Accelevation’s $660M IPO: Olympus Is Selling the AI Shovels
A $660 million IPO where 71% of the shares are being sold by existing owners is not a growth story. It is a very expensive question: how long can AI’s data-centre building boom last?
The owners are selling more stock than the company is issuing
Here is the bit most IPO coverage buries halfway down the page: Accelevation is offering 30 million shares at US$20 to US$24 each, but 21.36 million of them are coming from existing selling shareholders.
That is roughly 71% of the deal.
At the US$22 midpoint, Accelevation’s IPO would raise US$660 million in total. But only about US$190 million of that gross amount is tied to newly issued shares for the business. The rest is liquidity for people already on the register.
Nothing illegal. Nothing automatically sinister. But let’s stop pretending it means the same thing as a company raising US$660 million to go and build factories, hire people and take market share.
Olympus Partners bought Accelevation in early 2025. Now, barely two years later, it is taking a business built around data-centre power infrastructure to Nasdaq at a midpoint valuation of US$4.9 billion. Olympus will retain majority voting control after the float.
That is private equity doing what private equity does: buy an excellent business, improve or position it, then sell part of it into a market desperate for the theme.
And right now, the theme is AI.
Accelevation is not selling AI. It is selling the stuff AI cannot run without.
Accelevation, based in Miamisburg, Ohio, designs, manufactures and installs the electrical, mechanical and structural infrastructure inside data centres. Think power distribution units, remote power panels, thermal-management gear, factory-built systems and installation services.
No chatbot. No glamorous founder on stage in a black turtleneck. No promise that it will replace your accountant by Christmas.
Just the very unsexy equipment and field work required when a hyperscaler, cloud operator or colocation provider needs a building full of servers powered, cooled and commissioned properly.
That matters because the AI boom has collided with physical reality. You can announce all the GPU clusters you like. They still need electricity delivered safely, heat removed reliably, equipment installed on time and systems that do not fail when the customer turns them on.
Accelevation reported US$727 million in revenue for the 12 months ended June 30, 2026. Its revenue grew 147% in 2025, and it had roughly US$1.1 billion of backlog at June 30. Those are serious numbers for a business most ordinary investors had never heard of yesterday.
The company plans to list under the ticker ACCV, with Morgan Stanley and J.P. Morgan leading the book. It is expected to price in the week beginning September 28.
This is exactly the sort of IPO Wall Street loves when an investment theme gets hot: a real business with real revenue, attached to a giant capital-spending cycle, given a clean public-market label.
The catch is that real revenue does not make a business immune from a very stupid price.
The US$4.9 billion question is not whether AI is real
AI is real. The data-centre buildout is real. Power constraints are real. Anyone still arguing that point is having a different conversation from the one the market is actually having.
The proper question is whether the current spending boom produces durable, high-return demand for suppliers such as Accelevation — or whether it creates a few years of frantic ordering followed by a nasty hangover.
At the midpoint, the proposed US$4.9 billion market value is about 6.7 times Accelevation’s US$727 million in trailing 12-month revenue. That is not a dot-com valuation. But it is not a bargain-bin industrial multiple either.
Investors are being asked to pay for three things:
1. Growth continuing. A 147% growth rate makes everyone look clever. Repeating it gets harder as the revenue base grows. 2. Backlog converting cleanly. A US$1.1 billion backlog is valuable only if customers proceed, projects stay funded and margins survive the work. 3. AI infrastructure spending remaining rational for longer than markets usually expect. This is the big one.
When customers are scrambling for capacity, suppliers can sell speed, reliability and certainty. Those are premium products. When capacity catches up, customers start asking why the panels cost so much and whether another supplier can do it cheaper.
That is when the blokes who bought a story discover they actually owned an industrial business.
Industrial businesses can be brilliant investments, by the way. I like them when they have difficult-to-replicate operations, strong customer relationships, proper service capability and disciplined management. But the market must value them as businesses, not as spiritual extensions of Nvidia.
The overlooked detail: this is a test of private-equity timing
The loud story is AI infrastructure. The more useful story is Olympus Partners.
Private equity is very good at spotting an exit window. It does not need to believe the AI buildout is about to collapse. It simply needs to recognise that public investors are presently willing to pay handsomely for exposure to the picks-and-shovels side of the boom.
