Hark raised more than $700M in a Series A round in May 2026 at a $6B post-money valuation. Parkway Venture Capital led the round, with participation from NVIDIA, AMD Ventures, Intel Capital, Qualcomm Ventures, Salesforce Ventures, ARK Invest, Brookfield, Greycroft, Prime Movers Lab, Align Ventures, and Tamarack Global.

Before the Series A, founder Brett Adcock committed approximately $100M of personal capital to launch the company in late 2025. AT&T later disclosed a strategic investment alongside its August 2026 connectivity and device-certification partnership.

Total cumulative funding exceeds $800M across the founder commitment and Series A.

Hark is developing a personal-intelligence platform that combines proprietary multimodal AI models, persistent user memory, a computer-use agent, and planned AI-native hardware. The system is designed to complete tasks on a user's behalf across the open web rather than only answer questions.

Its first product, Handoff, is a computer-use agent. In response to a natural-language request, Hark creates a dedicated virtual computer with a browser, file system, and terminal. Handoff navigates websites visually by moving a cursor, clicking, scrolling, and typing. Tasks include ordering food through DoorDash, comparing products on Target or Costco, booking restaurants on OpenTable, searching for candidates on LinkedIn, reserving flights and hotels, and handling returns.

Because Handoff controls a browser visually, it does not require a formal API integration with each website. It can attempt to use any site accessible to a person, but website changes, bot detection, and CAPTCHAs create ongoing reliability challenges.

The platform also uses persistent memory and personal context, connecting to a user's accounts, saved addresses, payment methods, preferences, and interaction history to reduce the information required for subsequent requests. Hark intends for the system to become proactive over time, identifying calendar constraints or fare changes and proposing actions without waiting for a prompt.

Hark is also developing consumer hardware co-designed with its models for speech, vision, and ambient interaction. Each device is expected to include direct cellular connectivity through an AT&T partnership, allowing it to operate without a phone or Wi-Fi network. Former Apple product executive Abidur Chowdhury leads design, and the team of roughly 70 people includes alumni from Apple, Tesla, and Meta AI.

Hark is pursuing a vertically integrated, direct-to-consumer model spanning proprietary AI models, cloud-hosted agent execution, persistent user memory, and first-party hardware with embedded cellular connectivity. Users sign up directly through Hark's app and web interfaces.

Likely monetization includes recurring consumer software subscriptions, usage-based charges or tier limits for computationally expensive agentic tasks, hardware sales, and recurring device or connectivity plans. Each Handoff request requires model inference, a dedicated virtual computer, browser compute, network bandwidth, and potentially minutes of runtime, making unlimited flat-rate pricing risky. Hark will likely need credits, task allowances, or tiered plans even if the consumer interface remains simple.

Hardware serves primarily as a distribution and retention channel rather than a margin center. First-generation devices may be sold near cost or through AT&T carrier financing to expand the subscription base, with recurring intelligence services generating more durable revenue than one-time device sales.

Hark's cost structure includes foundation model training on gigawatt-scale NVIDIA infrastructure, dedicated virtual computers for each agent task, cellular consumer electronics development and manufacturing, and carrier certification and connectivity. The multi-year NVIDIA partnership covering Vera Rubin capacity and the AT&T device-distribution relationship indicate the capital required by this model. More than $800M of cumulative funding provides runway, but Hark must establish consumer willingness to pay and task-level unit economics before the integrated model can sustain itself.

Hark sits at the intersection of frontier AI models, computer-use agents, personal assistants, and AI-native hardware. It competes with incumbents that have large distribution networks and startups focused on individual layers of the stack.

OpenAI is the closest strategic analogue. Its consumer agent stack combines visual browsing, text browsing, terminals, authenticated sessions, and scheduled tasks inside ChatGPT, while the company is investing in custom chips and longstanding AI-native device manufacturing. OpenAI's installed user base and brand give it a distribution advantage over Hark, which must acquire users one by one. However, OpenAI CEO Sam Altman revealed in a court case in September 2026 that its long-awaited consumer device, designed in partnership with Jonny Ive's iO Products team, "is not coming anytime soon," possibly opening a window for Hark's strategy.

