AI Chips & Infrastructure

Taiwan is reportedly planning to invest billions of dollars in developing industrial parks in the US, with a focus on building a comprehensive supply chain for major players like Nvidia Corp. and other artificial intelligence hardware makers. The initiative is starting with a complex in Texas that would cost around $4 billion to build, according to Taiwan’s Minister of Economic Affairs Kung Ming-hsin.

The facilities are expected to help smaller suppliers, which are essential to the operations of Nvidia’s main production partner, Taiwan Semiconductor Manufacturing Co., and AI server assemblers like Foxconn Technology Group. These companies will be able to set up shop in the US, thanks to the new infrastructure. This move is seen as a strategic step towards strengthening ties between Taiwan and the US.

Taiwan is also tightening export controls on artificial intelligence chips to align with US restrictions. The country plans to add AI servers and cutting-edge semiconductors to its list of controlled items under the Foreign Trade Act. Shipping restricted AI hardware to China will become a criminal offense after the measures take effect, according to Minister Kung. This move is part of Taiwan’s efforts to comply with the Agreement on Reciprocal Trade signed with the US this year.

The development comes as several Taiwanese companies have recently received approval for significant investments in the US. Quanta Computer was approved to spend $973 million to support assembling servers and peripheral equipment in Tennessee, while Wistron was given the green light to invest $150 million in a U.S. subsidiary that assembles circuit boards. The company also plans to spend another $146 million on renovating a Texas plant.

TSMC’s planned investment of over $265 billion in the US is separate from Taiwan’s economy minister’s announcement last month that Taiwanese companies are planning an additional $20 billion of investments in the country, driven by strong demand for AI applications and semiconductor products. Nvidia relies heavily on suppliers like TSMC, SK hynix, Samsung Electronics, and Micron Technology for chip manufacturing and memory, while its major customers include Microsoft, Amazon, and Meta Platforms.

Hp has launched a mobile workstation that challenges the idea that serious ai models can only run in cloud environments. The hp zbook ultra g3a is built around amd’s ryzen ai max pro processor and packs up to 192 gb of unified memory, with a significant portion – 160 gb – allocatable to the gpu.

A shift towards running large-scale ai models locally on laptops or desktops could have far-reaching implications for design teams. They might be able to work more efficiently without relying entirely on cloud-based solutions. The conversation around ai in architecture, engineering, and construction has long centered on cloud-based options.

David mcafee from amd spoke with hp about the new workstation during its launch event. One key topic was how a local ai model would compete with other resource-intensive applications running at the same time. McAfee noted that performance can be affected when multiple tasks vie for resources, but he assured us that this isn’t insurmountable.

The zbook ultra g3a uses lpddr memory efficiently, but it cannot be upgraded later, which may become an issue as ai requirements continue to shift. Companies need to carefully consider their hardware needs when planning for the future and expect those needs to change over time. This mobile workstation is a testing ground for how businesses will approach hardware procurement.

McAfee’s conversation with hp also covered amd’s work on professional drivers for radeon gpus and ai-assisted coding. The company is developing drivers that can take advantage of its rocm platform, making it easier to port applications built for nvidia’s cuda to amd’s architecture. This could be a game-changer for businesses looking to adopt ai tools.

AMD envisions a future where users have multiple options for compute depending on their specific needs and workflows. They’re exploring everything from high-end desktops to cloud-based solutions, with the hp zbook ultra g3a sitting somewhere in between. The company sees this as an opportunity to provide more flexibility and choice for customers.

The discussion around bringing ai on-device also touched on data analysis tools and how they can be used by companies that don’t necessarily need a full-fledged local model running all the time. Llm models, in particular, could serve as judges or decision-makers within larger workflows, allowing businesses to get insights without having to set up complex cloud-based solutions.

A major expansion of their long-standing partnership has been announced by Amazon Web Services (AWS) and NVIDIA. The two tech giants are committing to deploy an additional 2 million NVIDIA graphics processing units (GPUs) across AWS’s global infrastructure by 2028, significantly boosting the capabilities for machine learning jobs and AI development.

