Hark raises $700M to build a universal AI interface

An outsized funding round for a promise that remains largely secret

Startup Hark, still unknown to the general public until recently, announced a $700 million Series A funding round, an amount that is exceedingly rare at this stage of development. The information, revealed by TechCrunch in its article devoted to the deal, immediately places the company among the most ambitious bets in artificial intelligence right now. The American outlet describes a very discreet company working on a “universal AI interface” capable of operating on top of existing products and services, with initial multimodal models expected as early as this summer.

The figure alone is enough to convey the scale of the announcement. In the startup ecosystem, a Series A theoretically corresponds to a phase where the product is still being built, where product-market fit is still being validated, and where the company has not yet deployed a massive commercial apparatus. Raising $700 million at this point in the cycle amounts to short-circuiting several of the usual stages of tech financing. That says as much about Hark as it does about the state of the AI market in 2025: investors are no longer funding only products, they are funding potential positions in the future architecture of computing.

According to details reported by TechCrunch AI, Hark is not positioning itself as a simple chatbot publisher or as a new foundational model provider in direct competition with OpenAI, Anthropic, Google DeepMind, or xAI. Its ambition is instead to become an orchestration layer, an intermediary interface capable of unifying access to software, digital services and, eventually, everyday uses around the same conversational and multimodal logic. Put differently: Hark appears to be aiming less at the raw production of intelligence than at controlling the entry point through which users will mobilize that intelligence.

This positioning is crucial. Since the shockwave triggered by ChatGPT at the end of 2022, an initial battle has played out around models, their size, their training cost, their benchmark performance, and their ability to attract developers and enterprises. A second then opened around conversational assistants, followed by agents capable of executing tasks in software environments. Hark’s promise fits into this third phase: that of the unified access layer, in other words the idea that a user will no longer want to open ten separate applications to book a ticket, manage emails, edit a document, launch a marketing campaign, review finances, or control a smart home.

The bet is simple to state, but considerable in its implications. If personal computing was long structured by operating systems, then by search engines, then by smartphones and their app stores, AI could reshuffle the deck by installing a new dominant interface: a cross-functional assistant that understands intent, chooses the right tools, aggregates data, and executes actions without the user having to navigate between software silos themselves. That is precisely the strategic position Hark seems to want to occupy.

This type of proposition recalls several historical precedents. Microsoft and Apple built empires by controlling the interface between the user and the machine. Google captured access to information through search. Meta imposed social distribution layers. Apple and Google then dominated mobile access through iOS and Android. Today, with generative AI, the next choke point could be the orchestration point between human intent and software execution. Whoever owns this layer could potentially steer usage, distribute traffic, extract a toll on transactions, and impose its integration standards.

The fact that Hark still has little public visibility is not insignificant. In AI, discretion is no longer necessarily a handicap when raising money. On the contrary, it can become a signal of scarcity, especially when investors believe a project sits at the intersection of three gigantic markets: the user interface, application infrastructure, and agentic intelligence. The record rounds raised by Anthropic, xAI, or Inflection have already shown that venture capital and major funds are ready to deploy massive sums even before mass adoption, as long as the promise touches on a future market standard.

In Hark’s case, the word “universal” is obviously the most ambitious, and no doubt the riskiest. Tech history is full of attempts at universal layers that ran into platform fragmentation, diverging publisher interests, and regulatory constraints. But that is also what fuels financial enthusiasm: if such an interface truly works across existing products and services, it could become the equivalent of a next-generation browser, search engine, or OS, without having to rebuild the entire software ecosystem from scratch.

For investors, the reasoning is crystal clear. A company that became the default gateway to digital services would not merely sell a subscription. It could capture commissions, sell infrastructure, monetize premium integrations, offer enterprise packages, and above all accumulate usage data of immense strategic value. In a market already dominated by a few major platforms, the prospect of funding the next unavoidable intermediary is enough to justify checks that, just five years ago, would have seemed extravagant for a Series A.

What Hark says it is building, and why this promise is so attractive

According to information published by TechCrunch, Hark intends to offer a platform capable of interfacing with existing products and services, relying on in-house multimodal models whose first deployments are announced for the summer. The company is therefore not limited to a textual overlay. The term multimodal implies understanding and generation across several formats: text, image, audio, and potentially video or graphical interfaces. This capability is essential if the goal is to become a general interface for contemporary computing.