That is a different calculation.
Olympus acquired Accelevation last year, according to Axios. It will remain in control after the IPO, which means new public shareholders are not buying a business where control has meaningfully changed hands. They are buying a minority stake alongside a sponsor that has already chosen this moment to take money off the table.
Again: that is not an accusation. It is the deal.
But if you are considering buying an IPO like this, you need to read the capital structure as carefully as the revenue chart. Who is selling? Who controls votes? Who receives the cash? What does the company itself get to do with its portion?
Accelevation says it intends to use its net proceeds through its operating structure to repay debt, cover offering and organisational costs, and for general corporate purposes. Reducing debt can be sensible. It is not the same thing as pouring all the proceeds into a fresh growth engine.
The seller receives liquidity today. The public investor receives exposure to tomorrow.
That can work beautifully. It can also be a terrible trade if tomorrow arrives with lower growth, weaker margins and a valuation that assumed the boom would last forever.
Backlog is not revenue, and revenue is not cash
This is where founders and investors alike get themselves into trouble. They start treating the best available metric as if it were all the metrics.
Backlog is encouraging. It tells you customers have work planned. But it is not banked revenue, and it certainly is not free cash flow.
A company serving large data-centre projects has execution risk all over the place: customer construction delays, procurement changes, labour availability, supply-chain bottlenecks, project-cost overruns, changing equipment specifications and financing conditions. One delayed campus can move a meaningful amount of work from one quarter to the next.
Then there is working capital. Fast-growing infrastructure businesses often have to buy materials, build inventory, reserve capacity and carry costs well before the cash arrives. Growth can make a company look rich on paper while making it hungry for cash in reality.
That does not mean Accelevation has a broken model. It means investors should demand the boring answers before paying a premium multiple:
- How much of backlog is contractually committed versus forecast demand? - What are cancellation rights and typical delivery lead times? - What portion of revenue comes from repeat customers? - Are gross margins improving, stable or being sacrificed to win volume? - How much cash is tied up in inventory and receivables as revenue grows? - What happens if a major AI customer pauses a campus build for six months?
If management cannot answer those crisply, the AI label is doing too much heavy lifting.
The contrarian view: the hardware may be safer than the hype
Here is the bullish case that deserves respect.
The market has spent years rewarding software companies that promise to capture AI value someday. Accelevation sits closer to the actual construction site. Its customers are spending money now, its products solve physical constraints now, and its field-service capability cannot be copied by prompting a large language model.
That is a better starting point than a slide deck with a synthetic voice demo.
There is also a genuine moat in being able to design, manufacture, deliver and install complex infrastructure at speed. Plenty of companies can sell components. Fewer can take responsibility for the job getting done across a national footprint, under deadline, in an environment where delays can cost a customer millions.
If AI demand stays robust, the bottleneck may not be chips alone. It may be power delivery, cooling, skilled labour and deployment capacity. Companies that solve bottlenecks tend to get paid well.
But that is precisely why discipline matters. A good company at a silly price is still a bad investment. And a sponsor-backed IPO with a heavy secondary component deserves more scrutiny, not less, when the underlying industry is fashionable.
What this means for you
Whether you are a founder, operator or investor, take three practical lessons from Accelevation’s float.
First: follow the money, not the headline number. A US$660 million IPO sounds like the company is receiving US$660 million. It is not. Work out who is selling, who is getting cash and what the business will actually do with its share.
Second: build around a bottleneck. Accelevation is interesting because it serves a problem customers cannot wish away: getting power and cooling infrastructure deployed. In your own business, stop chasing fashionable labels and find the expensive operational pain where customers already have budget.
Third: never confuse momentum with permanence. A US$1.1 billion backlog and 147% growth are excellent. They are not a lifetime guarantee. Good operators use boom times to strengthen balance sheets, improve systems, lock in customers and prepare for the day buyers regain bargaining power.
That is the real test for Accelevation. Not whether it can get public while AI infrastructure is hot. Anyone can sell a hot story.
The test is whether, five years from now, it is still the company data-centre operators call when speed, power and uptime actually matter — after the hype merchants have moved on to the next shiny thing.