The move marks a significant milestone in the companies’ 16-year collaboration, which has already led to numerous innovations in the field of artificial intelligence. The partnership extends beyond just GPUs, with NVIDIA Vera CPUs being integrated into AWS’s platform as well. This will provide customers with a new compute option specifically designed for agentic AI workloads.

The massive influx of hardware signals a new phase in the race to build more powerful AI capabilities. To achieve this goal, the two companies are deepening their work across various areas, including AI factories, CPUs, networking, and robotics. Their aim is to deliver co-engineered solutions that can accelerate AI development at an unprecedented scale.

The deployment of 2 million additional NVIDIA GPUs will be a crucial factor in achieving this goal. This move complements Amazon’s existing custom silicon efforts, offering customers a wider choice for their AI workloads. Whether it’s NVIDIA’s specialized hardware or AWS’s Trainium chips, users can now select the best option to suit their needs.

The partnership also includes deploying 100,000 NVIDIA GPUs on secure AWS infrastructure specifically designed for federal and national security workloads classified at Impact Level 6 (IL6) and above. This initiative addresses global demand and competition in AI development, ensuring that sensitive information is handled securely.

By expanding their collaboration, AWS and NVIDIA are taking a significant step forward in the field of artificial intelligence. Their commitment to delivering co-engineered solutions will undoubtedly accelerate AI development at an unprecedented scale. As the companies continue to push the boundaries of what’s possible with AI, it remains to be seen how this partnership will shape the future of machine learning jobs and beyond.

OpenAI has announced the results of its first custom inference chip, Jalapeño. The company claims that this new architecture delivers industry-leading speed and efficiency in artificial intelligence (AI) inference tasks. According to OpenAI, Jalapeño can serve more AI work per unit of power while also returning responses more quickly than existing hardware systems.

The performance gains offered by Jalapeño are significant. In tests conducted on three public models - GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T - the chip delivered a better combination of throughput, power efficiency, and latency compared to leading commercially available AI systems. Specifically, it achieved 1.5 to 1.9 times more AI work per watt at peak throughput and 1.7 to 3.6 times lower end-to-end latency than the comparison systems.

The benefits of Jalapeño’s performance are not limited to just one or two specific models. The chip was designed from the start with a focus on serving modern and future language models, particularly interactive agents. As such, it excels in both prefill and decode phases, which have different bottlenecks. By minimizing data movement and communication delays, Jalapeño ensures that model state can be explicitly placed and kept local while activating the right combination of compute, memory, and networking for each inference phase.

Jalapeño’s development was also facilitated by AI itself. The team used OpenAI models to explore implementations, shorten design, measurement, and verification loops, and continuously iterate on model workloads. This allowed them to move from initial design to tapeout in just nine months. Additionally, AI helped optimize the chip’s arithmetic circuits, enabling more compute performance to be fit into the chip on schedule.

The architecture of Jalapeño is designed as a clear, predictable programming target for both humans and AI. Engineers can describe work through local tensors, explicit communication, and predictable synchronization, making it easier for AI to tackle parallel programming tasks. This flexibility was demonstrated when the team used Codex with GPT-Astra to bring three open-weight models not part of Jalapeño’s original production plan to high performance within two months.

The results from these tests are impressive. On Kimi K2.5 1T, for example, Jalapeño delivered approximately 1.5 times higher peak performance per watt and 3.4 times lower end-to-end latency than the comparison system. In internal testing, the advantage of Jalapeño widened further on frontier OpenAI models, suggesting that its architecture becomes more valuable as workloads grow larger and more demanding.

Jalapeño’s impact extends beyond just improving performance in AI inference tasks. By producing more useful work from the same power and hardware, it can help serve more demand and lower the cost of delivering a successful result. This is particularly important for OpenAI, which aims to ensure that artificial general intelligence benefits all of humanity.

The deployment of Jalapeño within OpenAI’s compute infrastructure is expected to begin by the end of this year. It marks the first generation of a multigenerational roadmap, with Gen 2 already in development and Gen 3 taking shape. Each subsequent generation will build on what has been learned and further advance both efficiency and speed.

The growing demand for AI requires more compute from every available source. To meet this need, OpenAI plans to continue deploying accelerators from NVIDIA and other partners for both training and inference workloads. The company’s mission remains unchanged: to ensure that artificial general intelligence benefits all of humanity.