Google can embed Gemini across Android, Chrome, Search, Gmail, Calendar, Maps, and its device portfolio. Its 2026 Android strategy recasts the operating system as an intelligence layer, with Gemini automating tasks and remembering real-world context. Apple controls a widely used consumer device surface and is integrating Siri AI across iPhone, Mac, Watch, and Vision Pro with on-device semantic indexing and App Intents. Both can bundle equivalent functionality at zero marginal cost to users.

Anthropic is a direct technical competitor. Claude can interpret screenshots and operate mice and keyboards, and the acquisition of Vercept in early 2026 added to that capability. Anthropic's approach focuses on enterprise and developer deployment rather than proprietary consumer hardware, allowing partners to embed computer use into controlled workflows.

Below the frontier-model layer, startups including Browser Use, Polar, Strawberry, and Aside combine off-the-shelf models with browser orchestration, retries, and session management. Their cost structures are lighter because they do not train large base models and allow customers to select whichever frontier model is currently best. Hark argues that jointly training the model and execution harness produces better reliability, while an open orchestration layer may improve faster and avoid model lock-in.

Amazon Alexa+ can order takeout, make reservations, book rides, shop, and control smart homes, and it is bundled at no additional cost for Prime members. Meta AI powers a multimodal assistant across messaging, social products, and Ray-Ban smart glasses, which have retail distribution through EssilorLuxottica, prices starting at $299, and millions of daily users.

Earlier dedicated AI devices such as Humane's AI Pin and Rabbit R1 struggled to establish durable demand, with Humane ultimately acquired by HP for $116M after weak adoption. Focused products such as Plaud's recording hardware and Limitless's Pendant have fared better by targeting specific use cases. Hark has more capital and vertical integration than the first generation of AI gadgets, but it faces the same question: whether a new device delivers enough recurring utility to complement or displace a phone.

Hark's TAM expansion depends on extending a single browser-agent product across consumer subscriptions, enterprise automation, hardware, and connectivity.

Handoff currently covers commerce, food delivery, reservations, recruiting, research, and travel. Each category gives Hark access to transaction workflows without requiring every service to provide an API. As persistent memory matures, Hark could move from discrete prompts to recurring workflows such as household purchasing, appointment scheduling, bill administration, and trip planning, extending the browser tool into life management.

Handoff could address business workflows in recruiting, research, purchasing, travel administration, and back-office operations that depend on fragmented websites or legacy software without reliable APIs. An API or managed agent runtime with administrator controls, approval checkpoints, and audit trails would allow Hark to monetize its model and browser infrastructure at higher utilization while collecting task data relevant to the consumer product.

Pricing could be tied to labor hours or completed outcomes rather than seat counts, giving small businesses and independent professionals a direct comparison between agent costs and human labor.

Hark's AT&T partnership provides access to carrier retail, installment financing, bundled data plans, and account billing for its forthcoming cellular devices. Untethered connectivity would allow Hark hardware to operate independently of a smartphone in settings where phones are inconvenient, including ambient home use, hands-free travel, and accessibility scenarios.

International expansion could use additional carriers and device distributors. A software-first rollout in new markets would let Hark assess demand and collect localization feedback before incurring inventory and certification costs, while carrier relationships could reduce the need to build device distribution independently.

Web-platform resistance: Retailers and platforms can identify, rate-limit, or block browser agents through bot detection, CAPTCHAs, and terms-of-service enforcement, as shown by Amazon's cease-and-desist challenge to Perplexity's Comet agent, which means Hark's open-web automation strategy may require negotiated merchant access rather than unrestricted GUI control.

Vertical-integration capital intensity: Training foundation models on gigawatt-scale infrastructure, operating dedicated virtual computers for every agent task, and manufacturing cellular consumer electronics combine capital-intensive businesses with different technical and economic cycles, any one of which could produce unfavorable unit economics or inventory risk before the integrated ecosystem reaches sufficient scale.

Delegated-action liability: A system authorized to log into user accounts, make purchases, send messages, and store persistent memory of browsing, payment, location, and behavioral data creates a concentrated privacy and security target, where a single breach or pattern of mistaken autonomous actions could undermine the consumer trust on which the value proposition depends.