Rebellions CEO Park Sung-hyun: K-Inference Will Break Nvidia's Monopoly

*Editor's Note: 'AI PRISM' (Personalized Report & Insight Summarizing Media) is an AI-based personalized news recommendation and summary service developed with support from the Korea Press Foundation. It selects and provides six customized news items by reader type.

1. Rebellions CEO Park Sung-hyun: "It's the NPU Era Now... K-Inference Will Break Nvidia's Monopoly"

- Key Summary: Rebellions CEO Park Sung-hyun stated at a special lecture at "Seoul Forum 2026" on the 28th that the open-source ecosystem can overturn Nvidia's monopoly in the AI inference market. The NPU designed by Rebellions touts 5 to 7 times the price efficiency of rival systems and focuses on inference optimization, and has already been deployed for SK Telecom (017670) A.'s call recording summary feature and KT Cloud's NPUaaS (NPU-as-a-Service) infrastructure. The company has also formed a global semiconductor alliance by attracting strategic investors (SIs) including Samsung Electronics (005930), SK hynix, ARM, Marvell, Amkor, and Pegatron, and is working with Saudi Arabia's Aramco on building sovereign AI (nation-controlled AI) infrastructure. "Rebellions' goal is to confidently stand against Nvidia and advance into the global market within two to three years," Park said.

2. Naver to Invest 1 Trillion Won Over Five Years to Build AI Creator Ecosystem

- Key Summary: Naver is launching a 1 trillion won impact fund over five years to expand its creator ecosystem and strengthen AI search competitiveness. Starting next month, the company will roll out "Naver Mate," a top-creator support program that selects 3,000 creators each month from blogs, cafes, Knowledge iN, and premium content, paying out 20 billion won annually in activity fees, with top-tier creators receiving up to 10 million won per month. AI Briefing has reached 30 million monthly users, while "AI Tab," beta-launched last month, surpassed 3 million cumulative users in just one month, and a new version of the image search tool "Smart Lens" is scheduled for release at the end of June. The ultimate goal is to implement a vertical AI agent (AI automation service specialized in specific fields) that completes shopping, venue reservations, finance, and real estate within a single service.

3. FuriosaAI Forms "Alliance" with Broadcom to Jointly Develop Next-Generation Inference Platform

- Key Summary: Korean AI semiconductor company FuriosaAI has signed a strategic partnership with global semiconductor firm Broadcom to jointly develop next-generation AI accelerators. The collaboration goes beyond simple joint chip development, aiming to build a hyperscale (mega data center-class) AI inference platform integrating AI computing, networking, and software. FuriosaAI's third-generation AI accelerator under development will feature a TCP (Tensor Contraction Processor) architecture and HBM4 and HBM4E memory, with Broadcom incorporating advanced packaging technology to begin sampling in the first half of 2028. "We will deliver industry-leading performance per watt even in ultra-large AI models and hyperscale agentic AI environments," FuriosaAI CEO Baek June-ho said.

4. Amid Easing of "Finance-Crypto Separation," Securities Firms Embrace Crypto Exchanges

- Key Summary: Amid financial authorities' easing stance on the "finance-crypto separation" principle, major Korean securities firms are accelerating their efforts to secure stakes in cryptocurrency exchanges. Three Samsung affiliates (Samsung Securities (016360), Samsung SDS, and Samsung Card) have decided to invest 612.8 billion won to acquire a 4.0% stake in Dunamu, while Hana Financial has agreed to acquire a 6.55% stake in Dunamu for approximately 1 trillion won. Hanwha Investment & Securities (003530) has raised its stake in Dunamu from 5.94% to 9.84%, becoming the third-largest shareholder, and Korea Investment & Securities is scheduled to hold a signing ceremony on the 29th to acquire a 40% stake in Coinone alongside global exchange OKX. With FSC Chairman Lee Eok-won signaling possible deregulation, competition for so-called "super apps" — which allow trading of stocks, crypto assets, STOs (security tokens), and stablecoins on a single platform — is becoming visible in Korea as well.

5. SK Telecom Introduces New AI Assistant That Automates Complex Tasks Independently

- Key Summary: SK Telecom has introduced a beta version of "A.Biz Cowork," an in-house AI agent that learns employees' work patterns to automate repetitive tasks. A.Biz Cowork operates in such a way that once a legal officer trains it on contract review criteria, it automatically organizes incoming contracts according to the same criteria, and it can also read dozens of specifications and automatically generate comparison tables by linking with collaboration tools such as Outlook and Teams. The company is also updating its company-wide AX (AI transformation) management platform "AXMS" to version 1.5, building an integrated management system covering AI project dashboards, training programs, and output sharing. "The use of AI is evolving beyond the stage of getting answers to questions, to the stage of teaching AI my own way of working," said Yoon Hyun-sang, head of A. planning at SK Telecom.

6. After Calling It a Virus... Microsoft Says "OpenClo Is 10 Times Faster Than Humans"

- Key Summary: Microsoft (MS) has officially announced the so-called "Lobster Project" to integrate OpenClo — an open-source autonomous AI agent platform once labeled as a "virus" by the company — into its enterprise subscription service M365. "Through open-source projects, AI can do more than 10 times the work of humans," MS Vice President Omar Shahine emphasized at a media briefing held in Mountain View, California, on the 27th (local time). Released on GitHub this January, OpenClo drew attention for enabling 24-hour-running agents to be built without development knowledge, but Microsoft's official position has been completely reversed just two months after CEO Satya Nadella likened it to a "virus" citing security vulnerabilities. However, MS maintains that its proprietary AI model Copilot and proprietary chip Maia are still needed in parallel to reduce server and power costs of open-source expansion.