Higher peak mixed TPS/kW ≈1.9×

85,448 vs. 44,960 mixed / kW

Lower end-to-end latency ≈1.7×

1.03 s vs. 1.80 s

Lower min TBT ≈2.7×

0.69 vs. 1.87 ms (1,459 vs. 535 tok/s/user)

More throughput at previous TBT ≈53.7×

22,935 vs. 427 mixed / kW (at 535.28 tok/s/user)

Higher peak mixed TPS/kW ≈1.7×

19,641 vs. 11,781 mixed / kW

Lower end-to-end latency ≈3.6×

1.65 s vs. 5.99 s

Lower min TBT ≈4.1×

1.43 vs. 5.90 ms (700 vs. 169 tok/s/user)

More throughput at previous TBT ≈104.3×

12,258 vs. 118 mixed / kW (at 169.41 tok/s/user)

Higher peak mixed TPS/kW ≈1.5×

18,195 vs. 11,862 mixed / kW

Lower end-to-end latency ≈3.4×

1.56 s vs. 5.31 s

Lower min TBT ≈3.8×

1.44 vs. 5.48 ms (694 vs. 182 tok/s/user)

More throughput at previous TBT ≈56.1×

6,744 vs. 120 mixed / kW (at 182.46 tok/s/user)

News has emerged that NVIDIA will be the exclusive provider of AI compute infrastructure for the PORTS-Pike Technology Campus in Pike County, Ohio. This significant development marks a major milestone in the growth of artificial intelligence (AI) and its increasing importance as an essential component of modern industry.

The campus is being developed by SB Energy, which has partnered with NVIDIA to secure land, power, and shell capacity for hosting AI compute infrastructure. OpenAI will be the primary customer for this infrastructure, utilizing it to deploy AI factories that leverage NVIDIA’s full-stack DSX AI factory platform.

NVIDIA’s exclusive agreement with PORTS-Pike Technology Campus is a response to the extraordinary growth in demand for AI, which has become an essential component of modern industry. As Jensen Huang, founder and CEO of NVIDIA, noted, ‘AI is becoming infrastructure – the foundation for intelligence in every industry.’

The campus will provide 8-10 GW of new energy generation capacity, resulting in 8 IT-GW of AI factory capacity. This significant investment will not only support OpenAI’s operations but also create tens of thousands of jobs and stimulate regional economic development.

SB Energy has committed to investing at least $4.2 billion in new regional grid infrastructure through a partnership with AEP Ohio designed to protect ratepayers. Additionally, the company will build on its originally announced $40 million community benefits fund by adding an incremental $40 million to support local priorities such as affordable energy, job creation and workforce development, and community and economic development.

NVIDIA’s investment in SB Energy is a significant step towards supporting the growth of AI infrastructure at scale. The company has committed to investing $1.5 billion in SB Energy, joining existing investors SoftBank Group and OpenAI. This investment will enable SB Energy to continue its evolution into a leading developer of AI infrastructure while supporting local communities.

The PORTS-Pike Technology Campus is being developed on private and federal land through collaboration with AEP Ohio, the U.S. Department of Energy, and the U.S. Department of Commerce. The planned capacity is expected to come online in phases beginning in 2028, marking a significant milestone for regional economic development.

OpenAI’s CEO Sam Altman expressed his enthusiasm for building AI infrastructure at PORTS-Pike Technology Campus, stating that ‘this site will have enough computing power to help millions of people use AI to do things we can only start to imagine today.’ The campus is expected to play a crucial role in driving scientific discovery and healthcare advances while supporting regional economic development.

The investment by NVIDIA and SB Energy marks a significant step towards creating an ecosystem for the growth and deployment of AI infrastructure. As Masayoshi Son, Chairman and CEO of SoftBank Group Corp., noted, ‘the next era of intelligence will transform every industry – and require infrastructure built at unprecedented speed and scale.’

SB Energy’s PORTS-Pike Technology Campus is reindustrializing the decommissioned Portsmouth Gaseous Diffusion Plant and surrounding area, bringing a new generation of jobs to Appalachian Ohio. The campus development has been designed in collaboration with AEP Ohio, the U.S. Department of Energy, and the U.S. Department of Commerce.

