Inference cloud startup General Compute raises $100 million, despite widespread fears about AI funding

Neoclouds are said to be under pressure as companies like Gartner Inc. and Elon Musk both predict a bloodbath in the emerging cloud infrastructure segment in the not-too-distant future, but apparently no one has told General Compute Inc. about it.

The cloud infrastructure startup announced today that it has just raised $100 million in a Series A funding round led by Blackbird and Abstract Ventures with participation from Fuse Venture Partners, Village Global and other investors, bringing its post-money valuation to $800 million.

The round may have caused a stir as General Compute’s competitors in the neocloud industry are currently going through a difficult time. This is best illustrated by the biggest neocloud of them all, CoreWeave Inc., whose shares recently fell to a 12-month low amid incessant talk about investors growing tired of funding what looks like an endless, multibillion-dollar build of artificial intelligence infrastructure.

Musk reiterated these fears last weekend when he spoke on Twitter, but warned that he could only receive the money because his company had alternative sources of income. “More than 200 neoclouds will experience Guppy-level death and consolidation,” Musk said. “Some will make it, but not many.”

Musk’s comments came a few weeks after respected analyst firm Gartner Inc. said up to 80% of all Neoclouds could disappear by the end of the decade. Some smaller neoclouds are already feeling the pressure. Earlier this month, Microsoft Corp. announced a multi-billion dollar deal with a startup called Nscale Global Holdings Ltd., which now calls into question the company’s plans for an IPO.

Despite all the negative headlines, Finn Puklowski, co-founder and CEO of General Compute, said in an interview with TechCrunch that he isn’t worried about what others are doing. He said Musk’s comments were likely an attempt to drive down the prices of graphics processors he wants to buy, and said fears of a collapse in the chip industry were “massively overblown” based on the demand his company was seeing.

General Compute also has a different business model compared to most neoclouds. Instead of focusing on general-purpose GPUs, the company specifically buys only inference chips from other chipmakers like Cerebras Systems Inc, SambaNova Systems Inc. and Positron AI Inc. Customers never have to worry about the hardware either. You simply set up a workload and press the start button, and General Compute routes it to the most appropriate chip. Customers can choose between dedicated bare metal servers or a service level agreement with guaranteed tokens per second.

Additionally, General Compute is not interested in building its own data centers, Puklowski said. Instead, it rents space in other companies’ facilities, which handles all the networking, power, cooling and maintenance. The startup also has money to play with. In addition to today’s funding round, the company secured $400 million in debt financing in July, most of which has not yet been spent.

Puklowski said the company has been running some workloads for customers, but the first large-scale chip implementations won’t go live until the first quarter of next year. More than 75 megawatts of capacity is planned for next year, and a major deal has already been completed that will bring in annual recurring revenue of over $100 million.

Blackbird’s James Palmer said silicon is no longer the biggest bottleneck in the AI ​​industry. Rather, what companies need is power. “Both the energy and silicon supply chains are growing slowly,” he added. “Finding creative ways through these constraints will determine who wins. Finn and his team are the ultimate coalition builders, and they think in decades, not days.”

The startup also received praise from the three chipmakers that supply it with inference chips. “We understand that inference performance depends on how the entire system is deployed and operated,” said Sean Lie, Chief Technology Officer of Cerebras. “We appreciate their approach and commitment to rapid and scalable implementation.”

Image: General Computing