Getting your news
Attempting to reconnect
Finding the latest in Climate
Hang in there while we load your news feed
AMD
Data center news, project activity, and monthly briefings for AMD.
AMD · Construction & power moves · 1
full tracker →Where AMD is securing land and power — each traced to primary filings.
Editor's picks
-
Assessing Sovereign AI: A Two-Pronged Framework
The article argues that sovereign AI is not a binary choice but a spectrum of partial sovereignty, and it proposes a two-pronged framework for assessing why states pursue it and how they build it across the AI stack.
- The piece compares five country cases — the United States, China, India, Singapore, and France — and maps their motivations across layers such as chips, compute, cloud, models, data, and applications.
- It highlights policy issues including interoperability, regulatory divergence, and the economic tradeoffs of building fragmented AI stacks; no single new project or deal is announced in the article, which is primarily an analytical commentary based on national documents and prior statements.
-
AWANI Talk Series: Friends to All, Enemy to None | Can Malaysia stay neutral in a polarised world?
Astro AWANI hosted a panel discussion on Malaysia’s foreign policy approach of “active non-alignment” and whether the country can remain neutral in a more polarized world.
- The discussion focused on Malaysia’s neutrality, trade and investment attraction, human rights, and ASEAN centrality, with panelists debating whether Malaysia can stay “friends to all, enemy to none” under rising geopolitical tensions.
- Speakers referenced China-US trade tensions, South China Sea, Myanmar, Cambodia-Thailand, Gaza, Ukraine, and Malaysia’s national action plan on business and human rights; they also argued that ASEAN is constrained by national interests, a weak Secretariat, and rotating chairmanship.
- This appears to be an analysis/opinion panel, not a first-time policy announcement, although it cites Prime Minister Anwar Ibrahim’s term “active non-alignment” and discusses recent government positions and reforms.
-
AT&T and Microsoft scale trillion-token workloads with Microsoft Foundry and AMD
AT&T has announced that it used Microsoft Foundry Managed Compute to build and scale its OTel2.0 telecom-focused AI models.
- AT&T used approximately 530 GPUs through Microsoft Foundry Managed Compute, including 430 AMD Instinct™ MI300X GPUs, to support OTel2.0 development across multiple GPU architectures.
- The company processed about 1T tokens for OTel2.0 and says using open-source models like Phi-4 for synthetic data generation saved tens of millions of dollars versus frontier models; Microsoft also said deployment could happen in days rather than weeks.
- AT&T deployed models including Phi-4, OSS-120B, and Gemma-4 for synthetic data generation, data preparation, reasoning workloads, and broader model development.
-
Edge AI Platforms
AMAX has announced its Edge AI platform portfolio and supporting services for moving edge AI systems from prototype to production.
- The portfolio includes 1U and 2U edge servers for low-latency inference, industrial compute, medical imaging, smart vision, and distributed enterprise workloads.
- Named products include ServMax X-110, ServMax X-114, ServMax A-1112T, ServMax X-216, and AceleMax AXG-2164IB, with the latter supporting up to four PCIe GPUs.
- AMAX also describes services covering product engineering, GPU system design, OEM customization, thermal & power validation, factory testing, manufacturing & QA, global deployment support, and lifecycle management.
-
What the OCI MSA Didn't Solve for AI Scaling
Scintil Photonics’ CEO Matt Crowley argues that the architecture for optical scale-up has been settled by the OCI MSA and that manufacturing — specifically heterogeneous integration — now determines who can scale beyond four wavelengths per fiber.
Main announcement/action: The article presents the argument that the Optical Compute Interconnect MSA (formed earlier this spring by AMD, Broadcom, Meta, Microsoft, NVIDIA, and OpenAI) settled on an NRZ modulation + wavelength multiplexing architecture (starting at four wavelengths per fiber), and that the remaining challenge to move to 8, 16+ wavelengths is industrial (manufacturing) rather than architectural. It identifies heterogeneous integration (bonding III-V gain material to silicon photonics wafers) as the manufacturing pattern that provides wavelength-scaling headroom and cites SHIP™ on Tower Semiconductor 200 mm lines and LEAF Light™ demonstrated in 8- and 16-wavelength configurations (with NVIDIA among Series B investors) as production proofs.