The planned capacity at PORTS-Pike Technology Campus is expected to come online in phases beginning in 2028. This significant investment will not only support OpenAI’s operations but also create tens of thousands of jobs and stimulate regional economic development.

NVIDIA’s exclusive agreement with PORTS-Pike Technology Campus marks a major milestone for the growth of AI infrastructure at scale. As Jensen Huang noted, ‘now is the time to scale the AI infrastructure that will power the next industrial revolution.’

The campus development has been designed to provide long-lived infrastructure for NVIDIA compute, enabling OpenAI to deploy the most productive AI factories that can be upgraded repeatedly with each new generation delivering more intelligence and better economics.

Goldman Sachs and JP Morgan served as financial advisors for SB Energy, while Morgan Stanley advised NVIDIA on its investment in SB Energy. The partnership between NVIDIA, SoftBank Group, OpenAI, and SB Energy is a significant step towards creating an ecosystem for the growth and deployment of AI infrastructure at scale.

Cisco has released its Stack Automation platform, which automates the deployment of GPUs, AI models, applications, and other infrastructure across hybrid environments. This new tool is designed to simplify the complex process of setting up modern infrastructure for enterprises.

The development of Stack Automation was a collaborative effort between Cisco and Quali, with the exclusive right to offer it through Cisco’s channels. The primary goal is straightforward: make setting up and managing AI infrastructure less manual and time-consuming, especially when multiple technologies are involved and different teams need to be coordinated.

Stack Automation combines Quali’s automation technology with Cisco Validated Designs and security best practices. It also integrates seamlessly with other Cisco products like Intersight and Nexus Dashboard, which will eventually become part of the Cloud Control platform for AgenticOps.

The Solutions Hub in Stack Automation is a comprehensive resource featuring technologies from both Cisco and third-party vendors. This catalog spans networking, security, observability, virtualization, AI frameworks, enterprise applications, including popular tools like Splunk, Red Hat OpenShift, NVIDIA NIMs, VMware, Qumulo, and Veeam.

For channel partners, Stack Automation presents a significant opportunity to standardize and automate their deployment processes. The validated blueprints are designed to simplify infrastructure planning and deployment, allowing them to integrate existing investments in Terraform, Ansible, Python, Shell, and Helm into governed workflows without having to start from scratch.

Customers increasingly ask solution providers to set up complex environments that span multiple areas such as networking, compute, storage, security, virtualization, and applications. By automating the deployment of these full-stack environments, Stack Automation can save businesses weeks or even months spent on infrastructure setup before they can begin using AI.

The platform’s ability to automate complex deployments while incorporating the collective expertise and validated designs of Cisco and its partner ecosystem into every workflow is a key benefit, according to Jeremy Foster, senior vice president and general manager of Cisco Compute. Customers should not have to spend weeks assembling infrastructure before putting AI to work.

Stack Automation offers different tiers for users: Essentials supports standard software deployments on a subscription basis, while the Advantage subscription adds features like Blueprint Designer, GitOps integration, advanced customization, and agentic design capabilities. Full-stack solutions, including Cisco AI PODs, are scheduled for availability in October.

The automation of complex deployments through Stack Automation has the potential to save businesses weeks or even months spent on infrastructure setup before they can use their AI tools effectively.

A developer has pushed the boundaries of what’s thought possible for large language models by running one on an extremely low-cost microcontroller. The LLM in question, with its massive 28.9 million parameters, runs entirely locally on an ESP32-S3 chip that costs just $8.

The project uses a trick borrowed from Google’s Gemma models called Per-Layer Embeddings to fit the model into the microcontroller’s memory. This technique stores approximately 25 million parameters in flash memory as a lookup table, which dramatically reduces memory requirements during inference and lets the full model occupy just 14.9MB.

According to the developer, this is the first known demonstration of Google’s Per-Layer Embeddings concept being adapted to hardware as constrained as a microcontroller. The ESP32-S3 chip has only 512KB of SRAM, 8MB of PSRAM, and 16MB of flash storage, but it manages to run the model smoothly.

