Wolfspeed: From Bankrupt to Innovator?
Artificial intelligence (AI) is transforming yet another semiconductor sector investors rarely discuss: advanced semiconductor packaging.
As AI models grow larger and demand greater memory bandwidth, the packaging technologies that connect processors and memory are becoming a critical component.
Modern AI processors depend on stacks of high-bandwidth memory (HBM) placed directly beside the GPU (graphics processing unit) and connected through an “interposer,” or advanced substrate.
Wolfspeed (NYSE: WOLF), emerging from bankruptcy September 2025, and with a new management team, is bidding to take a leading role in this evolution. That explains the recent announcement in which the company announced its 300-mm diameter silicon-carbide (SiC) wafers could support large interposers and thermal structures for next-generation AI. Today, most SiC wafers still are 150 to 200 mm in diameter. Bigger wafers save money and/or accommodate larger structures.
HBM’s rise drives a new packaging market
Driven by the rapid adoption of AI accelerators, the high-bandwidth memory market is expanding at a pace rarely seen even in the semiconductor industry. HBM revenues are projected to rise from approximately $12 billion in 2022 to about $65 billion by 2027, a five-fold increase in just five years. Measured by “bits shipped,” the growth is even more striking. HBM output, measured in one-gigabit equivalents, is projected to rise from roughly 1.3 billion units in 2022 to more than 40 billion by 2027, a more than 30-fold expansion.
Each HBM stack must be placed next to the GPU inside a complex package, meaning that the expansion of the HBM market simultaneously drives demand for advanced packaging to integrate processors and memory in increasingly dense configurations. In other words, the growth of HBM is simultaneously creating an entirely new market for the packaging technologies that connect GPUs and memory.
Si today, SiC tomorrow?
Today’s AI accelerators rely on silicon (Si), not silicon carbide (SiC), interposers to connect GPUs with multiple stacks of HBM. These interposers provide extremely fine routing density that allows processors and memory to exchange data at extraordinary speed. As the number of memory stacks increases and GPU power consumption rises, however, silicon has challenges.
More power consumed means more heat generated. Silicon’s thermal conductivity—its ability to shed heat—is roughly 150 W/m·K. Silicon carbide, by contrast, has thermal conductivity between 370 and 490 W/m·K, more than twice that of silicon, while also offering significantly greater mechanical stiffness. These properties help explain why silicon carbide is widely used in high-power electronics, and why it is being contemplated as an interposer material for AI chips.
The illustration below shows how Wolfspeed envisions silicon carbide being used inside future AI packages. GPUs and HBM stacks would be mounted on a silicon-carbide interposer positioned above the package base and mechanical support frame. In this architecture, the interposer serves not only as a high-density interconnect layer but also as a structural and thermal platform capable of spreading heat across the package.
In effect, Wolfspeed is proposing that silicon carbide—long used in power electronics for its thermal and structural properties—could become part of the material foundation for the next generation of AI packaging.
Importantly, Wolfspeed is not proposing that silicon carbide immediately replace silicon. Instead, the idea is to start by using hybrid architectures that combine silicon logic with SiC structural layers to help manage the thermal and mechanical stresses of increasingly large AI packages.
A new path
Semiconductor progress historically has been defined by shrinking transistors to squeeze more of them on each chip. Now, AI systems are forcing a new phase of innovation centered on system integration and packaging. GPUs, HBM stacks, chiplets, and high-speed interconnects must coexist inside extremely dense modules that dissipate enormous amounts of heat. Silicon carbide could be a big part of the solution.
Wolfspeed’s announcement was not about a near-term product introduction, but a strategic bet on the future architecture of AI systems. If the growth trajectory of HBM continues, the packaging technologies surrounding those memory stacks could become one of the most important—and least understood—segments of the semiconductor supply chain.
As memory bandwidth and power density continue to rise, the materials used in advanced packaging may become as important as the processors themselves.
So, should we bring WOLF back from exile and restore it to our portfolio? We will address that question in an upcoming issue of the Gilder Technology Report.
P.S. Join George Gilder and Eagle Publishing at FreedomFest: “Think Independent” July 8-11 at Caesars Forum Las Vegas! Claim $100 off the current Regular Attendee or VIP Full Pass rate when you register with discount code EAGLE100. Or call Hayley to register over the phone at (855) 850-3733 ext. 202.