Nvidia has spent 2026 buying every layer of the AI stack it does not already dominate: a long-term partnership with Corning for optical connectivity capacity, a strategic partnership and $2 billion investment with Coherent for silicon photonics manufacturing, and a strategic collaboration with Lancium for power infrastructure. One gap remains. The company still relies on an independent supplier for the signal-conditioning and fabric-switching silicon that determines how fast data actually moves between accelerators inside a rack. That supplier is Astera Labs.
The strategic logic is no longer theoretical. Astera is already collaborating with Nvidia on NVLink Fusion, working with a hyperscaler customer to enable its custom compute to be deployed over the NVLink ecosystem, with a product ramp expected in 2027. Nvidia’s VP of Engineering said publicly that the combination of NVLink Fusion and Astera Labs’ connectivity solutions gives customers access to scale-up interconnect technology in a proven rack-scale architecture, with “more choice in bringing semi-custom AI systems to market faster.” That is not a vendor relationship. That is a dependency.
The product moat is real and widening. On October 8, Astera announced Aries 7, what it claims to be the industry’s first signal conditioning portfolio spanning both copper and optical links for the PCIe 7 standard. The expanded portfolio offers hyperscalers and OEMs connectivity that now extends reach from centimeters of PCB trace to 50-plus meters over optical cables spanning multiple racks. Then on October 12, the company unveiled Orion: a rack-scale reference architecture built on open standards, using a 320-lane PCIe switch fabric capable of connecting up to 160 accelerators per pod. OEM and ODM partners are expected to offer Orion-based systems beginning in Q1 2027.
Astera is not the only connectivity name building a moat around AI rack architecture. Ciena has been quietly capturing AI networking spend with a $7 billion backlog and pluggable optics acquisitions that serve the same hyperscaler buildout.
That last detail is the one Nvidia should find most uncomfortable. Orion addresses AI infrastructure challenges with a production-grade reference architecture that decouples rack-scale designs from accelerator and CPU selection. A compute-agnostic rack standard, built by a third party, is exactly the kind of open-platform wedge that could eventually compete with Nvidia’s own DGX appliances at the infrastructure layer. Acquiring Astera would end that risk before it compounds.
The financial case is strengthening fast. Astera posted record quarterly revenue of $392.4 million in Q2 2026, up 27% sequentially and up 104% year over year. The company then guided Q3 revenue to $540 million to $560 million. Q3 2026 results are due after the close on November 3, 2026. Dollar content per XPU is at about $1,000 within AI racks, and the Scorpio X ramp has not fully hit those numbers yet.
The pricing pressure on rack-scale systems adds another dimension to Astera’s valuation case. Hyperscalers absorbing 15% system price increases on Nvidia-stack equipment have a structural incentive to diversify connectivity suppliers before 2027 shipments hit the books.
Astera currently carries a market cap of about $50 billion. That is real money, but Nvidia is sitting on a market cap north of $5 trillion and has already demonstrated it will write large checks when the strategic logic is tight. The Mellanox deal in 2019, which locked down InfiniBand networking, is the cleanest precedent: Nvidia paid $6.9 billion for a network fabric company and spent the next five years turning that bet into a structural advantage. Astera Labs is a version of that trade, priced for today’s connectivity bottleneck rather than 2019’s.
Traders should watch the November 3 earnings call closely. Revenue heading toward $550 million a quarter, Aries 7 sampling beginning in Q4 2026, and Orion shipping through partners in early 2027 will collectively put a floor under ALAB’s valuation. Each quarter that passes without a deal raises the acquisition price. Nvidia knows this. The question is whether the DOJ’s current posture on semiconductor M&A slows the timeline or merely adds paperwork.
