Eliyan Raises $145M at $1B Valuation for AI Chip Links

What happened

Eliyan, a startup working on technology to speed up how data moves between AI chips inside data centers, has raised $145 million in a venture round that values the company at $1 billion 1. Reuters reported the deal on July 29, framing it as a bet on solving one of the more stubborn engineering problems in AI infrastructure: the bottleneck created when massive AI workloads need to shuttle data between processors faster than today's interconnect technology comfortably allows 1.

The funding lands amid a broader wave of AI-related venture activity that spans wildly different bets — from chip-level plumbing to agentic software platforms to content detection tools to Chinese AI labs racing toward IPOs. Taken together, the week's deals sketch a map of where investors currently see opportunity in AI: not just in building bigger models, but in the infrastructure, verticals, and safeguards that surround them.

Arrakis, a startup building agentic AI tools for industries like aerospace, energy, logistics, and manufacturing, emerged from stealth with $38 million in funding, according to Fortune, which covered the raise as an exclusive 2. Pangram, a company focused on detecting AI-generated content, raised $9 million and used the moment to launch a new detection model, Pangram 4, alongside an AI image-detection tool still in research preview, TechCrunch reported 4. Meanwhile, two of China's most closely watched AI labs were in the news for very different reasons: DeepSeek reportedly paused a fundraising round targeting roughly 10 billion yuan as it moves toward a potential IPO, amid heightened scrutiny and sensitivity around its use of Nvidia chips, per Seeking Alpha 3. Moonshot AI, by contrast, closed a $3.5 billion round that pushed its valuation to $35 billion, with Seeking Alpha noting its Kimi-K3 model is being positioned as a rival to offerings from OpenAI and Anthropic 5.

Where the reporting agrees

Across these accounts, there's a consistent throughline: capital is flowing heavily into AI, but investors are increasingly differentiating between layers of the stack. Eliyan's raise, per Reuters, is squarely aimed at hardware-level infrastructure — the physical constraints of moving data between chips — rather than at model-building itself 1. Arrakis and Pangram, as described by Fortune and TechCrunch respectively, represent a different investor appetite: applying existing AI capabilities to specific industries or specific problems, like verifying whether content was machine-generated 24. And the DeepSeek and Moonshot stories, both via Seeking Alpha, agree that Chinese AI labs are simultaneously racing to raise enormous sums and racing toward public listings, even as they operate under geopolitical scrutiny tied to chip access 35.

No outlet disputes the basic shape of any of these deals — the reporting is not contradictory so much as it is covering distinct, non-overlapping stories that happen to share a common moment and theme.

Where it doesn't

Because these five reports cover five separate companies rather than one shared event, there isn't a direct factual conflict to adjudicate — no outlet offers a competing account of Eliyan's valuation or funding amount, for instance. The more notable divergence is in framing and certainty. The DeepSeek item is explicitly sourced to reporting that the round was paused, a characterization Seeking Alpha attributes to unnamed reporting rather than presenting as company-confirmed fact 3. Moonshot AI's $3.5 billion raise, by contrast, is presented with more confidence and specificity, down to the $35 billion valuation figure and the framing of Kimi-K3 as a direct competitor to Western frontier labs 5. Arrakis's debut is labeled an exclusive by Fortune, meaning the terms and framing come primarily from one outlet's access rather than independently corroborated reporting 2.

What it adds up to

The throughline that best withstands scrutiny is that AI investment has fragmented into distinct bets rather than a single narrative of model-scaling dominance. Eliyan's $1 billion valuation for chip-interconnect technology signals that infrastructure bottlenecks are now viewed as investable problems in their own right, not just engineering footnotes 1. That sits alongside enterprise-vertical plays like Arrakis 2, trust-and-safety tooling like Pangram 4, and the geopolitically charged fundraising paths of Chinese labs like DeepSeek and Moonshot AI 35. None of these stories overturn each other; they collectively describe a venture market still willing to fund AI at nearly every layer, even as scrutiny around chips, valuations, and content authenticity intensifies.