Despite its limitations, the model can generate short, simple stories with reasonably coherent structure. It was trained on Microsoft’s TinyStories dataset and is not intended for tasks like answering questions or writing code. The project’s GitHub repository includes all necessary files – firmware, training scripts, quantization pipeline, wiring instructions, and experimental results.

The generated text is written directly to a small display connected to the board at roughly 9.5 tokens per second. This implementation significantly outperforms earlier projects that ran LLMs on microcontrollers with much smaller models containing around 260,000 parameters – it’s about 100 times larger and raises important questions about how such large models can be made to fit on hardware with limited memory.

The project’s success has far-reaching implications for the development of AI-generated content. While this particular implementation focuses solely on text generation – rather than creating images directly – its potential applications extend into other areas, including image generation and beyond.

A new project called NightRun allows users to run local large language models (LLMs) directly on a Raspberry Pi’s bare metal, eliminating the need for an operating system like Linux. This development is significant because it enables users to run AI-generated images and other tasks without relying on cloud services or powerful hardware.

The creators of NightRun acknowledge that LLMs deserve serious ethical consideration, but running them locally on one’s own hardware can alleviate many concerns. To achieve this, centoslinux built NightRun using Rust programming language with the assistance of Claude AI. The project runs as a UEFI application at boot, pulling the LLM from removable storage and copying it into RAM to execute.

NightRun supports various lightweight LLMs on Raspberry Pi 5 models, including Llama 3.2-1B Instruct, which can run on the 4GB model. The 8GB and 16GB versions of the device support additional LLMs such as Granite 4.1-3B and Qwen3-4B Instruct 2507. These models are capable of running more complex tasks due to their increased memory capacity.

Although NightRun is not truly running on bare metal, it operates within UEFI Boot Services. This allows the application to accept keyboard input, read from storage, and output a terminal to a display without incurring significant overhead. The creators emphasize that even lightweight operating systems supporting peripherals are more demanding than UEFI Boot Services.

Users can try NightRun by downloading the files and installer script from GitHub. The installation process involves creating a removable storage image using the provided instructions. This project demonstrates the potential for running AI-generated images and other tasks locally on single-board computers without relying on cloud services or powerful hardware.

A major expansion of their strategic partnership has been announced by SK Group and NVIDIA, with a comprehensive initiative worth over $500 billion. This collaboration will span the development of AI factories and next-generation memory solutions. The two companies have signed letters of intent to formalize this agreement, which covers various aspects of AI infrastructure construction and supply.

The announcement builds on decades-long technology partnership between SK Group and NVIDIA, including recent plans for SK Telecom to build a 2-gigawatt-scale AI Cloud in Korea. This cloud will utilize the NVIDIA DSX platform and deploy NVIDIA Vera Rubin accelerated computing powered by SK hynix HBM4. The first AI factory is expected to come online in 2027.

The new partnership aims to accelerate large-scale AI infrastructure development, including sovereign, physical, agentic, and enterprise AI services. Joint efforts will focus on addressing the increasing demand for AI across the Asia-Pacific region, particularly in South Korea. This collaboration enables SK Telecom to invest in and expand its large-scale AI infrastructure while making capital-intensive AI infrastructure available to a broader range of customers.

The partnership also involves SK hynix entering into a long-term AI memory agreement with NVIDIA. As part of this deal, NVIDIA will secure a stable supply of next-generation AI memory, allowing the company to optimize solutions for evolving infrastructure demands. This includes large language model training and agentic/physical AI applications.

SK Group Chairman Chey Tae-won emphasized that competitiveness in the AI era depends not just on effective utilization but also on intelligence production capacity. He stated that by leveraging SK hynix’s AI memory capabilities, SK will collaborate with NVIDIA to build a world-class AI factory, helping Korea become a global hub for AI innovation.

NVIDIA CEO Jensen Huang noted South Korea has all the necessary ingredients to become an AI powerhouse: world-class networks and data centers, leadership in chip technology, and vast industrial scale. He stated that together with SK Telecom and SK hynix, they are building a new generation of AI factories powering Korea’s next wave of growth.

The NVIDIA Vera Rubin DSX AI Factory will be built on the full-stack architecture, integrating accelerated computing, systems, software, and partner technologies to deliver maximum energy efficiency at minimum cost. This infrastructure development is crucial for addressing the increasing demand for global compute services.

