Brett Adcock Just Raised $700M for a 70-Person Startup. The AI Device Race Is Back.
Hark raised $700M at $6B with 70 employees. Adcock's startup builds AI models and hardware together from day one. Inside the bet, the Nvidia-AMD investor thesis, and why this isn't Humane.
Three months after Hark announced its existence in March 2026, the company raised $700 million at a $6 billion valuation. The round included Nvidia, AMD Ventures, Qualcomm Ventures, and Intel Capital — the four largest semiconductor companies with significant AI hardware stakes — alongside Salesforce Ventures, ARK Invest, Brookfield, Greycroft, Align Ventures, Prime Movers Lab, and Tamarack Global. Hark has 70 employees and no shipping product.
The implied valuation-per-employee is $85.7 million. The closest recent comparable: OpenAI was valued at approximately $29 billion with roughly 500 employees in early 2023, implying $58 million per employee — before ChatGPT had become a mass-market product and before the company had billions in annual revenue. Hark is being valued higher per capita than OpenAI was before its consumer inflection point, with no product and no revenue.
This is not irrationality. It is a specific kind of hardware venture bet — the kind that Bloomberg described as Adcock's third act — premised on founder track record, semiconductor industry strategic alignment, and a genuine technology thesis about why previous personal AI devices failed. The question is not whether the bet is rational. The question is whether Hark's specific combination of advantages is sufficient to do something that Humane, Rabbit, and several other well-funded teams could not: build a personal AI device that people actually want to carry instead of their phone.
Brett Adcock's Third Act
The case for Hark begins and ends with Brett Adcock's track record, which is genuinely unusual for a hardware founder at this stage.
Adcock founded Archer Aviation in 2018, an electric vertical takeoff and landing aircraft startup that went public via SPAC in 2021. Archer has since received FAA Part 135 air carrier certification, signed commercial agreements with United Airlines, and produced aircraft that fly. The trajectory has been slower than early projections — this is invariant for hardware companies navigating aviation regulation — but Archer is not dead. It is building aircraft that fly.
In 2022, Adcock founded Figure, a humanoid robotics company. Within two years, Figure raised over $225 million, signed a commercial deployment agreement with BMW for its manufacturing operations, and delivered Figure 01 — a robot that can perform real production tasks at a real facility. The 2026 humanoid robotics renaissance named Figure among the small cohort of companies actually shipping hardware to production customers rather than demo stages.
What Adcock has demonstrated across both companies is not that he builds cheap or fast. Archer and Figure have both been delayed relative to initial investor timelines. What he has demonstrated is that he builds things that actually work. Both companies are doing what they said they would do, on timelines that are longer than projected but not abandoned. For a venture as speculative as Hark — a company with no product, building in a category that has failed multiple times — that track record is genuinely meaningful signal. Most hardware founders with two prior companies at this stage have one failure on the record. Adcock does not.
What "Universal AI Interface" Actually Means
Hark's product thesis is simultaneously ambitious and deliberately underspecified. The company describes itself as building "highly intelligent, multimodal AI systems and native hardware devices designed to serve as a universal interface between humans and machines." The system is proactive, personalized, and capable of interacting through speech, text, vision, and persistent memory.
The closest frame of reference is the concept that drove Humane and Rabbit: a personal AI device that functions as an ambient computing layer above the smartphone. But Hark's approach differs from both failures in one critical structural dimension: they are building foundation models, software systems, hardware, and interface together from the ground up, rather than layering AI capabilities onto existing hardware or purchasing model access from third parties.
This is a substantially harder technical bet. It is also a substantially more defensible one, and the distinction matters.
The fundamental failure mode of both Humane AI Pin and Rabbit R1 was that their AI capabilities were constrained by third-party model access. Their devices were essentially wrappers around GPT or other foundation models, giving them no ability to improve core intelligence over time. When the underlying models underperformed — especially in the real-world, always-on context that ambient hardware requires — the hardware makers had no recourse. They were entirely dependent on OpenAI's or Anthropic's roadmap and pricing.
Hark is betting that the tight integration between proprietary hardware and proprietary foundation model — optimized together from day one — produces a qualitatively different user experience than the wrapper approach. The logic is the same logic Apple used when designing the iPhone: not attaching software to generic hardware, but designing hardware and software to enable each other's capabilities.
The hiring evidence is explicit about the hardware design ambition. The company recruited Abidur Chowdhury, the lead designer for the iPhone at Apple and most recently the designer who introduced the iPhone Air during Apple's 2026 keynote. You do not hire the person who designed the iPhone Air to iterate on an existing device category. You hire them when you believe you are building a new one.
Why the Semiconductor Industry All Signed Up Simultaneously
The most revealing element of Hark's investor list is not the total amount or the valuation. It is that Nvidia, AMD Ventures, Qualcomm Ventures, and Intel Capital all participated simultaneously. These companies compete for AI chip market share in nearly every segment. They do not typically co-invest.
