Baseten: The Six-Year Overnight Success of the AI Inference Gold Rush

Key Facts

• Baseten is a San Francisco-based AI model deployment platform, founded in 2019 by Tuhin Srivastava, Amir Haghighat, Philip Howes, and Pankaj Gupta, that saw its valuation climb from $825 million in February 2025 to a reported $11-13 billion by mid-2026.
• Who founded Baseten: Tuhin Srivastava (CEO), Amir Haghighat (CTO), Philip Howes (Chief Scientist), and Pankaj Gupta, in 2019, all with over 15 years of collective machine learning experience.
• How fast has its valuation grown: From $825M in February 2025, to $2.15B in September 2025, to $5B in January 2026, to a reported $11-13B in a round being finalized around June 2026 — a roughly 15x increase in under a year and a half.
• What does Baseten actually do: It provides infrastructure for deploying, optimizing, and scaling open-source, custom, and fine-tuned AI models in production, running on cloud providers like AWS and Google Cloud rather than its own data centers.
• Who uses Baseten: Over 100 enterprises including Descript, Patreon, and Writer, serving as inference infrastructure for AI-native startups and technical teams.

Six Years of Building Before the World Noticed

Baseten's own account of its rise is refreshingly candid about how long the underlying work actually took. Announcing its $150 million Series D round in September 2025, the company's blog post opened with a specific framing: "After more than six years building, we are uniquely positioned to solve today's AI scaled inference problem". Founded in San Francisco in 2019 by CEO Tuhin Srivastava, CTO Amir Haghighat, Chief Scientist Philip Howes, and Pankaj Gupta — a founding team with over 15 years of combined machine learning experience — Baseten spent years developing deployment infrastructure for a market that, at the time of its founding, had nowhere near the scale or urgency it would develop after ChatGPT's late-2022 launch reshaped enterprise appetite for generative AI overnight.

That six-year gap between founding and its most explosive valuation growth places Baseten in a similar pattern to Fireworks AI elsewhere in this series: a company that built genuine technical depth in AI inference infrastructure well before the broader market recognized how valuable that specific capability would become, then experienced a dramatically compressed valuation catch-up once demand caught up to the underlying technology.

A Classic "Picks and Shovels" Business in the Middle of a Gold Rush

Yahoo Finance's coverage of Baseten's September 2025 round frames the company's business model using the enduring gold-rush metaphor directly: "This venture epitomizes a classic 'picks and shovels' model, thriving amid a booming AI landscape by assisting organizations in deploying, managing, and scaling AI applications". That framing captures a structural advantage several infrastructure companies profiled in this CrackTheDeck series share: rather than betting on which specific AI application or model wins, Baseten profits from the sheer volume of AI development activity happening across the entire industry, regardless of which particular companies or products ultimately succeed.

More specifically, Baseten focuses on inference — Yahoo Finance describes this as "the method by which trained models apply their knowledge to produce forecasts and decisions", distinct from the training phase where a model is initially built. Rather than maintaining its own physical data centers, Baseten deliberately builds on top of infrastructure from established cloud providers including Amazon and Google, per CNBC's February 2025 reporting — a capital-efficient strategy that avoids the enormous fixed costs of building and operating physical data centers directly, instead focusing engineering resources on the software layer that makes inference on top of that borrowed infrastructure faster, cheaper, and more reliable than customers could achieve building it themselves.

Built on Nvidia's TensorRT-LLM for Genuine Performance Gains

Baseten's technical architecture leans heavily on Nvidia's hardware and software stack. An Nvidia customer case study describes the partnership directly: "Baseten leverages NVIDIA GPUs and NVIDIA TensorRT-LLM to provide machine learning infrastructure that's high-performance, scalable, and cost-effective," with the company's stated mission being to "provide machine learning (ML) infrastructure that just works". That case study also describes the flexibility of Baseten's customer offering: clients can bring their own custom models or select from a range of pretrained models, then deploy them in production served on Baseten's own infrastructure stack.

The company's own Series D announcement lays out three specific technical pillars underpinning its platform: "Fastest models," meaning a fast runtime with dedicated latency and throughput optimization tools; "Interchangeable compute," meaning the ability to seamlessly draw on cross-region and cross-cloud capacity for both raw scale and resiliency against outages; and "Flexible, open, and Pythonic runtimes," providing developers maximum control and visibility into how their deployments actually run.

