Deepgram funding: Every round, investor, and valuation (2026)

Deepgram sells the ears and the voice of AI software. Its APIs turn speech into text, turn text back into speech, and run complete voice agents that can answer a phone call, take a drive-thru order, or transcribe a doctor's appointment. Developers pay by usage, and large companies buy annual contracts or run the models on their own servers.

Deepgram has raised about $229 million across six funded rounds since it went through Y Combinator in 2016, all of it equity. Its latest, a Series C led by AVP in January 2026, valued the San Francisco company at $1.3 billion.

For most of its life, Deepgram raised small rounds and said it was close to self-funding. The investor list mixes venture firms with the companies that sell, buy, or power its technology, including Nvidia, SAP, Twilio, and In-Q-Tel. Let's dive into Deepgram's funding history.

Deepgram's first five rounds brought in less than $90 million over six years. The Series C, closed after the company had turned cash-flow positive, raised more than all of them combined.

Deepgram started as a side project of two University of Michigan physicists, Scott Stephenson and Noah Shutty. Stephenson had worked on dark matter detection, where deep learning was used to pick faint signals out of noisy data, and the pair applied the same idea to audio. The company was first incorporated as Lexika, the name that still appears on its SEC filings.

Y Combinator took Deepgram into its Winter 2016 batch. The seed round that followed in September 2016 paid for what TechCrunch called a "Google for sound," a search engine that could find spoken words inside hours of recordings. Nvidia's GPU Ventures unit added a strategic check in 2018, at a time when Deepgram was one of the early startups training speech models end to end on Nvidia chips.

Wing VC led the Series A in March 2020, when Deepgram had moved from search to transcription for businesses. Its pitch was accuracy on messy, industry-specific audio, such as call center recordings, through models trained for each customer. In-Q-Tel, the venture firm funded by the US intelligence community, followed three months later with an investment tied to a technology development agreement.

Tiger Global led the $25 million Series B in February 2021, at the height of its fast-moving growth investing. Madrona then led a $47 million extension that SEC filings show closed in two parts, in October 2021 and November 2022. Combined, the Series B came to $72 million.

Stephenson said the extension came at the company's highest pre-money and post-money valuation yet, though he did not give the number. By then Deepgram had about 300 customers, including Spotify, Auth0, and NASA, and 146 employees.

Deepgram went three years without a new round. When it returned, the market had changed: large language models had made voice agents practical, and speech recognition had become the first and last step of every one of them.

AVP led the Series C, announced in January 2026 at $130 million. The company's Form D shows $143.2 million sold to 27 investors, which suggests later closes or stock issued in the OfOne acquisition announced alongside it. Stephenson told TechCrunch that Deepgram was cash-flow positive and "weren't out looking for a raise" when the offers came in.

In August 2026, EDBI, an arm of Singapore's state-backed SG Growth Capital, invested as part of an extension to the Series C. Neither side disclosed the amount.

Deepgram became a unicorn about ten years after its founding, slow for a company in AI. The Series C is the only round with a disclosed price.

The jump to a billion-dollar price came with the voice agent boom rather than with steady step-ups between rounds. Generative AI startups now often reach unicorn status in months, while Deepgram took about a decade. At the Series C, it was valued at roughly 5.7 times the money it had raised to date.

Revenue gives the price more support than at most AI companies of this size. Deepgram said in August 2026 that it had passed $100 million in annual recurring revenue, which puts the January valuation at about 13 times that later run rate. Research firm Sacra estimates gross margins above 85%, a third-party figure the company has not confirmed.

Its best-known rival is priced far higher. ElevenLabs, which built its business on text-to-speech before moving into voice agents, closed an $11 billion Series D a month after Deepgram's round.

Deepgram's backers fall into three groups: the venture firms that led its rounds, the companies that build on or sell its technology, and a set of institutional and university funds that joined the Series C.

AVP led the Series C as a new investor. The firm invests in growth-stage technology companies and backed Deepgram after it had already reached profitability on a cash-flow basis, which made the deal closer to a growth equity bet than a venture one.

Wing VC led the Series A and Tiger Global led the first Series B tranche. Madrona, the Seattle firm that was an early backer of Amazon, led the 2022 extension through partner Karan Mehandru. All three returned in the Series C.

Alkeon Capital, a crossover fund that invests in both private and public companies, put a large share into the 2022 extension and returned in 2026. Funds managed by BlackRock also joined both rounds. These are the investors most likely to stay through an IPO.

Nvidia invested in 2018 and again in the Series A and Series B. Deepgram trains and serves its own models on Nvidia GPUs, so each dollar the company spends on compute feeds Nvidia's data center business.

In-Q-Tel's 2020 investment came with a development agreement, and the firm returned in the 2022 extension and the Series C. Its backing signals use by US government and intelligence agencies, which need transcription that can run on their own servers rather than in a public cloud.

The corporate investors are also distribution partners. SAP first invested through its SAP.iO accelerator in 2020 and returned as SAP itself in the Series C. Its ownership is very different from Deepgram's, with SAP's founder-anchored shareholder base still led by co-founder Hasso Plattner.