Their simultaneous participation tells you something important about what each company believes: that Hark represents a potential new computing platform category, and that being absent from the founding capitalization is a larger strategic risk than co-investing alongside competitors.
| Investor | Strategic Interest in Hark |
|---|---|
| Nvidia | Custom AI silicon for personal devices — Hark already runs Nvidia B200s in its data center |
| AMD | Counter-position to Nvidia's dominance; access to next-gen personal AI hardware specs |
| Qualcomm Ventures | Mobile and edge AI chip positioning; Snapdragon has dominated mobile compute for a decade |
| Intel Capital | Edge compute and AI PC positioning as Intel attempts to recapture relevance in AI silicon |
| Salesforce Ventures | Enterprise ambient AI interface — if Hark becomes the workplace AI layer |
| ARK Invest | Long-duration technology platform bet aligned with transformative technology thesis |
| Brookfield | Capital deployment into long-cycle technology infrastructure |
This investor composition is the semiconductor industry's collective hedge against a new platform transition. The logic mirrors the early smartphone era precisely: every major chipmaker competed for Apple and Samsung design wins because whoever won those relationships defined the hardware requirements for billions of devices over the following decade. The company that wins the personal AI hardware category will define the chip requirements for the next generation of personal compute devices.
A relatively small financial commitment to Hark is low-cost insurance against being locked out of specifications that matter. None of these semiconductor companies are expecting the typical venture fund return profile. They are buying strategic optionality.
The AI Hardware Graveyard: What Hark Is Not Doing
The failure of Humane AI Pin and Rabbit R1 is worth understanding in specific detail, because Hark's strategy appears designed in direct response to each failure mode.
Humane raised $240 million and launched the AI Pin in 2024 at $699 plus a $24 per month subscription. It was discontinued less than a year after launch. The failure was multi-layered: the form factor required users to project a laser display onto their palm, battery life was two to three hours, and the AI capabilities — entirely dependent on GPT-4 — were slower and less capable than simply using a smartphone. The device offered no compelling reason to exist. It did fewer things than a phone, did them worse, and cost more.
Rabbit R1 failed along similar dimensions. Its core technology — the "Large Action Model" that supposedly learned to use apps autonomously — appeared on analysis to be running GPT-based automation that any standard smartphone app could replicate. The hardware became a liability rather than an advantage.
Hark's structural differences from both failures are threefold:
First, proprietary foundation models. Hark is training its own multimodal models, scheduled for release in summer 2026. The intelligence layer is not dependent on third-party access and can be continuously improved based on Hark's specific hardware and use-case requirements. This is the most critical structural difference from its predecessors.
Second, hardware designed for the model. The existing AI hardware failures placed AI capabilities into hardware designed for different purposes. Hark is designing hardware and software simultaneously. When each is built to enable the other, the result is qualitatively different from attaching a voice interface to a screen or a projector.
Third, no premature hardware commitment. Hark has not announced a hardware ship date. The plan is models in summer 2026 — a concrete deliverable that can be evaluated — followed by hardware on an undefined timeline. This sequencing avoids the mistake that damaged both Humane and Rabbit: committing to hardware delivery dates before the AI capabilities that would make the hardware compelling were ready.
The Valuation Math and the Structural Risk
A $6 billion valuation for a company with 70 employees and no product is a bet on three simultaneous variables: founder track record, technology thesis, and market timing. The first is better than average for hardware venture. The second is internally coherent but extremely difficult to execute. The third is genuinely uncertain.
The AI venture capital barbell that has emerged in 2026 identifies two defensible categories of AI investment: foundational infrastructure at hyperscale, and highly specific vertical applications with proprietary data moats. Hark doesn't fit neatly into either category. It is attempting to be foundational — a new computing platform — which is the highest-risk position in the barbell and also the highest potential return if successful.
The technology thesis requires Hark to build foundation models that are competitive with frontier labs staffed by thousands of researchers and backed by billions in compute budgets, while simultaneously designing hardware that users will prefer to interact with over their existing smartphone. Both parts of the thesis are individually difficult. Combined, they are among the most technically ambitious bets being made anywhere in 2026.
Market timing concentrates the uncertainty. The personal AI device category has failed twice in recent memory, with well-funded companies and serious teams. The case for Hark is that those failures were premature — the models weren't capable enough, the hardware wasn't designed for the models, and the form factors were wrong — and that 2026-2027 represents the window when all three conditions can be met simultaneously. The countervailing case is that smartphones are already optimal personal AI interfaces for most use cases, and that adding dedicated hardware to the stack creates friction rather than reducing it.