SK Telecom plans to build a 2-gigawatt NVIDIA Vera Rubin DSX AI factory that will serve as a major hub for global compute demand. The partnership between SK Group, NVIDIA, and SK hynix marks significant progress in establishing comprehensive AI infrastructure across Asia-Pacific regions.

A team of developers has successfully fitted a large language model (LLM) with 28.9 million parameters onto an ESP32-S3 microcontroller, demonstrating the potential for AI to run on even the smallest devices. This achievement is particularly noteworthy given the tiny size and limited RAM of the ESP32-S3.

The LLM in question can tell coherent stories and keep track of variables, showcasing its ability to perform complex tasks despite being confined to a relatively small space. For context, this model has roughly a quarter of the parameters found in OpenAI’s first ChatGPT version, which had 117 million parameters.

So how did they manage to pack so many parameters onto such a tiny device? The key lies in an innovative technique called Gemma’s Per-Layer Embeddings. This approach recognizes that most language model parameters are stored in a large embedding table, rather than being computed on the fly.

The solution was to store this 25 million-row table in flash memory and read only a few rows at a time when needed. By doing so, they were able to free up fast RAM for more critical tasks, requiring only around 560 kilobytes of dense core space.

This clever use of memory-mapped XIP (execute-in-place) allows the model to run efficiently on-chip, processing approximately nine tokens per second. While this may not be as fast as larger AI models, it’s a remarkable feat considering the ESP32-S3’s limited resources.

The project is now available for download from the ESP32-AI GitHub page, allowing developers and enthusiasts to experiment with their own LLMs on small devices.

The tech company Moreh recently showed off its distributed inference solution, the MoAI Inference Framework, running on AMD graphics processing units (GPUs) at a conference in San Francisco. This demonstration highlighted how well the framework works with large language models on systems equipped with 32 AMD Instinct MI300X GPUs across four nodes.

At this AI event, Moreh presented a live test of its GLM-5.1 LLM, which is powered by the MoAI Inference Framework. Visitors got to see and use the chatbot themselves, and they gave feedback on how fast it responded and how well it performed in real-world situations. This hands-on experience let attendees verify that the framework works as promised.

The demonstration showed key performance metrics in real time, like GPU usage, tokens per second, time to first token, and time per output token. This level of transparency helped visitors understand what the system can do and how well it performs. People at the event praised the live deployment of a big model on AMD GPUs with production-grade service levels.

Global leaders in AI and companies that buy these services were very interested in the system’s fast response times and stable performance. Moreh’s MoAI Inference Framework is seen as one of the first distributed inference solutions built specifically for the AMD ecosystem, which helps solve a major problem: the cost of running big AI models.

The CEO of Moreh, Gangwon Jo, said that this event gave customers around the world a chance to see firsthand how well large language model inference can be done on AMD GPU environments. He added that Moreh will keep working on its AI infrastructure software so enterprises can run their services as efficiently as possible, no matter what kind of GPUs they’re using.

Moreh’s distributed inference and computing technologies are designed to make running big AI models much cheaper, which should help more people use these systems around the world. This technology addresses a major challenge in the industry by making it easier to set up an efficient system for inferring things from data. Moreh also develops its own engine for AI infrastructure.

Moreh is strengthening its presence in the global market through partnerships with big tech companies, including AMD and Tenstorrent. The company’s innovative approach to running distributed inference on AMD GPUs marks a significant step forward for the industry.

Despite its ambitious goal to remove more carbon than it emits by 2030, Microsoft has seen a significant increase in greenhouse gas emissions. The company’s operational emissions rose by 25% in 2025, largely due to the construction of massive data centers to support its AI push. This surge in energy demand is putting pressure on the tech giant’s sustainability efforts.

The rapid expansion of data centers is driving up electricity consumption. As Microsoft continues to build more facilities, it needs increasing amounts of power to keep them running. The company has been using fossil fuels for some of this energy, which contributes significantly to its emissions.

One factor that has contributed to the increase in reported emissions is Microsoft’s decision to suspend purchases of renewable energy credits (RECs). RECs are certificates that allow companies to claim they have used clean power. By not buying these credits, Microsoft’s reported emissions rose accordingly.