Twilio, whose communications APIs carry millions of phone calls, is both a customer and a Series C investor. Twilio raised about as much venture money as Deepgram before it went public in 2016, and how Twilio is owned today shows where that path can lead. ServiceNow Ventures joined for the same reason: voice is moving into the customer service software ServiceNow sells.

The University of Michigan, where the founders met, invested in the Series C alongside Columbia University and Stanford University. Citi Ventures, Princeville Capital, and Alumni Ventures also took part. The Series C Form D lists five directors: Stephenson, chief technology officer Adam Sypniewski, Zach DeWitt, Dave Wilner, and Elizabeth de Saint-Aignan.

Deepgram spends less on compute than the frontier labs because it builds smaller, specialized models. The Series C was meant to speed up work the company could already pay for, not to cover losses.

Most of the spending goes into the models. In 2025 Deepgram released Nova-3 for transcription, Nova-3 Medical for clinical notes, the Aura-2 voice model, a Voice Agent API, and Flux, a speech recognition model built to handle people interrupting each other. Flux TTS, its newest text-to-speech model, followed in August 2026.

The company runs its own inference rather than renting it from inference clouds such as Together AI, valued at $8.3 billion in 2026. That lets it sell the same models in its cloud, in a customer's private cloud, or fully on premises. Banks, hospitals, and government agencies often require the last option.

The Series C made acquisitions an explicit use of the money. Deepgram bought OfOne, a Y Combinator startup whose voice agents take drive-thru orders, when it announced the round. OfOne's founder now runs a Deepgram for Restaurants unit.

It had already bought Poised, an AI communication coach, in 2024. In 2025, Stephenson said the company looked at about one small acquisition target a week, typically teams of around ten people strong in model training and data.

Deepgram opened EU and India API endpoints in 2025. The EDBI investment came with plans for an Asia-Pacific headquarters in Singapore, led by former Salesforce executive Sriram Ved, plus more low-latency computing capacity in the region. The company says its API requests from Asia-Pacific grew 96% in a year.

The company also spends on a "Powered by Deepgram" program for companies that build products on its APIs, and on a voice AI hub in San Francisco. Deepgram says more than 1,400 organizations and 200,000 developers use its platform, including Twilio, Jack in the Box, Granola, and Vapi.

Deepgram makes money from usage. Developers pay as they go for each minute of audio transcribed or spoken, while larger customers sign annual plans that Sacra puts at a starting point of about $4,000 a year for growth customers and $15,000 for enterprise. The Voice Agent API is billed by the hour of conversation.

That model is why the company could run lean between rounds. Deepgram says it ended 2024 cash-flow positive, after years in which usage on its platform roughly tripled annually. Voice agents raise the revenue per call, since one agent conversation uses transcription, a language model, and speech generation at once, and the Voice Agent API packages all three into one hourly price.

Deepgram's growth plan is to own the whole voice agent stack: listen, think, and speak. Its first markets are contact centers, restaurants, healthcare, and financial services. In healthcare it sells transcription and agents next to specialists such as Hippocratic AI's patient-call agents. Its newest region is Asia-Pacific, where it sells in more than 20 markets. The OfOne deal shows the model it wants to repeat, with industry-specific agents sold on top of its own speech models.

The product roadmap points the same way. Flux and Flux TTS are built for live conversations rather than recordings, and the Voice Agent API lets customers plug in their own language model. OpenAI also sells speech models, paid for by OpenAI's far larger funding rounds. Deepgram's pitch against it is speed, cost, and the option to run everything privately.

Stephenson said in 2025 that he aimed to take Deepgram public within two to three years, which points to 2027 or 2028. The main risks are price pressure and bigger rivals. Open-source models such as Whisper have pushed basic transcription costs toward zero, and Google, Microsoft, Amazon, and ElevenLabs all sell voice AI with far more capital behind them. The Series C buys time to compete on agents, where the margins are higher, before transcription alone stops paying the bills.

Deepgram has raised about $229 million, all of it equity. That includes six funded rounds from a $120,000 Y Combinator investment in 2016 to a Series C that the company announced at $130 million and its SEC filing shows at $143.2 million. Deepgram itself has described its total as more than $215 million, using the announced figure.

Deepgram's lead investors are AVP, Madrona, Tiger Global, Wing VC, and Y Combinator. Other backers include Alkeon, BlackRock-managed funds, Nvidia, In-Q-Tel, Citi Ventures, SAP, Twilio, ServiceNow Ventures, EDBI, Princeville Capital, Alumni Ventures, and the University of Michigan, Columbia University, and Stanford University.

Deepgram was valued at $1.3 billion post-money in its January 2026 Series C. It is the only valuation the company has disclosed. The August 2026 extension from EDBI did not come with a new price.

No. Deepgram is private and has not filed for an IPO. Its CEO has said he aims to take the company public around 2027 or 2028.

Deepgram is owned by its founders, employees, and investors. Co-founder Scott Stephenson is CEO. The largest outside holders are likely the firms that led its rounds, AVP, Madrona, Tiger Global, and Wing VC, along with Alkeon, though exact stakes are not public.