Three hundred sixty million, the first stock in large models bought a building

What hidden asset optimization strategies are behind AI · Zhipu’s property purchase?

“First Stock in Large Models” Zhipu has invested several hundred million yuan, aiming for a building.

Recently, there was a change in business registration information on Tianyancha App: the wholly owned major shareholder of Beijing Red Diamond Technology Development Co., Ltd. was changed from Anshi Red Diamond Investment Co., Ltd. to Zhipu, with a committed investment of 445 million yuan. The core asset of Red Diamond Technology is the Diamond Tower located in Haidian District, Beijing. This is an office building in a low-density industrial park, standalone, with a total construction area of about 22.7k square meters, including one basement level and three above-ground floors.

On April 14, earlier, Zhipu announced that it planned to acquire related assets of the Diamond Tower for no more than 361 million yuan, including 81.62 million yuan in cash and assuming about 279 million yuan of related debts of the target company.

Since Zhipu listed on the Hong Kong Stock Exchange in January this year, with the “Lobster” hype soaring and the Token economy exploding, its stock price has been rising steadily. On April 13, during trading, Zhipu’s stock price once broke through HKD 1,000, with a market value exceeding HKD 430 billion. As of May 8, Zhipu closed at HKD 923 per share, with a market cap of HKD 22.7k.

This “new star” on the Hong Kong stock market has been listed for less than four months, and every investment move it makes attracts market attention. Its first large expenditure after going public was spent on buying property.

On the books, Zhipu doesn’t seem to be short of money. Its IPO raised as much as HKD 4.3 billion (about RMB 3.7 billion). According to the company’s financial report, by the end of 2025, the company still holds cash and cash equivalents of about RMB 2.26B. The total of these amounts is nearly RMB 6 billion, so 360 million yuan only accounts for 6%.

However, from the performance perspective, although Zhipu is in a period of rapid growth, its profitability remains insufficient. In 2025, Zhipu’s total revenue is projected at 724 million yuan, a year-on-year increase of 131.9%, but net loss reaches 4.72B yuan, an increase of 59.5% year-on-year. Just last year, Zhipu spent 3.18 billion yuan on R&D.

While R&D is still heavily “burning money,” why is Zhipu making such a large investment in real estate? Regarding the purpose of this transaction, Zhipu explained in the announcement: it intends to use the property as the company’s headquarters for self-use, to meet daily administrative and large model business operation needs; at the same time, leveraging the potential value of the property to support long-term strategic planning, optimize asset structure, and enhance risk resistance.

Not only Zhipu, but many tech companies, during rapid growth phases, almost always acquire real estate as their core office space. In February, ByteDance’s Beijing Yun Yue Changshi Technology Co., Ltd. acquired two parcels in Zhongguancun Street, Haidian District, at a bottom price of 2.8 billion yuan. A month later, ByteDance’s wholly owned Beijing Yunxiu Changshi Technology Co., Ltd. bought a third-phase site in the Shuangying Science Park on Xuesheng Road, Haidian District, for a total of 3.3 billion yuan.

According to Bai Wenxi, Vice Chairman of the Chinese Enterprise Capital Alliance, in the context of tightening financing environments and valuation logic shifting from “market dream” to “cash flow,” holding core location real estate is a defensive move to optimize the asset-liability statement. Tech companies mainly buy land for self-use, with investment as a secondary purpose. The deeper signal is that the industry has shifted from reckless expansion to a pragmatic stage focused on cost reduction and efficiency, using heavy assets to gain longer strategic buffers.

Zhipu, which is growing revenue while incurring losses, also seems to be signaling to the market through property purchases that it has made long-term strategic arrangements for the future.

Large model companies need to deploy high-density R&D data centers and computing clusters, which require renovations to office buildings’ power, cooling, and other systems. From this perspective, buying property is still cost-effective. Anshi Assets bought the Diamond Tower from Oceanwide Capital, the owner at the time, for about 850 million yuan in 2021. Now, Zhipu paid less than half that price to acquire it, settling in—considered a good deal.

Bai Wenxi told Caijing Tianxia that for tech companies, moving from “rental offices” to “heavy asset anchoring” is also a necessary business practice. He explained that large model enterprises need stable, scalable office environments to lock in core R&D teams. Leasing faces uncertainties like rent fluctuations, while owning property helps turn labor costs into “organizational assets” and can serve as collateral to expand financing channels.

Behind this, Zhipu may also have a broader strategic intent—acting as a “landlord,” it can attract more related companies and build its large model ecosystem. The location of Diamond Tower may also support this idea: it is situated in Zhongguancun Software Park Phase I on West Wusong Road, Haidian District, near R&D centers of giants like Baidu and Tencent, only 8 kilometers from Tsinghua University, where Zhipu originated. From 2004 to 2021, IBM China Research Institute was also based here, renting for 17 years.

Zhipu has been actively building a full industry chain ecosystem for large models. In 2024, it partnered with Xinglian Zhaoji, Jingcheng Yebei to jointly invest 260 million yuan to establish Xinglian Capital, investing in early-stage projects across the large model supply chain. By the end of 2025, Xinglian Capital had invested in over 30 AI companies. In May this year, Guoli Technology, the largest independent AI computing cluster provider in China, launched a Hong Kong IPO, with Zhipu and Xinglian Capital among its investors.

（Author | Yang Yi, Editor | Wu Yue, Image Source | Visual China, Content from Caijing Tianxia WEEKLY）