Meta And AMD Sign 6 Gigawatt Data Center Deal

Meta Scales Compute Supply With A Power-Based Chip Commitment

Meta Platforms said on February 24, 2026 that it reached a multi-year framework with Advanced Micro Devices to deploy up to 6 gigawatts of graphics processing unit (GPU) capacity across next-generation artificial intelligence infrastructure. The arrangement is sized for as much as $60 billion of chip purchases over five years, a demand signal that can move semiconductor earnings expectations and data center capital spending.

Advanced Micro Devices shares rose more than 10% in premarket trading after finishing the prior session at $196.60, reflecting renewed confidence in non-Nvidia data center supply. The first-order market channel is equity repricing in semiconductors and data center hardware, with second-order effects in power costs and credit conditions for issuers funding heavy capex.

A Six Gigawatt Buildout Ties Deliveries To Data Center Timelines

The 6 gigawatt GPU deployment is framed around the electrical footprint of racks and clusters rather than a simple unit count, which makes power availability a gating item for revenue recognition. That framing matters because interconnect capacity, cooling buildouts, and grid queues can delay shipments even when purchase orders are in place.

Shipments to support the first gigawatt are scheduled to begin in the second half of 2026, setting a clear trigger for when the newest accelerators start to flow through reported results. The initial wave centers on a custom MI450-based GPU tuned for inference, the compute phase when a model responds to user requests, where the focus is AI inference compute and energy efficiency per query can dominate total cost.

Custom Silicon And Systems Put Efficiency At The Center

The first deployments pair the MI450-based GPU with sixth-generation AMD EPYC central processing units (CPUs), codenamed Venice, integrated into rack-scale systems built on the Helios architecture. By tying GPUs, CPUs, and software more tightly, the companies are targeting lower total cost of ownership, which can support higher utilization and steadier chip demand through 2027.

The software stack uses AMD’s ROCm platform, and the rack design was developed through the Open Compute Project to standardize components and reduce supply friction. Systems validation can lift pricing power if performance-per-watt targets hold during 2026 qualification cycles.

A Warrant Structure Adds Equity Optionality And Dilution Math

Alongside purchase commitments, the agreement includes a performance-based warrant covering 160 million AMD shares with an exercise price of $0.01, linking equity upside to commercial execution. The warrant can vest in tranches as delivery and technical milestones are met, and Meta has the option to build a stake of up to about 10% under the framework.

To Sign each tranche of the warrant, Meta must clear stock-price hurdles stepping up toward $600 per share, with an expiry running to 2031. For shareholders, that creates a parallel valuation driver: dilution timing, index flows, and hedging demand tied to when vesting conditions are satisfied.

Capex, Rates, And Power Inputs Become The Macro Transmission

The agreement lands as mega-cap capital expenditure remains elevated, with combined spending by Alphabet, Amazon, Meta, and Microsoft projected at least $630 billion in 2026, much of it directed to data centers and accelerators. Persistent buildouts can firm demand for electrical equipment, cooling, and copper-intensive infrastructure, while raising sensitivity to regional power prices and grid constraints.

If spending stays near current trajectories through late 2026, the rates channel can run through stronger investment demand and sticky construction costs; that can keep longer-dated Treasury yields responsive to growth surprises. If spending slows, investment-grade credit spreads can widen for hyperscalers on slower payback math even as some utilities benefit from steadier contracted demand.

Competition Broadens Beyond A Single Vendor

Meta said it will keep a multi-vendor approach and continue scaling in-house accelerators alongside external GPUs, which reduces single-supplier risk but increases pricing pressure across the chip stack. It Spotlights the shift toward inference capacity as a key profit pool, where software maturity and power efficiency can matter as much as peak throughput.

The arrangement also signals that large buyers are willing to embed equity optionality in supply deals, a feature Tredu flags as capable of amplifying swings when demand surprises. For chip peers, a faster AMD ramp can tighten near-term supply in advanced packaging and memory, influencing component pricing into 2027.

Base Case, Upside Case, And Downside Case Into 2026–2031

Base case: deliveries begin in the second half of 2026, AMD ramps the first gigawatt, and the remaining capacity scales with data center power availability, keeping revenue back-end loaded into 2027–2029. Equity markets price a steadier cadence of data center sales if early MI450 milestones clear on schedule.

Upside case: faster site bring-up in 2027 pulls forward deployment above one gigawatt, lifting cash flow visibility and compressing semiconductor risk premia. That outcome would likely tighten high-grade credit and support cyclical industrial inputs tied to data center construction.

Downside case: qualification delays, software bottlenecks, or power constraints push meaningful volume into 2028 and intensify pricing competition, slowing margin expansion. In that scenario, chip equities can re-rate lower on longer payback periods, and implied volatility rises as investors reprice the probability of reaching the higher warrant tranches.