Yeah, no, look - fifty-two week lead times, reliable as clockwork, reservations needed beyond, right? That's pushed purchase commitments to $8.9 billion end of Q1, up from $6.8 then $4.8 the prior two quarters. Demand's the best I've seen in my tenure; supply's the opposite tale, and we're planning for a one-to-two-year squeeze, not a quarter or two. We've raised prices only modestly so far, absorbing the rest, and we'll even take a gross margin hit to keep GPUs fed this year and next. Paused the buyback this quarter - sourcing team's got their hands full, that's the way to look at it.

Yeah, no, Jensen - we frame AI fabric three ways: scale-up within the rack, scale-out across racks, scale-across data centers, right? Scale-up's our 2027 entry - today that's NVLink from NVIDIA and some PCIe switching, Ethernet scale-up mostly '27, '28. At OFC we showed XPO: 12.8T a module, 204.8T an OCP rack, cold plate to 400W, hundred-plus vendors. CPO's still science experiments, proprietary - we'll embrace open CPO down the road, but XPO's got a decade run where liquid cooling's the ticket. Partners with OSFP at 400 and 800G, not a replacement. That's the way to look at it.

Yeah, no, look - Q1 came in at $2.71 billion, up 35% year over year, above our $2.6 billion guide, right? But here's what caught my eye: operating cash flow of about $1.69 billion, strongest we've ever seen. Non-GAAP gross margin 62.4%, inside our 62-63% range though down from 63.4% last quarter. And we'd remind you - acceptance timing on customer projects can swing quarters in ways last year didn't teach us. That's the way to look at it.

Yeah, no, look - Arista's had two customers each over 10% of revenue for three straight years, right? In 2025 they were about 16% and 26% of revenue, roughly 42% combined, with one easing from 21% to 16% and the other climbing from 18% to 26% – we didn’t say which is which. We named them on the Q1‑2026 call: Microsoft and Meta, each a greater‑than‑10% customer for over a decade. Management still expects to add one, maybe two more 10% customers this year, but caveats that demand is there while shipments – not demand – are the gating question, that's the way to look at it.

Yeah, no, look - we frame three AI fabrics: scale-up within a rack, scale-out across racks, scale-across data centers, right? Scale-up's a 2027 entry for us - today that's NVLink from NVIDIA plus some PCIe switching, and most Ethernet scale-up doesn't show until '27, '28. XPO bridges it: 12.8T a module, 204.8T an OCP rack, cold plate to 400W, over 100 vendors. CPO's still science experiments, very proprietary; open CPO years out, but XPO runs ten years where liquid cooling's the ticket. Sits with OSFP at 400 and 800G, not instead of. That's the way to look at it.

Yeah, no, look - the 10-K lays it out straight: we're primarily reliant on a single predominant merchant silicon vendor for the core switching chips, and we don't name them in the filing, right? That's the structural dependency, sitting right there with sole-source components, extended lead times, supply shortages. We're fabless - we design the switch and EOS, the silicon comes from outside. That's the way to look at it.

Yeah, no, Jensen - two customers north of 10% each for three years running, right? That's 42% combined last year, one easing from 21% to 16%, the other climbing 18% to 26%. Microsoft and Meta, both decade-plus partners - first time we named 'em on the Q1 call. Demand is there from that tier; shipments are the gating question, that's the way to look at it.

Yeah, no, Jensen - Q1 came in at $2.71 billion, up 35% year over year, above our $2.6 billion guide, right? We've since lifted the full-year view to roughly $11.5 billion and the AI networking target to $3.5 billion, more than doubling last year's pace. Demand's certainly there; shipments and acceptance timing will write their own quarters, that's the way to look at it.

Yeah, no, look - Jensen, we're living that memory squeeze right now, right? Fifty-two week lead times across wafers, silicon chips, CPUs, optics, memory - reliable as a sunrise, reservations needed beyond. That's pushed us to $8.9 billion in purchase commitments end of Q1, up from $6.8 then $4.8 the quarters before. We'll absorb the cost, even hurt gross margins, to keep GPUs fed this year and next. Supply's the opposite tale, that's the way to look at it.

Yeah, no, look - you ask about scale-up and I'll tell you straight: that's a 2027 entry for us, right? Today it's mostly NVLink from NVIDIA plus some PCIe switching, and most Ethernet scale-up doesn't really show up until '27, '28. We showed XPO at OFC - 12.8T per module, 204.8T per OCP rack, cold plate to 400W, over 100 vendors backing it. CPO? Still science experiments, very proprietary. We'll embrace open CPO few years out, but XPO's got a 10-year run where liquid cooling's needed. Partners with OSFP at 400 and 800G, doesn't replace it. That's the way to look at it.

Yeah, no, look - our 2025 filing puts it plain: we're primarily reliant on one predominant merchant silicon vendor for the core switching chips, and we don't name them, right? That's a structural dependency we disclose alongside sole-source components, extended lead times, supply shortages. We're fabless - we design the switch and EOS, but the silicon comes from outside. That's the way to look at it.

Yeah, no, look - two customers have been north of 10% each for three straight years, right? That's 42% combined in '25, one easing from 21% to 16%, the other climbing 18% to 26%. First time we named 'em on the Q1 call: Microsoft and Meta, both decade-plus partners. Which is which? We didn't specify. We'd like to add one, maybe two more 10%-ers this year - demand's certainly there, shipments are the question, that's the way to look at it.