Microsoft President Brad Smith and Chief Sustainability Officer Melanie Nakagawa acknowledged the tension between AI infrastructure growth and sustainability goals in a recent statement. They described it as ‘real’ and ‘productive,’ highlighting the need for more efficient solutions to meet growing energy demands.

The company is reevaluating some of its earlier commitments, including matching data center electricity use with renewable energy every hour. Smith and Nakagawa emphasized that this does not mean Microsoft is lowering its ambition but rather being more precise about what sustainability requires and willing to refine strategies as conditions change.

Microsoft’s situation reflects a broader trend in the tech industry. Many companies have scaled back their environmental promises following changes in U.S. government regulations and policies limiting sustainability initiatives. For instance, Chevron Corp. has signed an agreement with Microsoft to supply power from a natural gas-powered plant for its new data center facility in West Texas.

The tension between AI expansion and climate goals is intensifying as the industry’s energy demands continue to outpace efforts to reduce emissions. While Microsoft’s investments in carbon removal and renewable energy have increased, they have not kept pace with the massive energy needs of new data centers. The company had previously paused some REC purchases to focus on direct investments in clean energy infrastructure.

Microsoft has invested billions into carbon removal contracts and renewable energy projects but still faces significant challenges. The shift away from buying RECs reflects a broader push for more direct impact, though it has temporarily worsened the reported footprint.

With four years remaining until its 2030 carbon negative target, Microsoft will need to adapt quickly as conditions evolve.

The memory shortage is having far-reaching consequences for the tech industry, and even Google isn’t immune. The company’s own research arm, Google DeepMind, is feeling the pinch due to a lack of available memory chips.

The issue lies in the supply chain, which is heavily constrained. According to Demis Hassabis, CEO of Google DeepMind, ‘the whole supply chain’ for memory chips is under strain. This means that companies are struggling to get their hands on the necessary components to power their AI systems.

Hassabis pointed out that even with its own Tensor Processing Units (TPUs), which it produces in-house and leases to external customers through its cloud, Google still has to navigate the competitive memory market. ‘It still comes down to a few suppliers of a few key components,’ he said.

The three main suppliers of memory chips are Samsung, Micron, and SK Hynix. These companies are struggling to meet demand from AI hyperscalers like Google without sacrificing their longtime electronics customers. The problem is further complicated by the fact that AI companies require a different type of memory chip than PC manufacturers do.

Large language model producers need high-bandwidth memory (HBM) chips, which are in short supply. This shortage has led to increased costs and prices for products across the industry. Companies like Google are having to get creative with their spending on AI infrastructure and chips, but it’s not clear whether this will be enough to meet demand.

Google itself is projecting capital expenditures of $175 billion to $185 billion for 2026. This significant investment in AI infrastructure and chips suggests that the company is taking steps to mitigate the effects of the memory shortage. However, even with its own resources, Google may not be able to escape the constraints imposed by the supply chain.

The issue has far-reaching implications for businesses looking to leverage AI tools for their operations. As Hassabis noted, ‘you need a lot of chips to be able to experiment on new ideas’ and drive innovation in the field. The memory shortage is creating an ‘AI choke point,’ where companies are struggling to access the necessary resources to power their research and development efforts.

The situation highlights the complexities of the tech industry’s supply chain and the challenges that come with scaling up AI systems. As demand for memory chips continues to outstrip supply, it remains to be seen whether the industry will find a way to address this issue or if companies like Google will have to adapt their strategies in response.

New Taipei City, Taiwan-based Vecow Co., Ltd. has announced its participation in the recent AUTOMATE 2026 event held in Chicago. The company showcased its latest AI robotics platforms and deployment-ready solutions designed to accelerate industrial automation workflows.

Dell Technologies is leveraging strategic partnerships with major players in the tech industry to drive its initiatives. The company's collaborations with chipmakers, virtualization leaders, cloud providers, and leading software platforms have enabled it to develop a platform that integrates compute, storage, networking, and workloads across hybrid and multicloud environments.

A major bottleneck in real-time AI development has been the slow pace of text generation. This limitation affects responsiveness, increases serving costs, and makes interactive experiences difficult to achieve.