The AI boom is producing IPOs of record-breaking scale.
By Chen Mei
Editor: Wang Qingwu
本文字数3484
Human-written丨96% AI-generated丨4%
Solidigm, the U.S. subsidiary of South Korean semiconductor maker SK hynix, is considering an IPO as early as next year, with a target valuation of up to $150 billion.
Its $150 billion valuation would be nearly $100 billion higher than Arm’s roughly $54 billion valuation when it went public in 2023 and Cerebras Systems’ roughly $56 billion valuation at its 2026 IPO—potentially setting a new record for the largest IPO in semiconductor industry history.
A closer look reveals that, unlike most startups that spring up from scratch, Solidigm has a team forged from two camps. After its acquisition by SK hynix, former Intel employees and SK’s own ranks began to come together, creating one of the most formidable forces in the memory chip industry.
Five years: from Intel castoff to the biggest semiconductor IPO
Solidigm’s story begins with an acquisition that shook the storage industry.
In 2020, Solidigm was still part of Intel when its parent company abruptly “abandoned” it, announcing that it would exit the NAND flash memory and SSD businesses. SK Hynix subsequently acquired it for about $9 billion. After integrating the relevant businesses, Solidigm was formally established as an independent subsidiary in 2021. Lee Seok-hee, SK Hynix president and co-CEO, became Solidigm’s executive chairman and continued to lead the post-acquisition integration.
However, Solidigm struggled after Lee Seok-hee took over. From 2022 to 2024, the global memory chip market endured a deep cyclical downturn, with prices falling for several consecutive quarters. Solidigm also remained loss-making amid lengthy production-line adjustments and customer qualification cycles.
Data show that Solidigm recorded a net loss of 3.3 trillion won in 2022. That loss widened to 4 trillion won in 2023, bringing the combined loss for the two years to more than 7 trillion won, or about 37 billion yuan. By the end of 2023, Solidigm’s shareholders’ equity had fallen into negative territory, reaching -906 billion won, or about -4.76 billion yuan.
The heavy debt forced Solidigm to halt production across the board. In October 2024, Solidigm discontinued consumer SSD products including the P44 Pro and P41 Plus, effectively dismantling its consumer division.
The turning point came with the full arrival of the AI era. After enduring the pains of its transformation, Solidigm’s performance rebounded sharply in the second half of 2024, driven by a surge in AI data center demand and a spike in demand for enterprise SSDs, particularly high-capacity QLC products.
In the first half of 2026, the company generated 12.25 trillion won in revenue and 5.839 trillion won in net profit, sending its net margin soaring to 47.7% and turning its cumulative earnings positive for the first time. Meanwhile, Solidigm’s management underwent another reshuffle as the company moved to accelerate the growth of its AI storage business.
In May 2026, Solidigm announced the appointment of Xin Guo and Richard Chin as its new co-CEOs. Guo has a technical background and a PhD in electrical engineering from Yale University. He previously held senior leadership roles at companies including Intel, Numonyx and Spansion, where he led the development of NAND flash and SSD systems.
Richard Chin clearly has ties to the SK Group. He established SK Telecom’s first venture capital division, led the business integration after Hynix Semiconductor joined the SK Group, and served as SK Hynix’s chief sales and marketing officer.
The earnings turnaround, combined with a blend of former Intel executives and SK Hynix loyalists, has accelerated Solidigm’s push toward the heart of the AI storage market—and toward an IPO.
The biggest semiconductor IPO: What makes Solidigm worthy?
Looking back over the five years since Intel “abandoned” it, SK Hynix not only supported Solidigm through its earnings slump but also turned the acquisition into a highly rewarding investment.
At a target valuation of $150 billion, SK Hynix’s $9 billion investment would represent a return of nearly 17 times. Even factoring in subsequent investments in production lines and integration costs, this would rank among the semiconductor industry’s most successful acquisitions of the past decade. Most importantly, Solidigm’s enterprise SSDs complement SK Hynix’s HBM and DRAM, together forming a complete storage solution for AI servers.
But how can a storage chipmaker built through acquisitions and integration support a $150 billion valuation and become the biggest IPO in the semiconductor sector?
The answer may lie in two figures for Solidigm: a 47.7% net margin and 40% quarter-over-quarter shipment growth.
Counterpoint Research data show that enterprise SSDs accounted for 48% of global NAND shipments in Q2 2026, with AI inference workloads replacing training as the main driver of demand. Against this backdrop, SK Hynix ranked second globally in Q2 2026 with a 22% share of shipments, while Solidigm’s bit shipments rose 40% quarter over quarter.