A truly universal interface cannot be content with answering questions. It must understand an email, read a dashboard, analyze a screenshot, listen to a voice instruction, generate a structured response, and then act in third-party software. It must also manage context, permissions, user preferences, business constraints, and potential conflicts between applications. That implies an architecture far more complex than a simple conversational assistant.

Hark’s project is part of a broader trend: the idea that AI will not be just a generation tool, but a meta-layer of interaction with the whole digital world. Over the past year, several signals have converged. OpenAI is pushing ChatGPT as a work and research interface, with memory, connectors, document analysis, and browsing capabilities. Microsoft is integrating Copilot into Windows, Office, GitHub, and Dynamics in order to make the assistant a cross-functional entry point within its software empire. Google is trying to make Gemini a common layer across search, Android, Workspace, and its cloud services. Apple, for its part, is moving more cautiously but is seeking to inject AI into the system itself, which amounts to reinforcing the OS as an intelligent interface.

Hark nevertheless stands out through its angle. Where the giants start from already established platforms and graft AI onto them, the startup claims to be directly building a unified access layer designed from the outset for orchestration. That is both its strength and its weakness. Its strength, because it is not trapped by product legacy or an existing business model. Its weakness, because it controls neither a dominant operating system, nor a global office suite, nor a search engine, nor a mass social network. To exist, it will therefore have to convince users, developers, enterprises, and service providers alike to open their doors to it.

The promise of a universal interface appeals to investors for a structural reason: it responds to fragmentation fatigue. Today, even advanced users juggle messaging platforms, CRMs, creative tools, collaborative suites, browsers, payment platforms, HR services, business applications, and document repositories. Generative AI has shown that well-interpreted natural language can serve as a simplification layer. The dream is therefore this: instead of learning each piece of software, the user expresses an intent, and the system takes care of translating that intent into coordinated actions.

In the enterprise world, this scenario is particularly attractive. Large organizations spend billions every year on integration, training, support, and software maintenance. An AI layer capable of unifying access to existing tools could reduce adoption friction, accelerate certain processes, and make complex software environments more usable. If Hark manages to offer robust interoperability, security guarantees, and credible access governance, the commercial opportunity could be immense.

The announced timeline also deserves attention. Initial multimodal models “this summer” means Hark wants to quickly demonstrate that it is not just a financing story. In AI, media time is short and credibility is quickly measured by the ability to release a product, even a limited one. Investors have therefore probably funded not only a vision, but also a technological roadmap deemed advanced enough to justify accelerated scaling: hiring researchers, compute costs, data acquisition, distribution agreements, security, compliance, and go-to-market capacity.

The $700 million amount must also be read in light of the current costs of multimodal AI. Training or fine-tuning competitive models, building inference pipelines, ensuring low latency, negotiating access to GPUs, funding top-tier research and product teams: all of this burns capital at an unprecedented speed. A Series A of this size is not just a sign of confidence; it is often a necessity if one wants to compete from the outset with players backed by Microsoft, Amazon, Google, or major sovereign funds.

That leaves the most delicate question: what exactly does Hark mean by a universal interface? Is it a mass-market personal assistant? An API layer for developers? A unified work environment for enterprises? An agent capable of acting on the web and in software? A future conversational operating system? The vagueness may be strategic, but it cannot last forever. The broader the promise, the higher the expectations, and the more immediate the comparisons with the sector’s giants will be.

After chatbots, the battle is shifting toward agents, AI OSes, and the orchestration layer

To understand why investors can bet $700 million on a still discreet company, Hark must be placed within the current sequence of generative AI. The first phase, between late 2022 and 2023, was dominated by the surprise effect of chatbots and content generators. At that time, the market mainly rewarded the perceived quality of models, the virality of conversational interfaces, and the ability to convert curiosity into subscriptions.

The second phase, which began in 2024, focused on agents. The challenge was no longer just to respond, but to act: fill out a form, book a service, analyze a database, produce a report, fix code, or automate an entire business sequence. This evolution shifted attention toward memory, tools, connectors, planning, and reliability. The most promising products were no longer those that “spoke well,” but those that integrated effectively into real workflows.