The DeepSeek Moment: How a Chinese Model Release Boosted a U.S. Inference Startup

One of the more counterintuitive details in Baseten's growth story involves the release of DeepSeek's AI models in early 2025. CNBC's February 2025 headline described Baseten's $75 million funding round explicitly as coming "following DeepSeek's" emergence — a framing that reflects a broader dynamic across the inference infrastructure market: when a powerful new open-source model like DeepSeek's release demonstrates that state-of-the-art AI capability doesn't require paying a closed-model provider like OpenAI or Anthropic, demand surges for infrastructure companies like Baseten that specialize specifically in efficiently deploying open-source and custom models in production. That February 2025 round, led by existing investors IVP and Spark Capital, valued Baseten at $825 million and demonstrated, per CNBC's framing, that "venture capitalists believe tech's AI boom stands to benefit a plethora" of companies well beyond the foundation model labs themselves. At that point, the company employed approximately 60 people and served over 100 enterprises, including named customers Descript, Patreon, and Writer.

An Almost Unbelievable Valuation Curve: $825M to $13B in 16 Months

Baseten's subsequent fundraising pace accelerated dramatically. In September 2025, the company closed a $150 million Series D led by Bond, with new participation from CapitalG, that Yahoo Finance reported "almost triples Baseten's valuation to $2.15 billion". Just four months later, in January 2026, SiliconANGLE reported Baseten had raised a further $300 million Series E, co-led by IVP and CapitalG, with Nvidia reportedly contributing $150 million of that round directly, pushing the company's valuation to $5 billion — again roughly doubling from the previous round.

By May 2026, The Information reported Baseten was already in talks to raise a further $1 billion at an $11 billion valuation, with the company's annualized revenue reported to have "surged from $200" million in a short span. That process culminated, according to TechCrunch's June 2026 reporting, in a stunning $1.5 billion round representing "a 160% increase in valuation in less than half a year" from the January Series E. Notably, TechCrunch specifically flagged that this was structured as a "split-priced round" — a financing tactic where different investors in the same round pay different effective valuations, with some coming in at $13 billion and others at $11 billion, a technique TechCrunch described as one "startups are using to boost their headline valuation and make lead investors look good on paper". That round was reportedly co-led by Spark Capital, Sands Capital, Altimeter Capital, and Wellington Management — a detail worth flagging directly to readers, since it means the widely reported $13 billion headline figure reflects the more favorable end of a genuinely split-priced structure, not a single, uniformly agreed valuation across all participating investors.

TechCrunch's coverage placed this entire trajectory within the broader context of what The Next Wave newsletter dubbed the "inference gold rush," describing venture capitalists "pouring enormous amounts of money into companies building the inference layer" across the AI industry — with Baseten's specific curve serving as perhaps the single clearest illustration of just how aggressively that capital has been flowing into inference-layer infrastructure specifically, distinct from the also-enormous but separately tracked capital flowing into foundation model labs themselves.

Some Customers Push Back on Pricing and Lock-In

Not every account of Baseten's platform is uniformly positive. A comparison published by Spheron Network specifically criticizes aspects of the company's commercial model, stating that "Baseten's replica-hour pricing and Truss lock-in push teams to look elsewhere," in an article specifically comparing ten alternative ML inference platforms across GPU cost and other criteria. That kind of competitive commentary is worth noting directly: Baseten operates in a genuinely competitive inference infrastructure market alongside companies like Fireworks AI, Modal, and Together AI, and its specific pricing structure and proprietary Truss deployment framework have drawn some criticism from technical teams evaluating alternatives, even as the company's overall growth trajectory and enterprise customer base have continued to expand rapidly.

Why Baseten's Story Matters Beyond Its Own Valuation

Baseten's rise captures something distinctive about how value has been distributed across the AI industry's build-out: the company itself doesn't build foundation models, and it doesn't build consumer-facing AI applications — it builds and operates the infrastructure layer that sits between those two things, letting other companies actually get their AI systems into the hands of real users reliably and cost-effectively. That position, six years in the making before the market caught up to its value, has let Baseten capture an outsized share of the capital flowing into AI infrastructure specifically, a pattern echoed closely by Fireworks AI's similarly dramatic valuation curve, and one likely to continue shaping how investors think about where durable value accrues across the broader AI technology stack.

FAQ

Source: CrackTheDeck Research.