Background and details: The piece contrasts prior eras (discrete lasers; silicon photonics with off‑wafer lasers) with the current heterogeneous integration era, notes that discrete-laser assembly scales poorly for hyperscale (e.g., a 16-wavelength source multiplies lasers and alignments across fibers), and references OFC 2026 where multiple vendors requested SHIP™ extensions across device categories. It emphasizes adding wavelength-scaling headroom as a line item on supplier evaluation sheets and states that teams delaying this consideration will need to redesign across two generations.
-
Economic Consequences of Section 232 Tariffs on Semiconductor Imports
The Trump administration issued an executive order in January 2026 imposing a 25 percent Section 232 tariff on certain advanced semiconductors.
- Main action: The January 14, 2026 executive order imposed an immediate 25 percent ad valorem duty on a narrow category of advanced computing chips and signaled a possible broader phase 2 at a “rate of duty that is significant.” ITIF models that a sustained 25% tariff would produce a cumulative $1.6 trillion loss in U.S. GDP (3.9%) over 10 years, reduce ICT consumption by $12.5 billion, and lower GDP per capita by $170 in year 1 and $4,825 cumulatively by year 10. The United States imported $48.1 billion of semiconductors in 2025 (baseline used in analysis).
- Background and recommendations: ITIF recommends removing blanket semiconductor tariffs, extending the investment tax credit (ITC) — now 35% — through 2030 and expanding it to semiconductor research and design, and requiring future Section 232 tariffs to include annual reviews and automatic expiration. The report also urges that tariff offsets under phase 2 be made available to firms engaged in semiconductor R&D and AI/data-center investment. The ITC is noted as scheduled to expire at the end of 2026.
Recent news
-
Landlords flex rental power as data center viability becomes ‘everybody’s problem’
Facilities Dive reports that rapid data center development is changing lease negotiations, with landlords and lenders pushing for more favorable contract terms and greater attention to financing and future upgrade needs.
- CBRE said primary market supply in H1 2026 rose 33.7% YoY to 10,903 MW, while vacancy fell to 1.4% as new capacity was absorbed immediately; under-construction capacity rose 24.8% to 7,481 MW.
- Peter Bergan of Vinson & Elkins said hyperscale and neocloud leases are becoming more finance-driven, with termination rights being removed or limited and developers focusing on shell design, power, and rack/load capacity for future growth.
- The article cites CBRE’s midyear report, released Aug. 27, and discusses North American market conditions, including community resistance, zoning delays, and power/fiber constraints.
-
Nvidia, MediaTek Bring Custom Chips to AI Racks
Nvidia has announced that it is opening its rack-scale AI infrastructure and NVLink Fusion technology to custom accelerators from hyperscalers and other AI developers, and it has expanded its partnership with MediaTek to support pre-validated custom XPU designs.
- The platform lets customers design custom XPUs on a pre-validated NVLink Fusion foundation and integrate them into Nvidia-connected AI factory infrastructure built around the MGX ecosystem.
- Nvidia also invested $3.5 billion in convertible bonds issued by MediaTek; the companies did not name any customers or announce specific data center deployments.
- The broader context is Nvidia’s effort to make NVLink interoperable beyond its own processors, while rival approaches include Ethernet and UALink; the MediaTek partnership also extends to local AI computing and automotive systems.
-
Private AI cloud, agentic infrastructure dominate VMware Explore
Broadcom has announced VMware AI Factory, AgentMinder, and new TrueSource open-source security services at VMware Explore in Las Vegas.
- Broadcom said VMware AI Factory is a packaged AI infrastructure stack built on VMware Cloud Foundation, supporting Nvidia and AMD accelerators, certified servers from Cisco, Dell, Lenovo, Supermicro, and more than 150 models.
