To The Moon

The question the Tiger thread is asking is the right one. Thursday's session produced one of the more confusing divergences of the year. NVDA closed up 8.74% after delivering revenue up 106% year on year, data centre sales up 117%, and a forecast for roughly 70% top-line growth next year. CFO Colette Kress explicitly named rising memory costs as the reason gross margin is heading from 75% today to a trough of 71 to 72% in Q4. High bandwidth memory now accounts for an estimated 30 to 40% of the cost to build an AI accelerator. The world's most important AI chip company just told the world, on an earnings call with full legal accountability, that memory suppliers have pricing power over it.

SNDK closed down 0.96%. MU closed down 0.32%. SK Hynix gained 2.27%.

That divergence deserves a straight answer rather than rationalisation.

The Three Reasons Memory Lagged

The first reason is that the cycle is already substantially priced in. Over the past month alone, the memory complex has absorbed Xiaomi's margin hit attributed to rising component costs, Intel GPU supply constraints up 48%, NVDA passing memory price increases of more than 15% through to customers, and Samsung's 110 trillion won payout plan failing to impress the market. Each of those was a demand confirmation signal. The market has been pricing memory pricing power for months. When the confirmation finally arrives from the source, the reaction is muted because the trade is crowded and the news is expected.

The second reason is NVDA's own gross margin guidance. NVDA's gross margin should fall from 75% today to a trough of 71 to 72% in Q4 before recovering to a 72 to 73% range in fiscal 2028. High bandwidth memory now accounts for an estimated 30 to 40% of what it costs to build an AI accelerator. The market is reading this in two simultaneous directions. Bullish for memory: NVDA is being squeezed on margins because memory suppliers have pricing power. Bearish signal embedded in the bullish read: if memory costs are compressing NVDA's margins this aggressively, at some point NVDA will push back, find alternatives, or reduce order volumes. That second thought is keeping a ceiling on memory stocks even as the first thought provides a floor.

The third reason is the tariff overhang. Reports emerged Thursday that the Trump administration is considering new semiconductor tariffs. That headline hit simultaneously with the NVDA report. Any tariff structure that creates friction in the supply chain for memory components introduces uncertainty about demand volumes that the NVDA gross margin compression story does not resolve.

What the NVDA Numbers Actually Mean for Memory

Set aside the one-day price action and look at what the NVDA earnings actually imply for memory suppliers over the next two quarters.

The most profitable hardware business on the planet is temporarily routing a few billion dollars a quarter to the three companies that make the one component it cannot make for itself. That is not a cyclical blip. That is a structural transfer of margin from the AI chip ecosystem's most profitable layer to its most constrained one. When NVDA cannot make its own memory and its customers cannot build without it, the pricing power stays with the memory suppliers until alternative supply materialises at scale.

The deeper Q4 trough is the more informative number. If the 71 to 72% guide holds up when Nvidia reports on November 17, it is a timing story with a tidy ending. On the other hand, a sliding range would mean that the memory makers are taking more profit than Nvidia budgeted for. That is the specific number to watch in November. If NVDA's actual Q4 gross margin comes in below 71%, it means memory suppliers are extracting more than even the pessimistic Q4 guide assumed. That scenario is the most bullish possible outcome for MU and SNDK even though it reads as bearish for the headline NVDA number.

MU reports its next quarter on September 29. That print, the first major memory earnings after NVDA's explicit acknowledgement of memory cost pressure, is the direct test of whether the pricing power NVDA described is flowing through to actual revenue and margins. Analyst expectations for MU's September quarter currently sit around $6.70 to $6.82 billion in revenue. Any upside to that figure, against the backdrop of NVDA explicitly saying memory is its margin constraint, re-rates the entire complex simultaneously.

Apple CEO Tim Cook, Amazon CEO Andy Jassy, and Space Exploration Technologies CEO Elon Musk all highlighted the same pressure point: rising memory costs driven by soaring demand. Four separate CEOs from four separate companies have now named memory pricing power on earnings calls this cycle. That is not a coincidence or a talking point. It is the supply-demand reality of a market where DRAM supply fulfils roughly 75 to 80% of demand today, falling toward 60% in 2027.

The Pick Level From Here

At current prices, SNDK at $1,483 and MU at $930, the thesis has not changed but the near-term setup has a specific structure worth understanding.

For SNDK, Bank of America analyst Wamsi Mohan raised his price target on SanDisk to $2,500 from $2,100 while reiterating a Buy rating, expecting strong pricing conditions to extend through mid-2027, albeit with moderating quarter-over-quarter growth rates. The $2,500 target implies 69% upside from current levels. Bernstein analysts believe that even in a worst-case scenario of a memory price collapse worse than 2010, the long-term agreements should significantly mute earnings downside in 2029 and 2030. With 60% of volumes covered by these agreements, SanDisk's fiscal 2030 earnings per share would be $214 even if average selling prices fell from peak levels. The downside is contractually floored. The upside is leveraged to every NVDA margin compression quarter between now and November.

The specific entry logic: SNDK between $1,400 and $1,500 is the zone where the contracted revenue base provides a valuation floor that the spot NAND price cannot undercut. Above $1,700 before NVDA's November print confirming actual Q4 gross margin is where the risk-reward deteriorates. The thesis is not broken by Thursday's sideways session. The thesis is strengthened by NVDA's explicit acknowledgement. The stock simply needs September 29 MU earnings and November 17 NVDA actuals to convert the acknowledgement into confirmed margin transfer.

For MU at $930, Trump has called Micron "one of the hottest companies" after a "massive" deal, adding a policy tailwind to the structural demand story. The forward PE of just 5 to 6 times looks strikingly cheap against Wall Street's $1,507 target. CEO Sanjay Mehrotra says multi-year Strategic Customer Agreements will lock in durability. With HBM4 shipping and HBM4E targeting calendar 2027 volume, $1,600 requires only modest multiple expansion. [Yahoo Finance](https://finance.yahoo.com/markets/stocks/articles/why-best-may-yet-come-150013709.html) The September 29 earnings date is the most important near-term catalyst. It is the first major memory print after NVDA's margin compression guidance, and it will either confirm or challenge whether pricing power is flowing through to actual revenue.

For SKHY, Thursday's outperformance of plus 2.27% against flat peers is the most informative data point of the session. As the primary HBM supplier to NVDA, SK Hynix is the most direct beneficiary of the 30 to 40% HBM cost share in AI accelerators. When NVDA's gross margin compresses because of HBM costs, the P&L line item that compressed it sits inside SK Hynix's revenue. Thursday's outperformance is the market beginning to price that read-through more specifically than it prices the broader NAND story.

The Honest Read on Thursday

The divergence on Thursday is not a crack in the memory thesis. It is the market asking a reasonable question: at what point does NVDA's margin pain become a demand reduction rather than a pricing confirmation? The answer is not visible yet. NVDA's $108 billion revenue forecast for next year, driven by AI accelerator demand that it explicitly says is supply-constrained, is not the profile of a company about to reduce memory orders. It is the profile of a company that needs more memory than the market can produce and is paying above its preferred price to get it.

The sideways session on Thursday is the price of being in a trade where the thesis is confirmed and the valuation is already elevated. The next confirmation event is September 29. The event after that is November 17. Between now and then, the entry at current levels is more compelling than it was at the post-Investor Day peak of $1,667 for SNDK and $971 for MU.

The pricing power is not in question. The only question is when the stock price catches up to the revenue reality that NVDA just disclosed.

I am not a financial advisor. Trade wisely, Comrades!