Analysis
CAPACITY TEST:

AMD Drops HBM For LPDDR5X In Versal Memory Package Shift

Newsroom brief

AMD is moving its Versal Premium Gen 2 memory-on-package adaptive SoCs from HBM to LPDDR5X after HBM2e supply limits forced the earlier Versal HBM family toward discontinuation.

Verified against source materialEdited by SendTech Times Chips & Compute Desk
AMD Drops HBM For LPDDR5X In Versal Memory Package Shift
Image source: ServeTheHome

AMD Moves Versal Memory Packages To LPDDR5X

AMD is introducing a Versal Premium Gen 2 Memory on Package lineup that puts LPDDR5X memory on the chip package instead of continuing the HBM path used by earlier programmable-logic products.

ServeTheHome described the AMD devices as adaptive SoCs for FPGA customers that want a more integrated product and faster development than designs using off-package memory.

The launch is notable because AMD's programmable-logic group had previously used HBM to give Xilinx and AMD customers more bandwidth inside tightly integrated packages.

The new lineup changes that memory choice rather than only adding another Versal variant.

AMD is making the shift as AI hardware demand has tightened HBM supply for other product lines.

The account says AMD's Instinct accelerators and EPYC CPUs benefited from the AI boom, while the programmable-logic division lost access to the HBM2e supply needed for Versal HBM devices.

The product is aimed at a different constraint from AMD's AI accelerator portfolio.

Programmable-logic customers often buy devices for long support windows and embedded designs, so a memory technology that disappears quickly can become a procurement risk even when it offers strong bandwidth.

The LPDDR5X move gives AMD a way to keep an on-package memory option available after older HBM2e supply dried up.

HBM2e Supply Forced A Product-Line Deadline

ServeTheHome reported that ServeTheHome reported that Xilinx offered Virtex UltraScale+ HBM FPGAs from 2018 with up to 16GB of HBM memory.

ServeTheHome reported that ServeTheHome reported that AMD later sold Versal HBM adaptive SoCs with up to 32GB of HBM2e memory onboard.

Those products targeted customers that needed both integrated memory and higher bandwidth than ordinary DDR memory could reasonably provide.

That long-life product model became harder once HBM supply shifted.

The account says AMD had already announced support for normal UltraScale+ FPGAs until 2045, but on-package memory products also depend on memory suppliers.

ServeTheHome said SK hynix supplied the first generation of HBM, while Samsung and Micron joined with HBM2.

The memory cycle advanced on a shorter timeline than AMD's usual FPGA support horizon.

ServeTheHome said HBM went through three major specifications in its first nine years, and older HBM types were phased out on shorter cycles than DDR memory as high-performance processors drove demand for newer generations.

Customers Still Need Deployment And Availability Detail

The trade article says AMD began discontinuing the Versal HBM family in September 2025 because HBM2e was no longer available at the scale needed for the product.

The new LPDDR5X version arrives on the final day to place orders for Versal HBM chips, making the memory-package switch a supply-chain response rather than a routine portfolio refresh.

The shift also keeps AMD exposed to a narrower proof point.

LPDDR5X can make the package more self-contained for adaptive SoC customers, but it is not the same bandwidth story as HBM.

AMD has to persuade FPGA buyers that integration, availability and product life matter more for this use case than staying with scarce HBM2e.

The public details still leave commercial evidence missing.

AMD did not disclose named customers, order volumes, pricing, shipment dates, production-ramp capacity or independent benchmarks for the Versal Premium Gen 2 Memory on Package chips.

Share this article
inXf

Related articles

More
SK hynix Expert Frames AI Chip Race Around Power, Water And Megafabs
Chips & Semiconductors

SK hynix Expert Frames AI Chip Race Around Power, Water And Megafabs

An SK hynix Newsroom expert column argues that AI semiconductor competition is moving beyond faster chips toward megafab capital, grid delivery, industrial water and memory supply stability.

Micron Rally Signals AI Memory Shortage In Data-Centre Buildout
Chips & Semiconductors

Micron Rally Signals AI Memory Shortage In Data-Centre Buildout

Micron's market rally is being driven by AI data-centre memory demand, with TechCrunch citing a $1.27 trillion market cap, a 236% one-month stock gain and supply shortages expected to last into 2027.

Micron And Anthropic Put AI Memory Supply Inside Claude’s Compute Plan
Chips & Semiconductors

Micron And Anthropic Put AI Memory Supply Inside Claude’s Compute Plan

Micron and Anthropic signed a strategic agreement covering AI memory and storage architecture, supply, Claude adoption inside Micron and a Series H investment.

AT&S Backs Malaysia AI Substrate Expansion With Customer Commitments
Chips & Semiconductors

AT&S Backs Malaysia AI Substrate Expansion With Customer Commitments

AT&S plans a customer-backed Kulim expansion for IC substrates and advanced PCBs, naming AMD as one customer as AI hardware demand moves deeper into the semiconductor supply chain.

