When growth starts looking gravitational
I have seen plenty of AI stocks produce arguments. Nebius has managed something more entertaining: it has produced two entirely different realities.
Nebius built the rocket. Investors are still debating the fuel
In one, Michael Burry sees the familiar ingredients of a late-cycle infrastructure boom, with depreciation potentially concealing the economic cost of rapidly ageing GPUs. In the other, Wall Street sees one of the fastest-growing AI infrastructure businesses in the market, backed by enormous customer demand, Nvidia's strategic involvement and a balance sheet increasingly built for the next phase of the build-out.
Both cannot be completely right. That is what makes Nebius interesting.
The numbers explain the excitement. $NEBIUS(NBIS)$ generated $1.36 billion of trailing revenue to June 2026, compared with $529.8 million for FY2025 and just $91.5 million in FY2024. Its annualised run-rate revenue has reached $3 billion, versus $1.25 billion at the end of 2025 and only $90 million a year earlier.
That is not normal growth. It is what happens when a company straps a rocket to an income statement.
But rockets require fuel, and Nebius is consuming an extraordinary amount of it.
The depreciation question nobody can spreadsheet away
The most important debate surrounding Nebius is not whether AI demand exists. Clearly it does. The more difficult question is whether the accounting life assigned to the infrastructure supporting that demand reflects its true economic life.
That is the heart of the depreciation argument.
Nebius' trailing depreciation and amortisation has risen to $863.8 million, from $468.5 million in FY2025 and just $85.8 million in FY2024. At the same time, property, plant and equipment has exploded from $891.5 million at the end of 2024 to $6.47 billion at the end of 2025 and $14.9 billion by June 2026.
That is an extraordinary acceleration in capital intensity.
The bear case therefore deserves more respect than simply dismissing it as another Burry contrarian trade. If GPUs and associated infrastructure become economically obsolete much faster than their accounting depreciation schedules imply, reported earnings can look healthier than the underlying economics.
And Nebius already provides an intriguing clue. TTM EBITDA is $258 million, yet depreciation and amortisation is $863.8 million. The gap is enormous.
This does not prove Burry is right. It does show why depreciation deserves to be treated as an economic cost rather than an accounting nuisance.
The financials are where the story gets uncomfortable
Revenue is exploding. The harder question is what growth costs
Nebius' gross profit has surged to $1.01 billion on a trailing basis, producing a 74.3% gross margin. That is a remarkable improvement from $363.6 million of gross profit and a 68.6% margin in FY2025.
That improvement is the strongest part of the bull case. Nebius is not merely buying GPUs and reselling compute at wafer-thin margins; the underlying economics are improving as the business scales. If management can sustain that margin profile while lifting utilisation, today's enormous infrastructure investment could eventually produce much more attractive returns.
Yet operating income remains deeply negative at $507.3 million.
That contrast is revealing. The business can generate impressive gross economics while still losing substantial money after the infrastructure and operating costs required to support its expansion.
Net income of $42.4 million looks superficially reassuring, but it needs considerable context. The company recorded $894.1 million of gains on sales of investments on a trailing basis. Without understanding that distinction, an investor could mistake a barely profitable company for one approaching mature profitability. The accounting department, in other words, has rather more interesting stories to tell than the headline earnings number.
Free cash flow is even less forgiving.
Nebius produced $5.26 billion of operating cash flow over the trailing period, but capital expenditure consumed $11.14 billion. The result was negative free cash flow of $5.88 billion.
That is the number I would put in unusually large type.
Nebius is not currently financing its growth from internally generated free cash flow. It is financing a gigantic expansion programme through debt, equity and customer-related cash flows. Financing cash flow reached $13.5 billion, while long-term debt stood at roughly $8.5 billion and long-term unearned revenue approached $5 billion. That unearned revenue matters: customers are effectively putting money into the machine before Nebius recognises all the associated revenue.
AI growth is expensive. Nebius is making the bill unusually visible
The balance sheet confirms the scale. Cash and equivalents stood at $8.04 billion at June 2026, while total debt had risen to $10.20 billion. Net debt was $2.15 billion.