The third phase, the one taking shape today, concerns control of the interface. If several models become good enough, and if agents already know how to use tools, then the central question becomes: where will the user express their intent? In what environment will they entrust their data, preferences, habits, permissions, and tasks? Who will decide which service is called, in what order, with what priority, and according to what commercial logic?

This is where Hark enters the picture. Its bet resembles that of a future intelligent browser, an agentic operating system, and an application integration layer combined into a single product. It is no accident that the term “universal” is used. It echoes a longstanding ambition in computing: to hide technical complexity behind a coherent and natural interface. Graphical interfaces simplified access to the machine. Search engines simplified access to the web. Smartphones simplified access to mobile services. AI could simplify access to digital action itself.

Competition is nevertheless already here, even if it does not always take the same form. OpenAI is seeking to make ChatGPT a personal and professional hub, with persistent memory, service connectors, analysis capabilities, and multimodal interactions. Microsoft is advancing an integrated strategy in which Copilot becomes the common layer across system, productivity, and cloud. Google has a unique advantage with Android, Chrome, Search, and Workspace, meaning several of the most-used entry points in the world. Apple can, if it chooses, impose an AI layer deeply embedded in iOS, macOS, and its hardware ecosystem. Meta is trying to install its assistant within its social apps, where it already benefits from massive usage frequency.

Alongside these giants, other startups are also trying to position themselves on adjacent building blocks: tool aggregation, proactive assistants, browser agents, voice interfaces, task orchestration, vertical copilots. Hark will therefore have to prove that it is not merely assembling existing trends under an appealing wrapper. Its advantage could come from more radical execution, better neutrality toward platforms, or an ability to become the Switzerland of application AI, where giants remain tempted to favor their own technology stack.

The parallel with web browsers is instructive. A browser does not need to own the sites it displays in order to become an unavoidable gateway. It only needs to offer the best access experience. In the same way, a universal AI interface would not need to own the applications it orchestrates, provided it delivers a superior experience in understanding, speed, personalization, and reliability. If Hark is aiming for this position, it is in fact trying to capture a meta-platform function.

But the history of platforms also shows that neutrality is difficult to preserve. As soon as an intermediary becomes powerful, it is tempted to steer distribution, prioritize certain partners, monetize access, or close certain interfaces. Software publishers, for their part, may fear being disintermediated. If the user goes through Hark to interact with a service, the service’s brand partially fades behind Hark’s interface. That is precisely what made search engines, app stores, and marketplaces so powerful, and so controversial.

The market is therefore rewarding Hark not because it has already won, but because it is positioning itself at a central tension point of the next digital decade. Funds are betting on the idea that after the race for models, the real rent could lie in the coordination layer between models, software, and users. If that hypothesis is correct, then $700 million does not represent an extravagance, but an entry ticket into a battle where the potential winners will be counted on one hand.

Why investors are paying so much for a still hypothetical strategic position

A $700 million Series A inevitably raises a question of financial discipline. How can such a sum be justified for a company still scarcely documented publicly? The answer lies in the specific logic of today’s AI: investors are no longer valuing only current revenue, but the probability of occupying a systemic control point in the future value chain.

The first factor is opportunity cost. The best AI talent is scarce, expensive, and often already captured by major labs or hyperscalers. Attracting a team capable of developing multimodal models, agent systems, robust connectors, enterprise security, and a mass-market interface requires considerable payroll. Recruitment packages for senior researchers, systems engineers, GPU optimization specialists, or security experts can reach levels once reserved for finance. A startup that wants to move fast must secure enough funding to recruit without constraint for several years.

The second factor is infrastructure. Multimodal AI is hungry for compute, storage, bandwidth, and software optimization. Even with favorable cloud agreements, training costs and especially large-scale inference costs can become dizzying. A universal interface, if it quickly finds its market, will have to process massive volumes of requests, maintain low latency, offer high availability, and manage persistent user contexts. In other words, it will have to behave like a global platform from its first serious iterations.

The third factor is the strategic window. In platform markets, timing matters almost as much as product quality. Investors know that if a startup wants to claim it can become an access standard, it must establish itself before existing major platforms have locked in usage. Yet OpenAI, Microsoft, Google, Apple, and Meta are all moving, at their own pace, toward integrated assistance experiences. Hark therefore needs the means to deploy quickly, forge partnerships, attract developers, and create enough usage habits to exist before the market consolidates.

