Two Asian signals hit the US semiconductor tape inside 48 hours. Reuters flagged on July 7 that DeepSeek is quietly building its own AI inference chip, aiming to lean less on both Nvidia and Huawei. Samsung had posted a Q2 operating profit up 1,810% year-over-year, driven by AI memory demand, only two days earlier. The market first hesitated. Then chip stocks rebounded hard on July 9, with Micron up 4.52%, AMD up 5.67% and the semi sector adding 3.1% for the session. What follows prices both the bull case for continued AI compute demand and the bear case that Samsung DeepSeek marks the first crack in Nvidia’s moat.
The Read
- Samsung DeepSeek double signal lands on the US chip tape, first via a Reuters DeepSeek chip leak on July 7, then via Samsung’s 1,810% Q2 profit jump.
- Chip stocks bounced back sharply on July 9: Micron +4.52%, AMD +5.67%, SanDisk +7.59%, Lam Research +6.01%.
- The bull case leans on $1.5T of AI infrastructure spending by 2027; the bear case flags Nvidia’s moat is now known to be under attack.
Reuters Leak on DeepSeek Chip Meets Samsung’s Q2 Blowout
Reuters reported on July 7 that Chinese AI lab DeepSeek is quietly developing its own AI inference chip. The project targets the serving side of AI, where a trained model actually answers user prompts. Three people familiar with the effort told Reuters the goal is to lean less on both Nvidia and Huawei silicon, while staying inside US export rules.
The project is still at the discussion and hiring stage. No foundry partner is named. No prototype has been shown. DeepSeek has reportedly been in talks with chip design, foundry and memory partners for about a year, and has hired chip design engineers privately.
Samsung posted its own signal in parallel. Its Q2 operating profit jumped 1,810% year-over-year, driven by AI memory demand. The stock still dropped 11% on the print. That gap flagged worries the beat had already priced in every bullish catalyst visible on the tape.
Taken together, the two signals landed as a stress test for the entire US chip complex. Nvidia and its ecosystem faced simultaneously a demand-side signal (Samsung memory) and a supply-side threat (DeepSeek building for itself). The reaction was messier than a single-catalyst move usually is.
Memory Names Rebound as Accelerator Trades Hesitate
The July 9 tape put clear water between memory names and accelerator names. Micron closed +4.52%, AMD +5.67%, SanDisk +7.59%, Lam Research +6.01%, and the semi sector added 3.1% for the session. Nvidia participated in the rally but did not lead it.
Memory chips run on different physics than AI accelerators. HBM (high-bandwidth memory) capacity is short across the industry. Even if DeepSeek ships its own inference chip, that chip still needs memory around it. The DeepSeek threat lands harder on Nvidia than on Micron.
AMD’s move is more revealing. AMD sits both sides of the fence, memory-adjacency plus MI-series accelerators. A 5.67% one-day rally at that price point suggests the market read DeepSeek as a competitive threat to Nvidia specifically, not to the compute industry itself, versus the bull view that DeepSeek merely reallocates the pie inside Asia.
The chip selloff that hit Applied Materials earlier in the week had already tuned the tape for volatility. The July 9 rebound worked as a mechanical reset for oversold names, with Wall Street focus rotating back to the AI trade.
Also on CFinance:
- Samsung DeepSeek Signals Rock Nvidia, Micron and AMD
- How to Build a 6-Month Emergency Fund: The Math
- Delta Air Lines Reports Q2 Friday with EPS Down 31%
AI Compute Demand Can Absorb the Chinese Diversification Push
The bull case rests on how deep the current AI compute buildout runs. Global cloud and AI infrastructure spending is projected to approach $1.5 trillion by 2027, a 40 to 50% annual rate of increase. That number frames every catalyst, up or down, that hits the tape from here.
Nvidia keeps roughly 60% pricing power in the accelerator layer via its design lock-ins and CUDA software gravity. SK Hynix, which supplies Nvidia, controls 60% of HBM. Even if DeepSeek internalises some inference chip work, the accelerator layer and the memory layer stay Western and Korean-controlled. The moat rebuilds around geography and IP, not around one chip vendor’s design.
Samsung’s 1,810% Q2 profit jump reinforces the point. AI memory demand is not slowing, and Samsung is the second HBM supplier behind SK Hynix. The bull view models Samsung and SK Hynix jointly harvesting the ramp-up from every AI trainer regardless of DeepSeek’s internal chip work.
Positioning tells its own story. Micron sits at 13x forward earnings versus SK Hynix at 6.8x forward earnings after its US debut, a valuation asymmetry that leaves room for Micron to compress lower or SK Hynix to reprice higher. Either scenario is a tailwind for the memory group.
First Cracks Show Up in the Nvidia Moat Story
The bear case sits on a simpler chart. Every large AI customer is now trying to build its own chip. Meta runs its MTIA line, Amazon leans on Trainium, OpenAI works with Broadcom, and DeepSeek adds another data point to a strengthening trend. What was once Nvidia’s exclusive customer list looks more like a set of design shops in the making.
The AI chip stocks slide of 5% earlier in the week already showed how sensitive the group is to any signal that dilutes Nvidia’s design lock-in. A confirmed DeepSeek chip would compound that pressure, especially if the design targets inference workloads where the CUDA software moat matters less.
Micron trades at 13x forward earnings, roughly double SK Hynix at 6.8x. The bear read is that Micron’s premium prices near-perfect execution and full HBM ramp-up. Any air pocket in AI capex, plus a DeepSeek chip that partially replaces external accelerators for inference workloads, and Micron’s multiple compresses fast toward SK Hynix’s level.
The asymmetry the investor has to price in, is that DeepSeek’s chip sits at the design and hiring stage, with no benchmark, no foundry partner and no prototype. But the market already rerated Nvidia off the Reuters leak alone. That gap between a speculative disclosure and a real book impact defines the bear case cleanly: the moat is intact today, but the moat is now known to be under attack from multiple sides at once.
More to come.
About the Author