- Broadcom also announced AgentMinder for AI-agent governance, upcoming AI gateways and agentic security features across VCF Private AI Services, Tanzu Platform, Avi Load Balancer, and vDefend, plus TrueSource Trusted Artifacts and TrueSource Data Services for open-source software security; several offerings are available today, while others are coming later this year or were already part of VCF.
- The article is a news report and analysis based on Broadcom statements and media briefings, not a standalone company press release; it also mentions Flexential building offerings on top of VMware AI Factory early next year.
-
Caught In The GPU Pricing Paradox, Can ESDS Deliver A Bumper IPO?
ESDS Software Solution has announced a ₹720 Cr IPO to fund expansion of its data-centre infrastructure and AI compute capabilities.
- The Nashik-based company’s fresh issue IPO opens tomorrow with a price band of ₹408 to ₹429 per share; ESDS plans to use proceeds for data-centre expansion, GPUs and servers.
- The company said server prices rose from about ₹25 Lakh when it filed its draft papers to about ₹1.8 Cr by the red herring prospectus stage, and that GPU, storage, firewall and networking costs also increased; it also disclosed a $1.25 Bn agreement with Australia’s Sharon AI for an 8,000-GPU cluster under a seven-year contract.
-
Reports: Google partnering with AMD for next-gen hybrid TPU
Google is reportedly working with AMD on the design of a next-generation TPU for AI and conventional computing workloads.
- The reported hybrid accelerator could integrate CPU cores directly into the TPU package, with a focus on reinforcement learning and agentic AI workloads.
- The article says Google already uses TPUs alongside standard x86 CPUs in servers, while AMD recently announced a partnership with Cerebras to combine processors for AI inference.
- This is a report about a rumored/industry-reported design effort, not a first-party Google or AMD announcement.
- Analyst Jon Peddie said the idea “makes good sense” because AMD has experience with custom APUs and Google “needs diversity.”
-
Data Center Hardware Highlights: August 2026
Data Center Knowledge has highlighted a shift in AI hardware coverage from GPUs toward infrastructure-first constraints across energy, minerals, grid access, and networking.
- The roundup says AI buildouts are increasingly limited by power, copper, grid interconnects, transformers, turbines, and permitting, while rack-scale integration and the network supercycle are becoming more important.
- The article also references several reported developments, including Nvidia’s $500B infrastructure bet, TSMC’s $265B Arizona campus expansion, and AMD’s Helios rack-scale AI system; it is commentary and synthesis rather than a single first-time corporate announcement.
-
Server prices to rise by up to 87% at OVHcloud
OVH has announced server and cloud price increases, with some gaming and bare-metal offerings rising sharply because of higher RAM and storage costs linked to AI demand.
- New prices take effect on Sept. 1 for new orders and Oct. 1 for renewals; the latest gaming servers rise 87%, High Grade bare metal servers rise 59%, and older 2024-spec High Grade models rise 26%.
- OVH also said optional extras are increasing: RAM and disks for new orders rose on July 1 by 127% and 89%, while from Oct. 1 renewals face further increases; existing customers can prepay by Oct. 1 to lock current prices for up to four years on servers already in production.
-
IT infrastructure shortages are real and lasting. Here’s how to cope
Network World has published an analysis of IT infrastructure shortages affecting enterprise AI, networking, storage, and compute planning.
- The article cites analysts and vendors saying memory shortages are driving lead-time extensions to 6, 12, or 18 months, with memory prices up 50% to 200%, PC prices up 35% to 45%, and some server prices up over 125%.
- It also reports enterprise network equipment pricing is projected to rise over 20% in 2026 and by a further 3% to 5% into 2027, with no price reduction expected until end-2027; examples and advice are drawn from Gartner, TD SYNNEX, WWT, Backblaze, Cisco, and Equinix.
-
Arm Chips May Get Their Own Virtual RAN Boost
Arm is described as expanding beyond smartphone chip licensing into AI, data centers, and mobile networks.
- The article says Arm’s latest earnings release lists vehicles and robotics, with AI as a major target, and notes Arm has moved into the data center by starting to make its own CPUs this year.
