AMD touts Meta as a data center chip customer, as it targets Nvidia with new chips like MI200, a family of server accelerators to boost ML and other workloads
AMD's Meta customer win marks an early foothold for its Instinct accelerator line in a data-center account long associated in this story with Nvidia-targeted ML competition. That foothold later broadened: Meta was among the companies planning to use AMD's MI300X accelerators, and it subsequently agreed to acquire up to 6GW of AMD Instinct GPUs in a much larger supply arrangement.
Meta's compute strategy was not exclusively external: related coverage says it planned to deploy second-generation in-house Artemis chips alongside commercially available GPUs. That makes AMD's role part of a broader effort by a major buyer to combine supplier silicon with custom hardware.
First-order effects
AMD gains Meta as a named data-center customer for MI200, giving its server-accelerator push a concrete deployment prospect against Nvidia.
Meta adds AMD as a supplier option for ML and other data-center workloads, rather than relying on a single accelerator vendor.
Second-order effects
Nvidia faces more direct competition within Meta's accelerator purchasing, while AMD must translate a customer announcement into hardware and software adoption at scale.
Meta's use of commercial GPUs alongside its own Artemis program gives it more leverage to allocate workloads across external and internally developed chips.
Third-order effects
If large cloud buyers keep pairing custom silicon with multiple merchant GPU suppliers, accelerator competition will be decided by how well hardware fits heterogeneous data-center fleets rather than by a single-vendor standard.
The later progression from MI200 to MI300X adoption and a major Instinct supply agreement indicates that customer relationships can compound into longer-term AI-capacity commitments.
The trend:AI infrastructure buyers are building heterogeneous compute estates that combine custom chips with competing merchant accelerators to diversify capacity and supplier leverage.
Amazon, Google, Microsoft. And now: Meta. AMD CEO Lisa Su tells @emilychangtv she's excited to count these giants as customers because they're doing “such unique things”... like building the metaverse. But, be patient: it's “not going to be one big bang” https://www.bloomberg.com…
Sleep on Lisa Su at your own peril. $AMD has signed a contract with $MVRS / $FB to build out their metaverse. AMD now has contracts with Meta / Facebook, $MSFT, $AMZN and $GOOGL. Once the deal with $XLNX closes, AMD will be a monster. https://www.cnbc.com/...
Competition is heating up in datacenter GPU land. Here's AMD's just-announced MI200 GPU compared to Nvidia's top of the line A100 https://www.youtube.com/... https://twitter.com/...
I think the deal with Meta (facebook) was overshadowed by product launches. They were a Tier1 Intel partner for YEARS, even getting their own CPU recently (cooper lake iirc with Bfloat16). The fact AMD stole them is amazing.
“The new AMD chip is designed to take on NVIDIA's A100 chip, which, along with other chips designed to speed up artificial intelligence, has helped make NVIDIA the most valuable US-listed semiconductor company.” Maybe MI200 will be prioritized in pytorch? https://www.reuters.com/…
Also, the fact that TSMC has so much visibility into semiconductor markets and trends, I wonder how much info sharing is shared between their Tier1's to help forecast growth in markets.
Strong new product offering from @AMD today that continues to position them well for server/datacenter. This market will thrive with increased competition to Nvidia. Investors are also happy, AMD stock is up 10% right now well above all-time highs. https://twitter.com/...
@GerberKawasaki @PalantirVision Apple changed the game with M1 silicon. it is so BIG & far reaching ->we have entered the era of custom silicon. Companies like GOOG with TPU & TSLA with DPU is already following. Hopefully MSFT up the game with SQ series. AMD is ruling the console…