Record-Scale AI Startup Funding Rounds Redefine Early-Stage Expectations
AI startup funding rounds are breaking previous norms as investors pivot toward infrastructure-layer bets with unprecedented ambition. Hark’s latest raise, an oversubscribed Series A exceeding USD 700 million (approx. RM3.2 billion) at a USD 6 billion (approx. RM27.6 billion) Series A valuation, underscores how capital is concentrating in a small set of platforms promising end‑to‑end control of the AI stack. Decart’s USD 300 million (approx. RM1.4 billion) financing round further illustrates the new bar for what an early-stage infrastructure company can command. These figures, historically associated with late-stage unicorns, are now appearing at the first institutional rounds. The strategic logic is clear: foundational AI hardware development, custom inference stacks, and vertically integrated platforms require enormous up‑front investment in compute, talent, and proprietary infrastructure. Investors appear willing to underwrite that capex to secure exposure to what they view as long‑term, infrastructure‑grade assets rather than short‑lived application-layer products.
Hark Bets on AI-Native Hardware as a Universal Human–Machine Interface
Hark is positioning itself as a vertically integrated AI company that fuses models, software, and AI-native hardware into a single platform. The company’s Series A, totaling more than USD 700 million (approx. RM3.2 billion) at a USD 6 billion (approx. RM27.6 billion) Series A valuation, is explicitly aimed at building hardware products that act as a “universal interface between humans and machines.” Rather than confining interaction to chatbots, Hark envisions agentic, multimodal systems that remember users, maintain long-term context, and operate across existing products and services. With roughly 70 employees and a new NVIDIA B200 data center to support training, Hark plans to debut its first models later this summer. Backing from a roster of major chipmakers and strategic investors signals confidence that AI hardware development—tailored around personalized, always‑on assistants—will form a defensible moat beyond generic, cloud‑only software offerings.
Decart Targets Real-Time Inference With World Models and DOS 2.0
Decart exemplifies the parallel funding thesis around real-time inference technology. The company, a vertically integrated AI research lab focused on real-time video and world models, has raised USD 300 million (approx. RM1.4 billion) in a round led by Radical Ventures with participation from major technology investors and strategic partners. Its core product, DOS 2.0, is engineered for ultra-fast agentic inference, reportedly delivering over 1,600 tokens per second—around eight times the industry average—and supporting full‑HD video at up to 100 frames per second across NVIDIA, Google, and Amazon hardware. Built on this foundation, Decart’s Lucy and Oasis model lines target distinct, high-value domains: sub‑30‑millisecond immersive experiences for gaming, e‑commerce, and advertising, and physically accurate real‑time simulation for robotics and autonomous vehicles. This focus on performance at the inference edge highlights why infrastructure‑centric platforms are drawing outsized capital commitments.
Vertical Integration and Hardware Shift from Nice-to-Have to Core Thesis
Across both Hark and Decart, a common pattern emerges: vertical integration is no longer an optional strategy but a central investment thesis. Hark is building a stack that spans proprietary AI models, context‑aware software agents, and dedicated AI-native devices, with the goal of creating deeply personalized interfaces that feel more like intuitive companions than tools. Decart is constructing a tightly coupled system where its DOS 2.0 inference engine, world models, and application‑specific offerings for gaming, commerce, and physical AI are engineered in concert. For investors, these vertically integrated architectures promise tighter optimization, better control over latency and reliability, and potentially stronger margins than software‑only AI plays that depend on third‑party infrastructure. As Series A valuations climb into multi‑billion‑dollar territory, the message is that future category leaders in AI may look less like standalone apps and more like full‑stack platforms owning everything from silicon to user experience.