Automakers Investing in Memory Chips: Look Beyond Market Cap

I spent the weekend digging into the news about GAC investing in ChangXin Memory Technologies (CXMT) and hit quite a few pitfalls. I originally just wanted to align the numbers in the news, but later realized it wasn't possible. Financial figures and engineering figures aren't on the same basis; looking at either alone leads to misjudgment. So I used Feishu Multi-Dimensional Tables to create a two-page evaluation sheet. The left side holds public disclosures, and the right side holds verification fields I commonly use when making compute infrastructure lists. The table is still rough, but at least it forces me to ask, "Has ROI been calculated?"

First, I looked at GAC's own draft. Words like "forward-looking investment participation" and "synergy channel" stand out on the page. There's a sentence in the middle:

"Opening up the synergy channel between whole-vehicle application scenarios and upstream chip R&D."

This sentence can't be scored in a financial table; it only makes sense in an engineering table. Then I went to financial articles to find IPO info. On July 27, CXMT listed on the STAR Market. The issue price was 8.66 RMB/share, opening at 49.50 RMB, with a first-day high open of 471.59%, closing market cap at 3.28 trillion RMB. Looking just at this, the story seems complete: early industrial capital enters, hard tech lists, book returns amplify. Many management teams stop here, thinking the automaker made a great bet.

Where I got stuck is also here. Public materials say GAC invested indirectly through an industrial fund managed by GAC Capital, but there's no specific ratio after penetration. Without the ratio, we can't calculate how much GAC actually earned, nor judge the weight of this money in the group's profit statement. In my tests, I can only mark the financial return column as uncertain; I can't draw direct conclusions. Another sticking point is automotive-grade validation. DRAM stands for Dynamic Random Access Memory. Simply put, it's temporary memory for data that loses everything when power is cut. Smart cockpits and autonomous driving involve massive amounts of data passing through it. But getting it into a vehicle is much more troublesome than plugging a RAM stick into a server. Behind it are power consumption, temperature, vibration, lifespan, safety redundancy, plus interface matching for cockpit domain controllers and autonomous driving domain controllers. The news says CXMT products include DDR4, DDR5, LPDDR4X, LPDDR5/5X, widely used in AI servers, cloud computing, and smart terminals. I found this surprising; it shows they're building a general-purpose storage foundation, with automotive chips being just one scenario. Once the general foundation is stable, automakers need to calculate if platformization can reduce pitfalls, with single-model costs amortized later.

Comparing options, I gave two schemes. Scheme 1 focuses only on IPO market cap and stock price gains. IPO is company fundraising upon listing. Pros: strong publicity, easy to report to the board. Cons: easily frames industrial layout as investment windfall. Scheme 2 focuses on matching DRAM with the whole-vehicle electronic architecture. Pros: puts compute, storage, cockpit, and autonomous driving on one engineering map. Cons: requires internal teams to supplement chip selection, testing, and supply chain validation. I've led platform teams and know this validation can't be solved just by procurement contracts. If embedded testing and supply chain engineers don't catch it, no matter how lively the investment department is, it's useless. GAC also disclosed sales exceeding 880,000 units from January to July, with independent brand sales breaking 400,000 units. Only with sufficient scale can validation costs be amortized. This direction is worth investing in; the key is who can compress the validation cycle. For automakers investing in memory chips, what's truly valuable is whether they can compress the validation cycle. Listed market cap is just a book result.

So the conclusion depends on the situation. For those in automotive electronics, platform engineering, supply chain, and investment strategy, this news is worth putting in internal reviews because it hints that upstream storage will be incorporated into the compute foundation, and the positioning of edge accessories needs re-evaluation. For those who just want to copy homework for stocks, I don't recommend it. Information is incomplete, ratios are opaque, engineering cycles are long, and judging based on first-day gains is too thin. Looking ahead, if automakers truly incorporate domestic storage into cockpit and autonomous driving platforms, whether this validation cycle can be flattened by sales scale—I haven't figured it out yet.