The fourth factor is the very structure of venture capital in AI. Over the past two years, funds have accepted very large rounds and high valuations for projects perceived as “category-defining.” We have seen this with the billions raised by Anthropic, with xAI’s financings, and with the massive investments made around infrastructure providers such as CoreWeave. The reasoning is binary: if the company becomes a standard, the potential return is colossal; if it fails, the loss is absorbed in a portfolio where a few winners pay for everything else.

Hark’s funding round also reflects a shift in the center of gravity of investment. In 2023, much capital was concentrated on foundational models themselves. In 2024 and 2025, investors are looking more closely at the distribution and orchestration layers. Models are becoming more numerous, more interchangeable on certain tasks, and sometimes more accessible through open-source or commercial APIs. In this context, competitive advantage shifts toward user experience, integration, and the ability to become the reference interface.

For funds, Hark thus represents a bet on the next scarcity. Models, over time, could become more commoditized than expected. By contrast, an interface that concentrates attention, habits, contextual data, and action permissions remains extraordinarily difficult to dislodge. That is what made Google strong in search, Apple in premium mobile, Microsoft in enterprise office software, or Amazon in online commerce. Once the gateway is installed, the rest of the ecosystem organizes itself around it.

This enthusiasm should nevertheless be tempered. A gigantic funding round also creates gigantic pressure. Hark will have to hit credible milestones very quickly: product demonstration, execution quality, security, initial adoption, revenue or at minimum strong usage signals. In AI, narratives turn quickly when technical reality does not follow. Highly publicized startups have already shown that it is not enough to promise revolutionary agents or interfaces; concrete problems of reliability, cost, and experience must be solved.

Regulatory risk is not negligible either. A universal interface that aggregates services, handles personal data, makes orchestration decisions, and acts on behalf of the user sits at the intersection of several compliance regimes: data protection, algorithmic accountability, cybersecurity, consent, competition. In Europe, the AI Act, the GDPR, the DMA, and the DSA already form a complex environment for any company seeking to become a major intermediary. To appeal to the European market, Hark will have to demonstrate that it can combine rapid innovation with robust compliance.

What this changes for companies, software vendors, and the French-speaking market

For French and European companies, the emergence of a player like Hark is far from anecdotal. The debate around AI has often focused on the models themselves, on compute sovereignty, or on comparisons between American and European solutions. But the interface battle is just as strategic. A universal access layer can redefine how employees use software, how customers interact with services, and how value is distributed among software vendors, integrators, and infrastructure providers.

In the French context, where many organizations combine legacy tools, specialized business software, international cloud suites, and strong regulatory requirements, the promise of a unified interface is appealing. Large groups, banks, insurers, industrial companies, and public administrations face high application complexity. An AI layer capable of navigating among these environments could improve productivity and reduce certain friction costs. But it also immediately raises governance questions: where does the data reside? Who controls activity logs? How are permissions managed? What guarantees exist regarding action traceability?

For French software vendors, especially B2B SaaS players, the arrival of a universal interface is both an opportunity and a threat. An opportunity, because a startup like Hark could bring them a new usage and distribution channel by making their features more accessible through natural language. A threat, because if the end user spends less and less time in the software’s native interface, the brand relationship and product differentiation risk eroding. Software then becomes a back-end capability, less visible and potentially more substitutable.

This partial disintermediation recalls what happened in other sectors. In travel, platforms captured the customer relationship to the detriment of many suppliers. In commerce, marketplaces absorbed part of brands’ visibility. In media, search engines and social networks controlled access to audiences. If AI becomes the new primary interface, software vendors will have to ask themselves how to remain visible, monetizable, and differentiated when the user no longer really “sees” their application.

For integrators and digital services companies in France, however, the movement could open up a significant field of work. Deploying a universal AI interface in complex environments will require consulting, integration, customization, security, and change management. Companies capable of connecting these new agentic layers to ERPs, CRMs, document management systems, HR tools, or industrial systems could benefit from new demand. The French-speaking market, highly structured around the digital transformation of large enterprises and the public sector, could become fertile ground for this type of deployment, provided compliance guarantees are sufficient.

