tan retweeted
Worst 60-day sell-off for tech momentum in market history, by far.
-57% for Jefferies' tech momentum index.
tan retweeted
Similar to the panic over DeepSeek R1, some uneducated people think Kimi K3’s use of linear attention (KDA) is bad for NVIDIA, HBM, DRAM, and networking because it has relatively lower KV-cache requirements. The opposite is true, and we explain why below. 👇️ 1/8🧵
tan retweeted
So Jensen did pre-pay a bit
Others will probably follow suit
At this rate, TSMC might spend $200B in the next 3 years (literally double compared to 2023-25)
As Ming Chi Kuo said, TSMC is the most capital-disciplined player in the AI supply chain
If they are willing to double their CAPEX then compute demand is tremendous
TSMC announced 2026 capex of US$52–56bn at today’s earnings call, with strong Nvidia demand as the key driver.
1. Ahead of the call, sell-side analysts and media generally expected 2026 capex of US$45–52bn, while the buy side was already at US$53–56bn.
2. TSMC is typically conservative on capex guidance, so final 2026 spending is very likely to beat buy-side pre-call expectations.
3. Most customers negotiate for guaranteed capacity. Nvidia CEO Jensen Huang negotiates for land. In November last year, he visited Tainan and told TSMC he was willing to pay to secure the reserved P10 and P11 land next to Fab 18. At the time, Fab 18 plans were only clearly defined for P1–P9, and Jensen Huang’s ultimate goal was to secure the P12 land as well.
4. Before Huang raised the idea of securing P10 and P11 land, buy-side consensus for TSMC’s 2026 capex was US$45–50bn.
5. Even with tight advanced-node and packaging capacity, TSMC stays highly disciplined on demand approval, often approving less than customers’ most optimistic requests. Want the most aggressive capacity? Put more cash down, book the land, and pay to grow capacity—just as Huang did. That said, with resources increasingly constrained across the board, even if other customers try to follow Huang’s playbook, securing the most aggressive capacity is becoming harder and harder.
tan retweeted
When a leader like $DELL is engulfing days worth of action it's a reminder to go outside and touch grass.
tan retweeted
Nebius $NBIS agreed to sell $1B+ of AI compute to Reflection AI through 2029. The deal gives Reflection AI access to Nvidia GB300 chips.
Reflection AI, founded by two former Google DeepMind researchers, also signed a multibillion-dollar compute deal with SpaceX last month and has reportedly discussed raising $2.5B at a $25B valuation.
Nebius already has compute agreements with Microsoft and Meta.
tan retweeted
Never been a better summer to deep research, place your trades, then disappear. The noise is louder than ever. Most of it won’t matter.
tan retweeted
To everyone who is feeling afraid after this recent market drawdown:
If you believe AI is "real", and will be creating levels of value humanity has never seen before, I want you to look at the bigger picture. Yes, we could still go down another 50% from here. Nobody can predict the short term.
But if you believe in AI (and I think you should), the evidence is everywhere: agents are here, AGI is plausible within the next year, autonomous driving improves weekly, robotics improves weekly, and new use cases keep showing up. AI is not fake. It's going to change business in every way, shape, and form.
You'll see headlines about Meta struggling to properly utilize agents, or selling off excess capacity and becoming a neocloud. That's a Meta/Zuckerberg specific problem with being able to utilize AI, and a lot of these headlines were completely taken out of context. Either way, it says nothing about the trend or the future. Every hyperscaler is still spending billions more every single quarter on capex. Robots, self-driving, LLMs, cybersecurity... all still accelerating.
The value being created right now is indescribable, and it'll keep compounding. The only thing you have to do is stay invested.
However, this is NOT the time to be on margin. You can't predict the future and you don't want to get liquidated. But completely sitting out this semiconductor rally, sitting out the AI infrastructure buildout because of negative headlines? Ridiculous. There's a gold rush happening right in front of you. This is, in my opinion, the best wealth-building opportunity in human history.
If you're still worried, zoom out. Look at the run we've had these past few months. Even if we gave all of it back, that doesn't mean AI isn't real. Rate hikes won't kill the thesis, and you will be a successful long-term investor.
I'm personally down nearly $1M from my peak. But through several years, I'm still up much more than that, and I know I'll make many times more over the next 5 years.
We're all going to make it. Sorry to everyone who's lost money in this downturn, I know how it feels.
tan retweeted
For those interested in $NOK ...
Nokia’s 6-inch exposure came through the Infinera acquisition, and it is a fundamentally different kind of bet. Coherent, Lumentum and AAOI are mostly laser and discrete-device stories. Nokia/Infinera is a PIC story, chips that fold lasers, modulators, amplifiers and detectors onto a single piece of InP.
That changes the manufacturing problem. The die is bigger, the process is more complex, and the yield math is harder, but the payoff is that far more of the optical system gets integrated onto one chip.
The new San Jose fab (the 6-inch site) backed by the CHIPS Act is moving through qualification now:
“I think I know the new fab is finally flowing gas, flowing chemicals. So it’s -- all the equipment is being turned up. We’re qualifying it this year. It will start production late this year, early next year.”
What that fab is built to produce is the key differentiator:
“A key thing for us in this... is the majority of the volume that we’re building is this, and particularly in that facility, is this photonic integrated circuit for pluggables.”