That is not a distressed balance sheet. But it is a very different proposition from a cash-rich software company selling AI subscriptions from a server rack that magically pays for itself.
Nvidia changes the competitive game
The $NVIDIA(NVDA)$ connection is potentially more important than the headline ownership percentage suggests.
Nvidia's disclosed passive 9.3% stake gives Nebius something competitors cannot manufacture overnight: strategic validation from the dominant supplier of the hardware at the centre of the AI infrastructure boom.
But Nvidia's involvement does not eliminate competitive risk.
Nebius operates between hyperscalers such as $Microsoft(MSFT)$, $Amazon.com(AMZN)$ and $Alphabet(GOOGL)$, which possess vastly greater financial resources, and specialist neocloud operators chasing the same scarce GPUs and customers. Its advantage therefore cannot simply be access to chips.
The differentiator has to become execution: obtaining capacity, deploying it quickly, keeping utilisation high and converting committed demand into attractive returns on capital.
That is a much harder business.
The valuation leaves no room for stage fright
This is where my enthusiasm becomes considerably more restrained.
At $277.68 per share on 14 August, Nebius had a market capitalisation of $76.1 billion and an enterprise value of $78.3 billion.
The valuation is extraordinary. The stock trades at 56.2 times trailing sales and 57.8 times enterprise value to sales. Its price-to-book ratio is 7.4 times.
The headline P/E of 1,795 times is almost comically high, although that metric is arguably less useful for a company still absorbing massive investment and carrying unusual gains in its earnings statement. More useful, in my view, is the cash-flow picture: the shares represent a business with negative $5.88 billion of trailing free cash flow and a $76 billion market capitalisation.
In other words, investors are paying today for a very large amount of tomorrow.
That can work. It just cannot tolerate many tomorrows going wrong.
The real AI race may be against depreciation, not competitors
The real competitive analysis
I think Nebius' strongest competitive advantage is its position in the rapidly expanding neocloud layer between conventional cloud computing and AI-native infrastructure.
Its weakness is the same thing.
The opportunity is enormous because customers increasingly want specialised AI capacity without committing themselves entirely to the hyperscalers. But the capital required to supply that capacity is enormous too.
The balance sheet illustrates the trade-off perfectly. Property, plant and equipment has risen to $14.9 billion, while long-term unearned revenue has reached almost $5 billion. That unearned revenue is an important and easily overlooked feature of the story: customer commitments are helping fund infrastructure before the associated revenue is fully recognised.
That is encouraging. It also means execution matters enormously. Capacity has to be delivered, utilised and monetised.
A delayed data centre is not merely an inconvenience when billions of dollars are being deployed. At this scale, a construction timetable starts looking suspiciously like an earnings forecast.
So, is Burry right?
I would not make the mistake of treating this as a referendum on Michael Burry.
The more interesting question is whether Nebius can grow quickly enough, and profitably enough, to justify today's valuation before the infrastructure it is buying becomes economically outdated.
The bulls have a formidable argument. Revenue is exploding, gross margins have climbed from 68.6% to 74.3%, gross profit has surged to $1.01 billion and annualised revenue has reached $3 billion. Customer commitments can help finance the build-out, while Nvidia's involvement adds another layer of credibility.
Just as importantly, improving gross margins suggest the underlying commercial model is becoming more attractive as scale arrives. If utilisation rises without a corresponding collapse in pricing, the operating leverage could eventually become substantial. That is the bull case I think deserves more attention.
The bears have an equally formidable spreadsheet. Trailing free cash flow is negative $5.88 billion. Capital expenditure has reached $11.14 billion. Debt has risen sharply. Operating income remains negative. And depreciation is becoming a huge economic variable rather than a footnote.
My conclusion is therefore deliberately uncomfortable: Nebius is a fascinating test of whether AI infrastructure can outrun its own capital requirements.
Wall Street is betting that it can.
Burry is betting that gravity eventually remembers its job.
The unresolved question is which one your portfolio can afford to be wrong about.
@TigerStars @Daily_Discussion @Tiger_comments @Tiger_SG @Tiger_Earnings @TigerClub @TigerWire
Comments