Here’s a look at the biggest semiconductor and AI hardware stories readers couldn’t stop clicking this week (April 27 to May 4, 2026). From bold new fab ambitions to major shifts in data-center strategy, the past few days made one thing clear: the chip industry is moving into a new era where supply chains, manufacturing economics, and AI-driven demand are reshaping nearly every decision.
TeraFab is already changing the wafer-fab landscape, even before the first chip ships
A newly announced chip fabrication venture called TeraFab, launched in partnership with Intel, is already rippling through the semiconductor ecosystem. The logic behind it is simple: when the market can’t reliably supply what you need, you stop waiting and start building.
That mindset has famously driven multiple “build-it-yourself” plays in other industries, and now it’s being applied to one of the most complex and capital-intensive arenas on earth: advanced semiconductor manufacturing. While it’s a disruptive move on paper, many industry watchers aren’t shocked. The pressures behind the decision—capacity access, strategic control, and long-term cost positioning—have been building for years.
Intel cancels three projects in two months as a multi-year reset begins
Intel is moving quickly through a major internal reshaping, shelving three projects in just two months as part of a broader reset led by Data Center Group executive vice president and general manager Kevork Kechichian, with backing from CEO Lip‑Bu Tan.
The goal: reposition Intel’s data-center playbook at a time when Nvidia’s AI dominance is rewriting infrastructure priorities and Arm continues to rise as a serious alternative in server CPUs. The strategy signals a push toward CPU-centric orchestration that works alongside GPUs and custom accelerators (ASICs), rather than fighting a head-to-head battle on a single front. Over the next two to three years, the outcome could influence everything from AI deployment costs to global data-center economics—and it may reshape how enterprises decide what hardware mix they standardize on.
Tata’s India chip fab pushes forward, but leadership churn and engineering challenges persist
Tata Electronics’ semiconductor fabrication project in Dholera, Gujarat—often described as India’s most ambitious attempt to build a homegrown chip manufacturing base—continues to show progress, but not without turbulence.
Construction activity and regulatory milestones point to momentum, yet repeated leadership exits and demanding site conditions highlight how difficult it is to execute a greenfield fab at global scale. The story unfolding in Dholera is a reminder that building a semiconductor ecosystem isn’t just about funding and ambition; it’s also about sustained technical leadership, steady execution, and solving complex engineering realities that don’t bend to timelines.
Why TSMC is holding off on ASML’s High-NA EUV, despite the hype
ASML’s 0.55 High Numerical Aperture Extreme Ultraviolet (High-NA EUV) lithography is designed to extend Moore’s Law and push advanced nodes further. Many expected TSMC to be the first major customer to adopt it. Instead, TSMC is waiting.
Company executives have indicated there are currently no plans to introduce High-NA EUV before 2029, with a straightforward reason: the cost is simply too high right now. But the decision also reflects something deeper—TSMC appears increasingly focused on competing through process integration mastery and cost efficiency, rather than racing to deploy the newest tools first. In other words, the next phase of foundry competition may be less about who buys the most advanced equipment, and more about who can deliver the best performance-per-dollar at scale.
Samsung reportedly reaches a key DRAM milestone with a single-digit nanometer-class working die
Samsung Electronics is said to have achieved a major memory manufacturing milestone by producing a single-digit nanometer-class 10a DRAM working die. The working die is reportedly being used to tune process conditions and accelerate yield improvement—one of the hardest parts of turning an advanced process into a reliable, high-volume product.
According to industry-sourced reporting, Samsung ran 10a process wafers in March 2026 and confirmed working dies during characterization testing. The development is notable not only for the scaling achievement, but also because it’s tied to key architecture and transistor changes, including a 4F Square approach combined with Vertical Channel Transistor (VCT) technology—an important signal of where leading-edge DRAM design and manufacturing are headed.
Cerebras targets an IPO after a turnaround year, with a partnership shaping its supply chain
Cerebras Systems filed for an initial public offering on April 17, 2026, with plans to list on the Nasdaq. The filing follows a sharp financial reversal: the company reported US$510 million in revenue and net profit of US$87.9 million in 2025, compared with US$290 million in revenue and a net loss of US$484.8 million in 2024.
The timing is especially interesting given the growing attention around demand for specialized AI compute. Just two days before the IPO filing, reporting indicated that OpenAI’s procurement agreement for Cerebras AI chips could potentially expand up to US$20 billion, and that the deal may include warrants allowing OpenAI to subscribe up to 10% equity in Cerebras. Meanwhile, Cerebras’ partnership with G42 is also drawing focus for how it influences supply-chain strategy and capacity planning—key factors for any AI hardware company trying to scale quickly.
Texas Instruments spotlights 800V power architecture for AI data centers ahead of Computex 2026
Power delivery is becoming one of the defining bottlenecks of AI infrastructure, and Texas Instruments is leaning into that reality with plans to showcase 800V power architecture-based AI data-center solutions at Computex 2026.
Beyond data centers, TI is positioning the same high-voltage approach for fast-growing applications such as humanoid robots, automotive systems, and edge AI deployments. Ahead of the event, TI executives have also been engaging with Taiwanese supply-chain partners to explore collaboration opportunities—another sign that the “AI boom” isn’t only about processors and memory, but also about the power systems that make next-generation compute feasible at scale.
