AMD Doubled Data Center Revenue and Guided 2027 to +80%: The Market Paid 6.5% Because HBM4 Now Sets the Margin

• Third-party content rides inside every rack: AMD's own 10-Q states that integrated rack-scale systems depend on other rack-level components from third parties, so a shortage outside AMD's silicon can cap its own shipments.
• Memory is the heaviest input, and AMD's MI450 carries 50% more HBM4 per accelerator than Nvidia's Rubin, 432GB against 288GB.
• Management is passing the bill on rather than absorbing it: Lisa Su said other component costs are seeing inflationary pressure and that AMD is working with customers so those costs are shared.
• Operating leverage, not gross margin, carries 2027: Jean Hu said operating expenses should grow at less than half the top-line rate, producing \u201cvery significant operating margin expansion in 2027.\u201d

• Memory is the tightest link: SK hynix expects demand to outrun supply through 2030, has long-term agreements with 14 major customers covering more than half of 2027-2028 capacity, and expects HBM4 to pass HBM3E in fourth-quarter volume and revenue.
• Samsung Electronics is described as the priority HBM4 supplier for AMD's MI450, while SK hynix holds the high supply share at NVIDIA. AMD's allocation is the contested one.
• Celestica designs and manufactures the scale-up networking switches inside Helios on Open Compute Project and Open Rack Wide form factors; revenue rose 83% to $5.83 billion at a record 8.5% adjusted operating margin, and the 2026 outlook was raised.
• AMD's own 10-Q warns that a small number of customers account for a significant share of revenue and receivables, so the second-source mandate concentrates both revenue risk and allocation risk.

We are also working with our customers to ensure that those costs are shared as we go forward.

• December quarter is the near-term test: about $17 billion of revenue with non-GAAP gross margin near 55% and Helios deployments stepping up substantially.
• Memory cost is the swing factor: SK hynix is taking M15X to maximum capacity by year-end and expects HBM4 to outgrow HBM3E on volume and revenue, which sets the price AMD pays for content.
• Cash conversion has to improve: free cash flow of $554 million was 4% of revenue in Q3, and AMD's 10-Q flags excess-inventory and prepayment risk if demand is overestimated.
• Competitive arithmetic: NVIDIA still converts 75% of revenue to gross margin and its Rubin line carries 288GB of HBM4 per GPU against 432GB on AMD's MI450, less memory cost per unit.
• Supply chain check: TSMC guides December revenue to $52.3 billion to $54.3 billion, and Celestica is manufacturing the scale-up switches, so both confirm that rack volume, not accelerator demand, is the binding constraint.

• 2027 requires about $94 billion of revenue on our arithmetic, more than double the company's own long-term growth target, with data center above 80% of the mix.
• The gross-margin math has to hold near 55% while mix shifts to racks; if the incremental dollar keeps printing near 51% gross margin, opex leverage below half of revenue growth is the only route to double-digit operating margin.
• Allocation risk is the thesis risk: AMD's HBM4 comes from a contested pool, with Samsung Electronics reported as priority supplier for MI450 and SK hynix holding NVIDIA's Rubin allocation, and none of it disclosed in AMD's guidance.
• Concentration compounds: the same 10-Q says a small number of customers drive a significant share of revenue and receivables, and the Helios adopters are the same buyers.
• Franchise optionality: 240-plus Venice commitments, the Helion open reference architecture and Celestica's scale-up switch design give AMD content in the rack beyond its own GPUs, and if that converts into switch and NIC share, mix improves rather than dilutes.

• Long the bottleneck: HBM4 supply from SK hynix, Micron and Samsung Electronics, leading-edge foundry from TSMC and scale-up switching from Celestica all get paid on AMD's rack volume regardless of which accelerator wins the socket.
• Own AMD for volume rather than margin: the guide is credible on revenue and thin on incremental gross margin and cash conversion, so the 2027 case rests on opex discipline, and the stock's 8% reaction has already discounted some of that optimism.
• Our view on the second-order effect: hyperscalers that own a qualified second source will use it to negotiate harder on price, which compresses the accelerator vendor's margin before it compresses the supplier's, and that pressure shows up in guidance first.