A Billion-Dollar Signal: Why These AI Funding Rounds Matter
Two specialist AI labs, Hark and Decart, have just secured a combined USD 1 billion (approx. RM4.6 billion) in AI startup funding rounds, underscoring how quickly capital is concentrating around real-time AI systems and infrastructure. Hark raised more than USD 700 million (approx. RM3.2 billion) in an oversubscribed Series A at a USD 6 billion (approx. RM27.6 billion) Series A valuation, while Decart closed a USD 300 million (approx. RM1.4 billion) round led by deep-tech investors. Instead of another wave of general-purpose chat interfaces, both companies are leaning into technically demanding domains: Hark with AI-native hardware and persistent, personalized agents; Decart with high-throughput video, world models, and next‑generation inference stacks. The message from investors is clear: the next competitive frontier will be won by startups that own the hardest parts of the stack—latency, context, and physical-world integration—rather than by yet more generalist text-only models.
Hark: Building a Universal Interface With AI-Native Hardware
Hark’s more than USD 700 million (approx. RM3.2 billion) Series A at a USD 6 billion (approx. RM27.6 billion) valuation is remarkable not just for its size, but for what it backs: a vertically integrated platform that spans models, software, and dedicated devices. Supported by investors including Nvidia, AMD Ventures, Intel Capital, Qualcomm Ventures, and others, Hark aims to create AI systems that serve as a “universal interface between humans and machines.” Instead of a generic chatbot, the company is developing agentic, multimodal models designed to remember who you are, retain long‑term context, and “speak your language” across the services you already use. Crucially, these capabilities will live on AI-native hardware specifically built for personalized, real-time AI systems. With roughly 70 employees and a new NVIDIA B200 data center for training, Hark is positioning itself as a hardware-software AI lab rather than a pure model API provider.
Decart: Real-Time World Models and High-Speed Inference
Decart’s USD 300 million (approx. RM1.4 billion) raise highlights investor appetite for world models inference and ultra‑low‑latency AI infrastructure. The vertically integrated research lab focuses on real-time video and world models, building both models and the underlying stack. Its DOS 2.0 platform delivers over 1,600 tokens per second for agentic inference—about eight times the industry average—and can stream full‑HD video at up to 100 frames per second across major cloud hardware. On top of DOS, Decart offers two model lines: Lucy, tailored for immersive, real-time experiences in gaming, e‑commerce, and advertising with sub‑30‑millisecond responses; and Oasis, targeting physical AI domains such as robotics and autonomous vehicles via physically accurate real-time simulation. With strategic ties, including Amazon as a customer, Decart is betting that the most valuable AI will be the kind that can see, predict, and act in the world in real time.
From Chatbots to Infrastructure: How Investor Priorities Are Shifting
These mega-rounds show investors rotating away from consumer-facing chatbots toward deep infrastructure and specialized research labs. Both Hark and Decart are optimized for real-time AI systems, not text-box interfaces. They are designing agentic platforms that manage continuous context, multimodal inputs, and time-critical decisions. This emphasis on vertical integration—owning models, inference stacks, and, in Hark’s case, AI-native hardware—contrasts with generalist model providers that primarily expose APIs. Capital is converging on teams that can solve technically hard, latency-sensitive problems: persistent assistants that live across devices, world models inference that can drive robots and vehicles, and video-native systems that respond in milliseconds. For founders, the takeaway is that differentiation increasingly means controlling the end-to-end stack, from silicon to software, rather than building thin applications on top of commodity models.
The New Playbook: Vertical Integration and Hardware as Moats
Hark and Decart illustrate a new AI startup funding template: secure large rounds early, then invest heavily in bespoke hardware, training infrastructure, and tightly coupled software stacks. Hark is explicitly avoiding a single-layer strategy, instead creating a universal interface that blends personalized models with dedicated devices designed for continuous, human-like interaction. Decart is doing something similar in a different domain, using DOS 2.0 as a foundation for high-speed inference that powers real-time video and physically grounded world models. In both cases, hardware isn’t an afterthought; it is a strategic moat that ensures their systems can run at the required speeds and scales. As investors double down on such vertically integrated approaches, AI startups that merely wrap existing models may find it harder to justify large rounds without a convincing story around real-time capabilities, infrastructure innovation, or hardware integration.