Nebius Group (NBIS +6.57%) sits at the center of every artificial intelligence (AI) hype checklist right now: graphics processing unit (GPU) cloud, hyperscaler partnerships, an Nvidia investment, and a stock that has multiplied in value over the last six months. The issue isn’t the story itself; it’s that investors now have to pay the valuation the market demands for it.
Rotating into Astera Labs (ALAB +2.03%) will give you AI infrastructure exposure with real operating leverage, cleaner unit economics, and more reasonable valuation assumptions.
Nebius is not a bad company. CEO Arkady Volozh separated Yandex’s non-Russian assets and built a legitimate GPU cloud platform from scratch — one that now counts Nvidia as a $2 billion strategic investor and runs AI compute infrastructure across Europe and the United States. The business is real. The question is what you’re paying for it.
Image source: Getty Images.
Nebius trades at roughly 99x trailing price-to-sales on approximately $530 million in revenue, with the stock up more than 522% over the past year. That multiple is not a reflection of what the company earns today; it is a reflection of what the market believes it will earn several years from now, assuming aggressive capacity expansion, high utilization rates, and sustained hyperscaler demand all arrive on schedule.
History has a consistent track record with stocks like this. When a company’s market cap is built almost entirely on future expectations, the path from current price to fair value is not a crash; it’s more of a long, slow grind lower as the business grows into its valuation over several years, while the stock goes nowhere. Revenue doubling from here would be impressive. It would also still leave the stock expensive. That is the trap.
I am not betting against Nebius. I just don’t want to own a stock where I need three consecutive years of perfect execution to break even. At this point, there is a better place to put this capital.
Today’s Change
(6.57%) $13.61
Current Price
$220.88
Key Data Points
Market Cap
$52B
Day’s Range
$207.80 – $233.70
52wk Range
$34.72 – $233.70
Volume
969K
Avg Vol
17M
Gross Margin
-765.63%
Astera Labs: The fabric under all of the GPU racks
Astera Labs sells the plumbing that sits one layer down in the stack: PCIe retimers, CXL controllers, and AI fabric switches that move data across GPU clusters. Its pieces are not visible in marketing decks, but they are essential once clusters scale beyond a few racks.
Last week, for the first quarter (Q1) of 2026, Astera reported revenue of $308.4 million, up 93% year over year and 14% sequentially. Generally accepted accounting principles (GAAP) gross margin came in at 76.3%, with GAAP operating income of $61.8 million and net income of $80.3 million, so the model already generates cash rather than burning it. Management guided Q2 revenue to a range of $355 million to $365 million, above prior consensus estimates of around $310 million, suggesting sustained demand from existing hyperscaler customers.
The product roadmap gives that guidance teeth. Astera’s new Scorpio X‑Series 320‑lane Smart Fabric Switch, built for PCIe Gen 6, is shipping to customers now, with production ramp scheduled for the second half of 2026. Management pegs the merchant scale‑up switch market that Scorpio targets at roughly $20 billion by 2030 as AI clusters grow in lane count and complexity. An existing warrant agreement gives Amazon the right to buy 3.26 million Astera shares if it purchases up to $6.5 billion of Astera products through early 2033, tying a large part of that ramp to one customer’s concrete spending plan rather than a generic “AI total addressable market.”
Risk profiles for both names
Both names face real risk. Nebius carries execution risk on a capital‑intensive build-out, concentration risk in a small number of large contracts, and the usual platform risk that comes with dependence on hyperscaler budgets and chip vendors. Astera carries customer concentration risk — with hyperscalers such as Amazon and other cloud providers as key accounts — plus competitive risk and in‑house solutions.
The difference is where the risk sits in the stack and in the stock’s pricing. With Nebius, a large portion of the risk sits in the multiple. The market is paying venture‑style prices for a company still in the cash‑burn phase, so even solid execution can lead to poor equity returns if the narrative cools. With Astera, the risk sits in the product and customer ramp. The valuation is rich, but recent profit metrics and the Scorpio X rollout suggest the market is paying for a visible roadmap rather than a distant promise.
For my AI infrastructure sleeve, I would rather underwrite interconnect hardware with current GAAP profitability and line‑of‑sight hyperscaler ramps than a GPU cloud platform priced as if every bull case scenario arrives on time.