Keep Reading

More Stories

Latest
Memory Prices Push US PC Shipments Down 7%Chips & SemiconductorsJul 2, 2026Memory Prices Push US PC Shipments Down 7%Omdia data cited by Tom Hardware showed US PC shipments fell to 15.8 million units in the first quarter of 2026, as memory and storage chip shortages hit entry-level laptops and pushed the market toward a forecast 14.4% contraction.Cloudflare Sets September Deadline For Mixed-Use AI CrawlersAIJul 2, 2026Cloudflare Sets September Deadline For Mixed-Use AI CrawlersCloudflare plans to block mixed-use crawlers from ad-supported pages by default from September 15, 2026, unless site owners change the setting. The policy pushes AI companies to separate search access from agent and training uses while Cloudflare expands publisher payment tools.WhatsApp Usernames Hide Phone Numbers But Scam Risk RemainsCybersecurityJul 2, 2026WhatsApp Usernames Hide Phone Numbers But Scam Risk RemainsWhatsApp is rolling out usernames and optional keys to reduce phone-number exposure, but security researchers warn that impersonation and social-engineering scams can move to handles, profile images and trusted-looking accounts.Starlink Discounts Memphis Plans Around xAI Data Centre DisputeCloud & Data CentersJul 2, 2026Starlink Discounts Memphis Plans Around xAI Data Centre DisputeSpaceX is offering Starlink discounts near xAI’s Colossus data centres in Memphis and Southaven, while lawsuits and permit disputes keep attention on power, noise and pollution claims around the AI site.MiCA Deadline Forces EU Crypto Firms To Choose Licences Or ExitCrypto/Web3Jul 2, 2026MiCA Deadline Forces EU Crypto Firms To Choose Licences Or ExitCoinDesk reported that the EU’s MiCA framework is now fully in force, requiring crypto firms serving the 27-nation bloc to hold a licence or stop operating. Industry lawyers and executives said the rulebook improves clarity, but warned that compliance costs could shrink roughly 3,000 registered providers to 300 or 400 licensed firms.UAE Gives Social Platforms 12 Months To Enforce Under-15 RulesCapital & PolicyJul 2, 2026UAE Gives Social Platforms 12 Months To Enforce Under-15 RulesThe UAE says social media platforms must build effective age-verification controls after a Cabinet resolution restricting under-15 access. Technology companies have 12 months before penalties apply, and officials said age-verification data must be deleted immediately rather than stored by platforms.Robinhood Opens Arbitrum Chain As Stock Tokens Go Live In 120 CountriesFintech & Digital PaymentsJul 2, 2026Robinhood Opens Arbitrum Chain As Stock Tokens Go Live In 120 CountriesRobinhood has launched the public mainnet for Robinhood Chain, a Layer 2 blockchain built on Arbitrum, and made Stock Tokens available through Robinhood Wallet in more than 120 countries. The company also introduced Robinhood Earn with an estimated 7% yield on USDG, but jurisdictional availability and control settings remain central limits.Nvidia Names $500 Billion US AI Infrastructure Plan But Leaves Timing OpenCloud & Data CentersJul 2, 2026Nvidia Names $500 Billion US AI Infrastructure Plan But Leaves Timing OpenNvidia says it and partners including TSMC, Foxconn, Wistron, Corning, Lumentum, Coherent and Amkor plan up to $500 billion of US AI infrastructure production. The account comes from Nvidia's own company blog; it names factories, suppliers and job figures, but gives no full production timetable for the programme.Springboards Tests Qwen 3 Model Against Repetitive LLM AnswersAIJul 2, 2026Springboards Tests Qwen 3 Model Against Repetitive LLM AnswersAustralian startup Springboards has built Flint on Alibaba’s Qwen 3 to produce more varied answers to open-ended prompts. MIT Technology Review’s article pairs the company’s claim with a NeurIPS-winning homogeneity paper and user cautions that the prototype still fails under pressure.AWS Announces $1 Billion Forward-Deployed AI Engineering UnitAIJul 2, 2026AWS Announces $1 Billion Forward-Deployed AI Engineering UnitAWS has announced a $1 billion Forward Deployed Engineering organisation that will send small engineering pods into customer environments for about 45 days. TheStreet reported that early users include the Allen Institute, Cox Automotive, the NBA, the NFL, Ricoh and Southwest Airlines.Meta Board Weighs Iran Influence Posts Left OnlinePoliticsJul 2, 2026Meta Board Weighs Iran Influence Posts Left OnlineMeta’s Oversight Board may examine whether Facebook and Instagram should have removed two pro-Iran posts that users flagged as possible co-ordinated inauthentic behaviour tied to state-sponsored influence operations.OCC Public Denials Raise Charter Risk For Fintech ApplicantsFintech & Digital PaymentsJul 2, 2026OCC Public Denials Raise Charter Risk For Fintech ApplicantsThe OCC plans to publish charter denial decisions, giving fintech and digital-banking applicants clearer examples of why filings fail. The guidance also raises the reputational cost of applying before governance, compliance and risk systems are ready.