated revenue — which is the backdrop for Baseten's reported push toward $26B.

Fireworks AI raised a $1.505 billion Series D in July 2026 at a $17.5 billion valuation, with Sacra estimating its annualized revenue near $800 million — both figures ahead of Baseten's confirmed numbers. Modal, the smallest of the three by revenue at roughly $300 million ARR, was reported alongside Baseten in the same September 2026 Bloomberg story as a candidate for a roughly $15 billion valuation. Together AI has reported figures near $1 billion in annualized revenue in earlier 2026 coverage, though more recent, well-sourced numbers were not available at publication. None of the four companies has filed to go public, and all are being priced almost entirely on growth rate rather than disclosed profitability.

Leadership Team

Bull Case / Bear Case

IPO Outlook

Baseten has not stated an IPO timeline. No underwriters have been named, no S-1 has been filed, and no executive has publicly discussed a target listing date, unlike several peers in this series who have at least floated a year. What the public record shows instead is a company staying private and raising progressively larger rounds: four capital events in roughly 18 months, with the most recent reported approach — the talks near a $26 billion valuation — surfacing just three months after the last round closed.

That pattern is common across the AI inference category right now. Fireworks AI, Together AI, and Modal are all still private and all still raising, which means the entire comparison set Baseten would be measured against at an IPO doesn't exist yet as a public-market cohort. Until one of these companies actually lists, there's no trading multiple to anchor whether $13 billion, or a future $26 billion, is a reasonable price for the revenue underneath it.