OpenAI's own Jony Ive device — a screenless ambient AI device with a late 2026 launch window and stated 100 million unit ambition — is the most direct near-term competitive threat. If OpenAI ships a compelling ambient AI device backed by the most capable frontier models and Jony Ive's industrial design credibility, Hark's market window narrows. Hark's advantage would then need to come from its integrated models-plus-hardware approach producing meaningfully better performance than an OpenAI device running the world's best external AI. That is a high bar by any measure.
What the Summer 2026 Models Actually Prove
The most important milestone in Hark's near-term timeline is not the hardware. It is the multimodal models scheduled for release this summer.
If those models demonstrate capabilities that compare favorably with frontier models on the specific tasks personal AI hardware is designed to handle — persistent memory across sessions, multimodal reasoning in ambient contexts, proactive assistance that doesn't require explicit prompting, personalization that improves over time without explicit training — the hardware thesis becomes substantially more credible. If the models are merely competitive with open-source alternatives available from Llama or Mistral, the $6 billion valuation faces significant revision pressure.
The model release is Hark's first concrete opportunity to prove its thesis with evidence rather than founder credibility and investor alignment. It is also the moment when the team's hiring choices will either confirm or refute the talent hypothesis: can a 70-person team building foundation models from scratch produce AI capabilities that compete with organizations 10× larger?
The hardware-to-model integration question is ultimately a question about whether the personal computing interface has room for a new platform. Smartphones absorbed the camera, the music player, the map, the calendar, and dozens of other single-purpose devices by being a better general-purpose platform. Personal AI hardware is betting that the smartphone, in turn, can be partially replaced by a more intelligent ambient layer. The iPhone's lead designer is now working on what that layer might look like. The chipmakers who built the iPhone's components all have stakes in whether it works.
Why This Matters Beyond Hark
Even if Hark fails — and the base rate for ambitious hardware startups suggests most bets in this category will — the round itself tells a structural story about where AI investment is moving.
The semiconductor industry's simultaneous investment in a personal AI hardware startup signals that the major chipmakers believe a new computing platform cycle is beginning. Platform cycles are rare: mainframe to PC, PC to internet, internet to smartphone. Each created massive new chip markets. If personal AI hardware represents even a partial platform transition — ambient AI supplementing rather than replacing smartphones — the silicon opportunity is enormous.
Hark's $700 million Series A also signals the end of the AI hardware graveyard narrative that followed Humane and Rabbit. Those failures created a brief consensus that personal AI hardware was a dead category — that the smartphone would absorb AI capabilities without a new form factor being necessary. The investor list for Hark's round suggests that consensus has been revised. The question is not whether personal AI hardware is a real category; it is who executes well enough to define what that category means.
Brett Adcock has built two hardware companies that deliver working products to real customers. The third company has raised more capital in its first major financing than either predecessor raised across multiple rounds. The thesis is clear. The team is real. The models will tell us in a few months whether the technology is ready. The hardware will tell us, eventually, whether the category was real all along.
Takeaway: Hark's $700 million Series A at a $6 billion valuation is either the most justified audacious bet in recent AI hardware history, or the beginning of the most expensive lesson in why personal AI devices keep failing. Brett Adcock's track record at Archer and Figure provides meaningful signal that he builds things that work, eventually. The simultaneous investment by Nvidia, AMD, Qualcomm, and Intel signals that the semiconductor industry believes a new personal computing platform is forming and wants inside early. The summer 2026 model release is the first real test — not of the hardware, but of whether Hark's integrated approach produces AI capabilities qualitatively different from what a wrapper company could build. The $6 billion question starts getting an answer then.
Frequently Asked Questions
What is Hark and what is it building?
Hark is an AI startup founded in late 2025 by Brett Adcock — previously the founder of Figure (humanoid robotics) and Archer Aviation (electric aircraft). The company describes its mission as building 'a universal interface between humans and machines' using multimodal AI that combines speech, text, vision, and persistent memory. Unlike previous personal AI device companies like Humane and Rabbit, Hark is developing its own foundation models, software systems, hardware, and interface layer simultaneously from scratch rather than licensing AI capabilities from OpenAI or Anthropic. The company raised $700 million at a $6 billion valuation in May 2026, led by Parkway Venture Capital with participation from Nvidia, AMD Ventures, Qualcomm Ventures, Intel Capital, Salesforce Ventures, ARK Invest, and Brookfield. Hark plans to release its first multimodal AI models in summer 2026, followed by hardware devices designed specifically to work with those models. The company currently has 70 employees and operates a data center running Nvidia B200 GPUs.
Who is Brett Adcock and why does his hardware track record matter for Hark?