For comparison, Nvidia’s net margin was about 55% in fiscal 2026, while most semiconductor companies’ net margins hover between 20% and 30%. By contrast, Solidigm, an enterprise SSD maker, has a net margin that exceeds those of many semiconductor companies and is closing in on industry giant Nvidia.
More importantly, Solidigm’s explosive growth is not just about “volume”—prices have surged as well.
In Q2 2026, contract prices for enterprise NAND rose approximately 75%–80% quarter over quarter. Take Solidigm’s own 122TB drive: the price of its D5-P5336 rose from $12,399 to $37,128. That is nearly a threefold increase in just nine months, equivalent to a rise in price per terabyte from about $101 to $302.
And this price surge is not the result of a short-term mismatch between supply and demand. It reflects a fundamental shift in the role of storage in the AI inference era.
Unlike training, AI inference requires frequent access to vast knowledge bases and contextual data. When processing a single user request containing 42,000 tokens, about 13.1 GB of KV cache space may be needed. This means storage is no longer just for “long-term data retention.” It is starting to temporarily hold the indexes and cached data being used by the “model,” taking on a role like memory.
Solidigm’s customer list bears this out: cloud providers and enterprise customers including CoreWeave, Vast Data, Dell and Tencent have brought it into their supply chains. In August this year, Solidigm also signed a multiyear agreement with AI cloud company CoreWeave, giving the latter priority access to enterprise SSD capacity.
In addition, the SK Hynix Group’s overall performance has bolstered Solidigm’s valuation. In Q2 2026, SK Hynix posted operating profit of 60.54 trillion won, up 557% year over year, with an operating margin of more than 76%—a record high for a single quarter. Meanwhile, in the first half of the year, SK Hynix Group revenue topped 100 trillion won for the first time, equivalent to about 494.4 billion yuan.
In its latest quarterly earnings report, SK Hynix explicitly said that increased investment in AI infrastructure was driving growth in server storage demand. Higher sales of high-value-added products—including HBM, AI server DRAM and enterprise SSDs—were contributing to improved performance.
The AI boom is producing the “largest” IPOs
Since the start of this year, valuation ceilings for technology companies around the world have continued to be shattered.
In June this year, Elon Musk’s SpaceX completed an IPO at a valuation of $1.77 trillion, setting a record as the largest IPO in history. Anthropic soon surpassed that record with a $2 trillion valuation, putting it in contention to become one of the largest IPOs ever.
Then there is Solidigm. The company, headquartered in Rancho Cordova, California, has held underwriting pitch meetings with several investment banks, marking a substantive step forward in its IPO process.
The three companies have clear common ground behind their bids to challenge for or become the largest IPOs in history: they occupy central positions in AI infrastructure.
First, consider SpaceX. Although it is a private aerospace manufacturer, its core business extends far beyond rockets and spaceflight. Space, connectivity through Starlink, and AI form SpaceX’s core competitive moat.
Building on its core space business, Musk has brought rockets, Starlink and AI together. In particular, after SpaceX acquired xAI, Musk took AI computing power directly into orbit. Generating AI computing power in space is expected to become the lowest-cost option within the next two to three years.
As the latest contender for the title of the largest IPO in history, Anthropic has also embraced the “AI infrastructure” narrative. This has led capital markets to view Anthropic as the “water, electricity and fuel” of future AI infrastructure, and to expect it to command a huge share of the global AI market. Ultimately, investors have assigned Anthropic an exceptionally high long-term valuation based on an “endgame” outlook. If Anthropic’s revenue reaches $200 billion in 2028, its price-to-sales (P/S) ratio would be just 10 times.
Solidigm is no different. Its “AI infrastructure” story casts it as “indispensable”: training requires HBM, while inference also needs vast amounts of fast, low-cost SSD storage to cache data. Without SSDs, AI servers cannot run.
But beneath the spectacle, there are concerns. After its listing, SpaceX shares fell as low as $104.83, nearly halving from their $185 offering price. Although they have since recovered to around $145, their future direction remains uncertain. After Anthropic disclosed its prospectus, the market was astonished by its $518 billion in cloud, computing and infrastructure commitments. At the same time, its net loss of about $42 billion stood out as one of the most striking figures.
For Solidigm, then, business is thriving during this upcycle. But if the industry enters a downturn or growth slows, the dramatic booms and busts of memory chips will be a fate it cannot avoid.
An original article by ChinaVenture