Celestica counts Amazon.com Inc., Google and Meta Platforms Inc. as customers.Fred Lum/The Globe and Mail
The boom in artificial intelligence is breathing new life into Celestica Inc. CLS-T, the Toronto-based global manufacturer that rode the excesses of the dot-com era a generation ago and crashed when the bubble burst.
Celestica now supplies equipment to the data centres and server farms that provide the computing power for AI models. Companies such as Amazon.com Inc. AMZN-Q, Google GOOGL-Q and Meta Platforms Inc. META-Q – all Celestica customers – are investing heavily in upgrading computing capacity partly for AI development, which has seen a surge of interest since OpenAI released ChatGPT a year ago.
“It’s not the sleepy old Celestica any more,” said chief executive officer Rob Mionis. “It’s a much different company than it was years ago.”
The company’s AI and machine learning business has more than tripled compared with 2022, and now accounts for about US$1.2-billion of Celestica’s estimated US$7.9-billion in revenue for this year, which is at the highest level in more than a decade. Investors have noticed. Celestica’s stock price is up about 141 per cent so far this year, closing Tuesday at $36.86. That’s still well below the October, 2000, peak of $128. Even so, the sale by Celestica’s long-time dominant shareholder, Onex Corp. ONEX-T, of its stake earlier this year has added long-needed liquidity to the stock.
But one big question still hangs over Celestica: Will this time be different? Many investors worry that AI is overhyped and valuations are inflated. The corporate world is still figuring out how to integrate generative AI, which suffers from reliability problems, while providers such as Google face huge development costs.
The era of generative AI has been compared with the dawn of the internet, which could ring alarm bells for anyone familiar with Celestica’s history. The company, a former IBM division bought by Onex in 1996, went public in 1998 as the dot-com boom heated up, surging to a $20-billion market valuation as it supplied fibre-optic cables and other equipment used to build out the internet.
But rosy forecasts for continued growth crumbled as network companies massively overbuilt capacity, and Celestica’s shares tanked, doomed to live in investor purgatory for nearly two full economic cycles. Its early flurry of acquisitions tailed off for a decade until Mr. Mionis joined in August, 2015.
Xenon stock surges as B.C. biotech’s epilepsy drug shows promise for treating depression
The parallels with the dot-com era are not lost on Mr. Mionis. But there are also key differences, he contends. “We feel it’s sustainable. We’re in the very early innings of this,” he said of AI development. Secondly, Celestica is a more diverse and better insulated company than it was two decades ago.
Mr. Mionis, an American with a background in private equity and aerospace, inherited a company that had experienced years of losses, job cuts and shareholder lawsuits.
He shifted Celestica’s orientation, moving it away from manufacturing low-margin products and competing on price to building more complex equipment that required engineering expertise. “We purposefully were reducing our revenue to reshape the portfolio while some of our peers were growing revenue,” said Mandeep Chawla, the company’s chief financial officer. Indeed, Celestica shed more than US$1-billion “non-strategic” annual revenue in the process, according to Mr. Mionis.
Celestica now has an improved mix of customer segments with higher-margin products, Mr. Mionis said. It makes equipment for the aerospace and defence industries, renewable energy providers, electric-vehicle chargers and medical devices.
The company also started servicing tech giants that operate massive data centres. “The Googles of the world are growing leaps and bounds, and continue to see a need for our engineering excellence,” Mr. Mionis said. Celestica counts the top five cloud computing firms (or hyperscalers, in company parlance) as clients and the segment accounts for 35 per cent of its business.
Those relationships put Celestica in a good position to jump on the generative AI wave. Mr. Mionis started to see more equipment demand from cloud computing firms a year ago, coinciding with the release of ChatGPT. “All the hyperscalers are in an AI arms race,” he said. “That’s when we really started to take off.”
Mainly, Celestica provides computing modules, network switches and data storage capacity for server farms, none of which is a straightforward manufacturing job. It incorporates custom silicon and has proprietary thermal and power management designs to help companies manage the extreme heat and electricity needed to run data centres.
Demand has been so great that Celestica has had to expand its manufacturing facilities. Some capacity was already freed up as a result of recent restructuring, but the company is also building an additional 100,000 square feet in Thailand and 80,000 square feet in Malaysia.
Mr. Mionis doesn’t seem particularly concerned about competition, owing to Celestica’s established customer relationships and engineering capabilities. “We’re somewhat first-to-market on a lot of these designs,” he said.
Artificial intelligence is just one reason for Celestica’s turn of fate. Thanos Moschopoulos, an analyst with BMO Capital Markets, said the company’s shift toward more complex and higher-value equipment has boosted margins. During the COVID-19-induced supply chain crunch, Celestica was also more adept than some competitors in sourcing key components. “Customers took notice and helped them gain share,” he said. “The business seems to be on much better footing today.”
Long-time shareholder Onex, which once had the dominant voting stake in Celestica, exited its position this year and off-loaded more than 18 million shares. That created more liquidity and allowed other investors to buy in, Mr. Moschopoulos said.
Celestica’s business still carries risk. The industry is intensely competitive and manufacturers have limited pricing power. Celestica’s customers are highly concentrated, too. “We view the lack of customer diversity as an inherent disadvantage, because losing a customer or a large program could lead to significant loss in revenues and cash flow,” S&P Global Ratings analyst Monysh Bandeally wrote earlier this year.
Despite Celestica’s recent good fortune, its stock is relatively inexpensive compared with its peers, though it has mostly closed a significant gap. Celestica trades at about 10 times its estimated earnings per share for next year, compared with 12 times for competitors. Whether more investors will buy in – and whether Celestica has truly left the past behind – depends largely on how long the AI boom will last.
Mr. Mionis is optimistic. “We’re at the beginning of a very long demand cycle,” he said.