The issue also touches on European digital sovereignty. If the next gateway to software and services is controlled by a handful of American players, Europe risks once again becoming dependent on a critical layer without a real local alternative. The debate is not limited to model training; it also concerns the interface that collects users’ intentions, contexts, and action flows. In that sense, Hark illustrates a broader issue: the value of AI will not reside only in models, but in the gateways that organize everyday usage.

For French-speaking consumers, the impact could translate into a radical simplification of certain digital uses: personal organization, administrative procedures, shopping, document management, multichannel communication, voice assistance. But this simplification comes with a potential price: increased concentration of data and intermediation power. A universal interface knows what the user is looking for, what they write, what they plan, what they buy, which tools they use, and how they arbitrate their choices. The quality of the experience will therefore have to be weighed against issues of privacy, competition, and freedom of choice.

In France, where sensitivity to data protection issues is stronger than in the United States, Hark and its competitors will likely have to adapt their messaging. The local market will not be satisfied with a promise of fluidity. It will demand commitments on hosting, governance, explainability of actions, and the ability to audit processes. In regulated sectors such as healthcare, finance, legal, or the public sector, model performance alone will not be enough. The interface will have to be provable, controllable, and integrable into strict compliance frameworks.

Beyond the announcement effect, Hark may foreshadow the next dominant architecture

The $700 million funding round announced by Hark obviously guarantees neither commercial success nor a lasting technological breakthrough. But it acts as a strong signal about the direction the industry is taking. The era when AI amounted to “asking a chatbot a question” already seems transitional. The next challenge is deeper: whether AI will remain one feature among others, or become the primary interaction layer between humans and the software universe.

If Hark succeeds, it could help crystallize a new computing model. In this model, applications do not disappear, but they move into the background. The user no longer learns a dozen interfaces; they converse with a single layer that understands their context, chooses the right tools, executes tasks, and returns results. Software then becomes “composable services” mobilized on demand. Value moves upward toward the orchestrator, the one that owns the continuous relationship with the user.

This perspective explains the intensity of the competition. Tech giants know that the current battle is not only about the best AI, but about the best access point to AI. A company that wins this position can redistribute traffic, impose its integration standards, select the partners that are highlighted, and capture a growing share of the value created downstream. That is why announcements like Hark’s resonate far beyond the world of venture capital alone: they concern the future hierarchy of the digital world.

Even so, several major obstacles remain. The first is agentic reliability. A universal interface cannot afford approximation when it acts on real software, customer accounts, or sensitive data. The second is interoperability. Promises of universality often run into incomplete APIs, closed interfaces, changing access conditions, and the strategic resistance of platforms. The third is trust. Entrusting a single layer with one’s intentions, histories, and permissions requires a level of credibility that is still rare in consumer AI.

The reaction of established platforms must also be taken into account. If Hark begins to emerge as a powerful intermediary, it is likely that some players will limit its access or develop more integrated in-house alternatives. The history of the web, mobile, and cloud shows that intermediary layers only prosper durably if they find a balance between usefulness to the ecosystem and perceived threat by incumbents. A startup can grow quickly, but it must avoid triggering too early a coordinated shutdown of the access on which its product depends.

From this perspective, Hark’s ability to position itself as an interface that enhances applications rather than replacing them will be decisive. If it presents itself as a neutral layer of simplification and distribution, it can attract partners and developers. If it appears as a machine for absorbing the customer relationship and commoditizing underlying software, it will quickly encounter resistance. The very term “universal” contains this ambiguity: it evokes both openness and centralization.

For the French-speaking market, the issue goes beyond Hark alone. This funding round is a reminder that the next major AI battle may not be fought only in model labs, but in interfaces, integration standards, and usage habits. European companies that fail to anticipate this shift risk finding themselves dependent on access layers designed elsewhere, with economic and technical rules they do not control. Conversely, those that understand the logic of orchestration early will be better able to negotiate their place, preserve their customer relationship, and adapt their products to a world where the user will first speak to an AI before entering, perhaps, the application itself.

Hark’s bet is therefore less that of a new chatbot than that of a silent overhaul of everyday computing. If the first multimodal models announced for the summer confirm the startup’s ability to make understanding, action, and interoperability work together, the company could become one of the most closely watched laboratories of this transition. And if the idea of a universal interface takes hold, the decade now opening could see the emergence of a new dominant layer, situated between the user and applications, capable of profoundly redistributing the value of software, cloud, and digital services, including in Europe’s most heavily regulated markets.