How shoddy crypto miner CoreWeave became the billion-dollar backbone of the AI ​​boom

At the heart of CoreWeave’s business are its data center engineers. The most experienced among them work like a special unit, flying from site to site to commission new data centers rather than working full-time on a campus. Venturo declines to say how many miles its crack engineering crew has logged, but says they installed about 6,000 miles of fiber optic cable last year. “I probably have the most interaction with this team, more than any other team in the company, simply because they are so incredibly important,” he says.

Some former CoreWeave employees say working at the company can be unnecessarily stressful. Employees are expected to always be available. Your personal phone numbers are visible on your company’s Slack profiles and cannot be removed. A former employee frustratingly says it allowed his colleagues to text him about work at odd hours.

Venturo says he introduced weekly massages as a perk for employees at CoreWeave’s New Jersey headquarters a few years ago after he first noticed back and neck pain at work. “If I feel this way, others feel the same way,” he says. (One of the former employees says that managers were usually the only ones who felt comfortable being treated in the office.)

In Venturo’s opinion, there is a reason for the high demands placed on employees. “Everything we do is about identifying and eliminating problems and blockages,” he says. “We push our teams to act quickly, and when things go wrong, we empower them by solving the problems together.”

Shortages in personnel and equipment as we grow have sometimes affected the reliability of CoreWeave’s services. The number of server outages began to rise with the move away from crypto because, until recently, CoreWeave only had one engineer focused on maintaining uptime, former employees say. According to two sources, the company sometimes did not have enough working GPUs to meet contractual obligations. “They never had enough spare parts,” says one. To compensate, the company skipped some tests of newly installed GPUs, the two sources say. “I think we have no choice,” a vice president said in an internal Slack message seen by WIRED. Venturo said the company’s testing platform was designed to ensure customers meet their timelines to get online.

CoreWeave’s fast pace has pushed some considerations — such as considering the company’s environmental impact — to the bottom, according to former employees. Venturo says most of CoreWeave’s data centers are powered by 100 percent renewable energy and that the company has not targeted data centers in areas such as flood plains that are not insurable. “It’s not so hectic that we end up in a situation where we have to make irrational decisions,” he says. When asked if CoreWeave would publish a sustainability report, like some other cloud companies doTo report its water usage and efforts to reduce supply chain emissions, Venturo told WIRED in October that it would have more to share soon. No report has been published yet.

Behind your co-pilot

If you’ve ever used ChatGPT, Microsoft’s Copilot or other generative AI offerings, or the Stable Diffusion image generator, you may have reaped the rewards of CoreWeave’s frantic labor.

According to sources, the company has assembled facilities for Microsoft using OpenAI is the host According to the chip manufacturer, a supercomputer data center for Nvidia in Plano, Texas. CoreWeave’s Plano campus cost about $1.6 billion and is about 42,000 square feet, about the size of the largest Ikea store in the United States. Another room dedicated to a single customer is in Oregon, where it was used to launch enterprise chatbot software Inflection AI. According to Venturo, these sites typically have 16,000 to 32,000 GPUs.

Source link