And management is upfront that integration cuts both ways:
“It’s a full photonic integrated circuit. So the die size is a little bit bigger. So that creates a different... yield dynamics.”
That’s the trade. A larger PIC removes assembly steps and packs more value onto the chip, but bigger die are more exposed to defects. The 6-inch platform hands Nokia more wafer area to work with. The open question is whether those large integrated devices yield well enough at scale.
Important to know that Nokia is building the whole chain domestically, not just the fab. The San Jose InP site is paired with an expanded advanced test-and-packaging operation in Pennsylvania (we just spoke about this a few days ago), because pushing far more wafers out the front of the fab is worthless if packaging becomes the next choke point. By combining larger wafers, more advanced tools and more total tool count, management has framed the build-out as a step-function increase in capacity for complex optical chips, timed to support the surge in 2027 demand.
Nokia’s angle is all about internal supply, integration, and control over its own optical-engine roadmap.
tan retweeted
well given the price action in today’s global equity markets it might be fair to say that people are a little nervous going into $MU earnings
which is a far better reason for a selloff than rate hikes…
crude hit BELOW $70 today and we are still talking about rate hikes?!
tan retweeted
SK Hynix stock is about to begin trading on the US exchanges on July 10th
SK HYNIX IS GOING PUBLIC IN THE 🇺🇸 IN JUST 3 WEEKS
SK Hynix is planning a $29.4 Billion US listing with trading set to begin on July 10th
This would be one of the top five share sales in history, comparable in size to Saudi Aramco's record 2019 IPO.
SK Hynix controls 57% of the global high-bandwidth memory market by revenue. HBM is the chip that sits inside every NVIDIA GPU powering the AI buildout. Without it, the data centers do not work.
The numbers behind the business:
Q1 2026 operating profit: $24.3 billion, a record high
Q1 2026 sales: $34 billion, nearly tripling year over year
Stock performance: +300% in Seoul this year
Market cap: over $1 trillion
The proceeds will fund additional capacity expansion and purchases of EUV lithography machines needed to make next-generation chips.
Lead underwriters: Bank of America, Citigroup, Goldman Sachs, JPMorgan.
tan retweeted
AI demand is outstripping Moore's law in the short run
Moore's law drove import prices of computers and semiconductors down by 52% between 2001 and 2020. (1/4)🧵
SK Hynix Throttles HBM4 Production… Seeks Additional Profit by Expanding Supply Constrained Commodity DRAM
SK Hynix is moderating the pace of its sixth generation high bandwidth memory (HBM4) ramp and placing greater weight on capturing the commodity DRAM market. With its HBM revenue share already exceeding 40 percent and an overwhelming lead established, the company is reallocating resources toward securing additional profit from commodity DRAM, where supply shortages are acute, rather than engaging in an aggressive capacity expansion race.
According to industry sources on the 23rd, SK Hynix is reportedly delaying somewhat the conversion of some fifth generation HBM (HBM3E) production lines that were originally slated to switch to HBM4. The aim is to expand its ability to respond to the commodity DRAM market, which currently posts higher operating margins than HBM, in order to secure additional profit. The industry view is that, having already established a solid position in the HBM market, the company sees no need to rush its transition to HBM4 and HBM4E (seventh generation HBM).
Behind this strategic shift lies a reversal in profitability between commodity DRAM and HBM. As of the first quarter of this year, the per gigabit (Gb) price of commodity DRAM still falls short of HBM, but the operating margin gap is estimated to have already widened to more than 15 percentage points (P). Daishin Securities forecasts that commodity DRAM operating margins could rise to a theoretical maximum of 90 percent within the year.
A source familiar with SK Hynix said, "From the standpoint of SK Hynix management, they cannot ignore the fact that a competitor (Samsung Electronics) is already earning enormous profits from commodity DRAM rather than HBM." The source added, "SK Hynix's HBM4 is still undergoing NVIDIA's quality certification process, and production forecasts for NVIDIA's next generation chip 'Rubin,' which will carry HBM4, are trending downward, so there is no reason to accelerate the HBM transition."
The view from overseas investment banks (IB) also supports this trend. Goldman Sachs assessed that it would be sufficient for SK Hynix to maintain a dominant position of more than 50 percent in HBM3 (fourth generation HBM) and HBM3E (fifth generation HBM) at least through 2026. Morgan Stanley pointed to memory wide pricing cycles, rather than a defense of HBM share, as the key driver of SK Hynix's value, and raised its earnings estimates by 56 to 63 percent on the basis of a forecast that DRAM average selling prices will rise 62 percent in 2026.
Indeed, in its first quarter earnings release SK Hynix stated that DRAM average selling prices (ASP) rose in the mid 60 percent range, and it presented a plan to focus on meeting demand for high density server modules and mobile products. Its signing of a three year DDR5 supply agreement with Microsoft (MS) is also interpreted as a move to secure long term earnings visibility in commodity DRAM.
On the other hand, as SK Hynix moves to adjust HBM4 volumes, the possibility of a share gain by its competitor Samsung Electronics is also growing. According to Counterpoint Research, SK Hynix's HBM market share stood at 57 percent in the fourth quarter of last year but is increasingly seen as likely to shrink, and some observers suggest that if Samsung Electronics succeeds in mass producing HBM4 in the second half of this year, SK Hynix's share could fall to the 50 to 60 percent range.