Brett Adcock is a serial founder who has now launched three hardware-adjacent companies. He founded Archer Aviation in 2018, an electric vertical takeoff and landing (eVTOL) aircraft company that went public via SPAC in 2021 and has since received FAA Part 135 air carrier certification with commercial agreements with United Airlines — a meaningful milestone for a deep-tech hardware startup. In 2022, Adcock founded Figure, a humanoid robotics company that raised over $225 million, signed a commercial deployment agreement with BMW, and delivered a working robot (Figure 01) capable of real manufacturing tasks. Both companies have been slower than initial projections suggested, as hardware companies typically are, but both have delivered working hardware to real customers. That track record — building things that actually work, even if delayed — is qualitatively different from the background of most AI hardware founders, who typically come from software or consumer product backgrounds without prior experience shipping complex physical devices. For Hark investors, Adcock's ability to attract serious capital, recruit hardware talent, and actually ship is the primary investment thesis.
Why did Hark raise $700M at $6B valuation with only 70 employees?
The $6 billion valuation for a 70-person company with no product implies an $85.7 million value per employee — a multiple that reflects founder premium rather than revenue multiple. Adcock invested $100 million of his own money at founding, which reduced early dilution and signaled conviction. The four semiconductor companies in the round (Nvidia, AMD, Qualcomm, Intel) are not investing on financial return expectations at this stage — they are making strategic bets on who will define the hardware specifications for personal AI devices, the same way chipmakers competed for Apple and Samsung design wins in the smartphone era. Hardware companies also require more capital per employee than software companies: building proprietary AI models requires significant compute expenditure, and designing custom hardware requires expensive engineering talent, tooling, and manufacturing partnerships. The $700 million round is not large relative to the capital intensity of the full build — it is a first major tranche of capital that will fund model development and initial hardware design, not production at scale. In hardware venture, large Series A rounds for credible founders are structurally different from equivalent rounds in software.
How is Hark different from Humane AI Pin and Rabbit R1?
Humane AI Pin and Rabbit R1 both failed for related reasons: their AI intelligence was entirely dependent on third-party model access (primarily GPT-4), their hardware form factors created friction rather than reducing it, and they offered no compelling reason to exist when smartphones already handled the same tasks more reliably. Hark's thesis directly addresses each failure mode. First, Hark is building its own foundation models rather than wrapping external APIs — this means the intelligence layer is not dependent on OpenAI or Anthropic and can be optimized specifically for the personal AI use cases Hark's hardware is designed to support. Second, Hark is designing hardware and models simultaneously, the way the original iPhone was designed — not by attaching AI capabilities to a generic form factor, but by designing each to enable the other. Third, Hark has not committed to a specific hardware ship date, avoiding the mistake of announcing hardware before the AI capabilities were ready. The recruitment of Abidur Chowdhury, the lead designer for the iPhone and the designer who introduced the iPhone Air, signals that the hardware design ambition is serious. Whether these differences translate to a product users actually want is unproven, but the structural mistakes of prior AI hardware attempts are explicitly being avoided.
Why did Nvidia, AMD, Qualcomm, and Intel all invest in Hark simultaneously?
The simultaneous participation of all four major semiconductor companies in a single startup round is unusual and strategically revealing. These companies compete for AI chip market share in almost every segment and do not typically co-invest. Their shared participation signals that each believes Hark represents a potential new computing platform category — personal AI hardware — that will require specialized silicon, and that missing it represents a larger risk than co-investing alongside competitors. The logic parallels the smartphone era: every major chipmaker competed for Apple and Samsung design wins because whoever won those relationships defined the hardware requirements for billions of devices. Nvidia's investment is particularly notable because Hark is already running Nvidia B200 GPUs in its data center, suggesting an existing relationship that Nvidia wants to extend into whatever hardware Hark eventually ships. AMD, Qualcomm, and Intel are hedging against that relationship — ensuring they have a seat at the table when Hark specifies its hardware requirements. For all four, a relatively small financial investment in a speculative startup is low cost insurance against being locked out of a new platform category.
When will Hark release its product and what is the timeline?
Hark's stated roadmap has two distinct phases. The first phase is the release of proprietary multimodal AI models, targeted for summer 2026. These models will combine speech, text, and vision capabilities with persistent memory, designed to function as a 'personal AI platform' that works with existing products and services — phones, computers, apps — before dedicated hardware is ready. The second phase is hardware devices designed specifically to run Hark's models and interface with the world in ways a smartphone cannot. Hark has not committed to a specific date for the hardware release, which is the right call given how consistently AI hardware companies have damaged themselves by committing to ship dates before capabilities were ready. The summer 2026 model release is the critical near-term milestone: it will be Hark's first opportunity to demonstrate that a 70-person team building foundation models from scratch can produce AI capabilities that compete with frontier labs employing thousands of researchers. If the models are compelling, the hardware thesis becomes substantially more credible. If they are merely competitive with open-source alternatives, the $6 billion valuation requires significant reappraisal.