- It also says Nokia has shifted from Intel to an Arm-based CPU made by Marvell Technology for Layer 2 plus in 5G RAN, while Marvell supplies a custom chip with Arm technology for Layer 1.
- This is an analysis/commentary piece referencing prior developments rather than announcing a first-time deal or product launch in the article itself.
-
New Data Center Developments: August 2026
This article is a curated monthly roundup of data center news, summarizing multiple announcements, reported deals, and policy actions across regions.
- It highlights several new project announcements and construction updates, including Meta’s 5 GW Hyperion expansion, OpenAI’s Project Camellia in Georgia, Iron Mountain’s 48 MW RCH-1 break ground, and Meta’s 1 GW Canadian campus in Alberta.
- It also covers policy and market developments such as New York’s permit pause for sites over 50 MW, Texas’s ERCOT queue audit, North Carolina’s tax change, Nebraska’s incentive block, plus large capital figures including €1.5 billion, $38.4 billion, $500 million, $400 million, and $10 billion across various projects.
-
AMD unveils AI GPU to challenge Nvidia’s Rubin
AMD has announced the Instinct MI455X, its new flagship AI accelerator, at its recent Advancing AI 2026 event, alongside the Helios rack-scale AI platform.
- The MI455X is the flagship of AMD’s Instinct MI400 family and is designed for large-scale AI training and inference; AMD says it delivers 320 billion transistors, 432GB HBM4, up to 40 PFLOPS FP4, and 23.3 TB/s of memory bandwidth.
- The chip uses CDNA 5, is manufactured with TSMC 2nm and 3nm process technologies, and is intended to improve tensor processing, cache bandwidth, memory movement, and execution efficiency for LLM and agentic AI workloads; shipping begins in Q3 2026, with volume ramping through Q4 2026 and H1 2027.
-
Hyperscalers Say AI Race Has Entered a New Phase
This article is an analysis/commentary on second-quarter earnings, not a first-time announcement from a single company.
- Microsoft emphasized execution speed, saying it opened 31 data centers in the quarter, 88 during fiscal 2026, and added roughly 1 gigawatt of AI capacity while cutting dock-to-live times by nearly 50%.
- Alphabet raised its 2026 capex outlook to $195 billion-$205 billion and said about 60% of capex goes to servers, with the rest to data centers and networking; Meta said it spent $31.1 billion on capex in the quarter and kept full-year guidance at $130 billion-$145 billion while building capacity through 2027.
-
Core Scientific Doubles AI Capacity to 1.1 GW in $14B AMD Deal
Core Scientific has announced a 15-year infrastructure agreement with AMD to expand its leased AI data center capacity to approximately 1.1 GW.
- The agreement covers 530 MW across five campuses and represents more than $14 billion in potential base contracted revenue; it begins in 2027 and gives AMD exclusive rights to reserve up to an additional 2 GW of future capacity.
- Core Scientific said the broader contracted AI portfolio now totals approximately 1.1 GW and more than $24 billion in potential contracted revenue; deliveries will begin at the Pecos, Texas campus in the first half of 2027, with development also advancing in Hunt, Texas; Muskogee, Oklahoma; Auburn, Alabama; and Dalton, Georgia through 2028.
-
Intel Layoffs Show Data Center Workers Aren’t Immune to Corporate Job Cuts
Intel has announced it is preparing another round of layoffs in its data center group while saying product commitments and roadmaps will not change.
- First reported by The Oregonian, Intel notified employees in its data center group about planned cuts; the company did not disclose the number of positions or whether Oregon employees would be affected.
- Intel said the reorganization is part of a broader strategy to become a more focused and efficient company and to ensure the group has the right roles and skills for long-term success; Intel also reported Q1 2026 data center and AI revenue of $5.1 billion, up 22% year over year.
- The article also notes broader industry layoffs at Oracle and Microsoft amid heavy investment in AI data center expansion, and includes analyst commentary from IDC about AI being used as a pretext for job cuts.