Special Features

Cloud Infrastructure Month

Supermicro CEO teases service to build or upgrade datacenters in six months

Doubles annual revenue, but warns repeating that will be hard while waiting until March for Nvidia Blackwell


Supermicro has teased a datacenter construction methodology that CEO Charles Liang claimed can create a small bit barn in six months or shrink the time to build bigger houses from three to two years.

Liang mentioned the methodology on the earnings call to discuss the IT infrastructure concern's Q4 and FY 2024 results, during which he revealed quarterly revenue of 5.3 billion and net income of $353 million – figures that jumped from $2.18 billion and $194 million from the same quarter last year. Full-year revenue of $14.94 billion more than doubled last year's $7.12 billion and net income jumped from $640 million to $1.2 billion.

The CEO pointed out that this year's Q4 revenue alone topped annual revenue from 2022.

But the news wasn't all good: Liang admitted costs for the direct liquid cooling (DLC) tech that he expects will differentiate Supermicro and fuel growth have been high – when it's been available – amid a component shortage.

Those shortages meant around $800 million of revenue couldn't be recognized in Q4 and put a dent in margins.

But the CEO was confident Supermicro's investments in extra production capacity can sort out those shortages and see it double the number of DLC-packing racks it makes – from 1,500 to 3,000 a month during this financial year. Liang said Supermicro's liquid cooled offerings cost no more than conventional kit, but consume less energy and therefore save users plenty of moolah.

Demand for that kit is expected to accelerate, leading Supermicro to forecast FY revenue of $26 to $30 billion.

The rapid datacenter building methodology – called Supermicro 4.0 Datacenter Building Block Solutions (DCBBS) – is among the things Liang thinks will help.

Liang said the solutions will be offered by year's end and "will significantly improve datacenters' Time to Online and cost, with full integration of AI compute, server, storage, switch, networking, rack, cabling, DLC, facility water tower, end-to-end management software, onsite deployment services, and maintenance."

The CEO's two-year build time prediction for the service referred to new AI-capable datacenters. The six-to-twelve-month forecast covered "smaller facilities or old datacenter transformations."

Supermicro does not, however, expect Nvidia's forthcoming next-gen Blackwell devices to be a big contributor to FY 2025 growth.

Liang told investors that Nvidia always pushes out the data on which newly announced products will debut, and that he expects that will be the case for Blackwell.

"I understand Blackwell may postpone – how much we don't exactly know because the new technology always can be pushed out, right?" The CEO noted he does not expect "any Blackwell volume" for calendar Q3 and only "very small … engineering sample volume" in the December quarter.

"The real volume, I believe, had to be March quarter next year. And that's why we foresee only $26 billion to $30 billion."

Financial analysts on the call were worried about declining margins, and Liang tried to reassure them that Supermicro will improve them with its expanded manufacturing capacity and bit barn building plans.

Shareholders didn't buy it: Supermicro opened at $616.01, and its shares spent the day in a band between $588.82 and $628.79. After the results announcement, the scrip spiked to $724.47 before settling to $535 in after-hours trading. It did that despite the biz also announcing a ten-for-one stock split to make its shares more accessible – a move usually made so that more small investors can get aboard a growing stock. ®

Send us news
2 Comments

Nvidia CEO whines Europeans aren’t buying enough GPUs

EU isn’t keeping up with US and China investments, AI arms dealer says

India, Nvidia, discuss jointly developed AI chip

Current capabilities mean local manufacturing is not likely – but a chip tuned to Indian needs could work

Hugging Face puts the squeeze on Nvidia's software ambitions

AI model repo promises lower costs, broader compatibility for NIMs competitor

AMD teases its GPU biz 'approaching the scale' of CPU operations

Q3 profits jump 191 percent from last quarter on revenues of $6.2 billion, helped by accelerated interest in Instinct

AMD targets Nvidia H200 with 256GB MI325X AI chips, zippier MI355X due in H2 2025

Less VRAM than promised, but still gobs more than Hopper

No-Nvidias networking club convenes in search of open GPU interconnect

Ultra Accelerator Link consortium promises 200 gigabits per second per lane spec will debut in Q1 2025

xAI picked Ethernet over InfiniBand for its H100 Colossus training cluster

Work already underway to expand system to 200,000 Nvidia Hopper chips

AI firms and civil society groups plead for passage of federal AI law ASAP

Congress urged to act before year's end to support US competitiveness

The troublesome economics of CPU-only AI

At the end of the day, it all boils down to tokens per dollar

European datacenter energy consumption set to triple by end of decade

McKinsey warns an additional 25GW of mostly green energy will be needed

UK gov report to propose special zones for datacenters, 'AI visas'

Vendors not keen on 'lengthy bureaucracy,' and cost when they try to hire skilled foreigners

Supermicro crams 18 GPUs into a 3U AI server that's a little slow by design

Can handle edge inferencing or run a 64 display command center