What 6,271 Failed Startups Reveal About China’s Tech Boom and Bust

Data Collection

Using the ITjuzi “death company” database, the author inspected the XHR requests, found the JSON API endpoint

https://www.itjuzi.com/api/closure?com_prov=&fund_status=&sort=&page=1

, and wrote a Python script to crawl the data.

def main():
    data = pd.DataFrame(columns=['com_name','born','close','live_time','total_money','cat_name','com_prov','closure_type'])
    for i in range(1,2):  # set number of pages to crawl
        url = 'https://www.itjuzi.com/api/closure?com_prov=&fund_status=&sort=&page=' + str(i)
        html = requests.get(url=url, headers=headers).content
        doc = json.loads(html.decode('utf-8'))['data']['info']
        for j in range(10):  # 10 companies per page
            data = data.append({
                'com_name': doc[j]['com_name'],
                'born': doc[j]['born'],
                'cat_name': doc[j]['cat_name'],
                'closure_type': doc[j]['closure_type'],
                'close': doc[j]['com_change_close_date'],
                'com_prov': doc[j]['com_prov'],
                'live_time': doc[j]['live_time'],
                'total_money': doc[j]['total_money']
            }, ignore_index=True)
            time.sleep(random.random())
    return data

The script successfully retrieved data for 6,271 closed companies.

Ten‑Year Survival Analysis

From the 6,271 records, 5,765 companies founded between 2010 and 2019 were selected for analysis. The data show a surge of new startups in 2013‑2014 followed by a wave of closures in 2016‑2017, with over 2,000 companies shutting down in 2017.

Various “wind‑up” trends such as mobile internet, live streaming, shared bikes, AI, and short video are illustrated, showing how many firms rode each wave only to burn through capital and collapse.

Even companies that seemed successful, like the short‑lived Panda Live, eventually fell victim to rising costs and market saturation.

Additional observations include “pseudo‑wind‑ups” such as shared charging stations and other low‑value sharing models that never gained traction.

The top “prayer” ranking (companies most remembered despite closure) is led by the infamous video platform QVOD.

Source code for the crawler is available on GitHub: https://github.com/zpw1995/aotodata/tree/master/interest/6217