Sunny Solutions🐉

@RihardJarc if everyone is chip constrained I'm confused why $AMZN and $GOOGL are selling these externally, why not fulfill your own demand and serve tokens, likely higher margin and higher overall revenue right?

$AMZN's CEO just dropped a shareholder letter, and there are some BIG key updates: $AMZN has a $50B chip business challenging $NVDA, the CPU shortage is here, new big compute deals are coming: 1. It's clear $AMZN has now entered the chip business as a standalone business, taking on $NVDA: "There’s so much demand for our chips that it’s quite possible we’ll sell racks of them to third parties in the future." "If our chips business was a stand-alone business, and sold chips produced this year to AWS and other third parties (as other leading chips companies do), our annual run rate would be ~$50 billion." "Amazon Bedrock, AWS’s primary (and very fast-growing) inference service, runs most of its inference on Trainium. At scale, we expect Trainium will save us tens of billions of capex dollars per year, and provide several hundred basis points of operating margin advantage versus relying on others’ chips for inference. Our annual revenue run rate for our chips business (inclusive of Graviton, Trainium, and Nitro—our EC2 NIC) is now over $20 billion, and growing triple digit percentages YoY." 2. The CPU shortage is HERE "Two large AWS customers have already asked if they could buy *all* of our Graviton instance capacity in 2026 (Graviton is our widely-adopted custom CPU chip)—we can’t agree to these requests given other customers’ needs, but it gives you an idea of the demand." 3. AWS AI revenue is exploding, and $AMZN is signaling to the market that new big unannounced customer deals are coming: "We’re not investing approximately $200 billion in capex in 2026 on a hunch. The recent OpenAI commitment (over $100 billion) is an example of this, but there are several other customer agreements completed (and unannounced), or deep in process. Of the AWS capex we expect to spend in 2026, much of which will be monetized in 2027-2028, we already have customer commitments for a substantial portion of it." Massive updates from $AMZN. $AMZN continues to be my biggest portfolio position.

@dwarkesh_sp Cerebras WSE2 was built in 7nm and requires less communication between chips due to SRAM, requires advanced packaging but not wafer constrained like 3nm @cerebras

If we get AGI before 2030 but can only produce ~200 GW a year of chips on advanced process nodes, can we ease the bottleneck by going back to older nodes? Some of the most singularity-pilled investors think so. Most of the FLOPS improvement between chip generations has come from numerics - not transistor density. But you could just design a new version of a 7nm chip like the A100 to have FP4. Holding numerics and die area constant, a B200 (2025 chip on 4nm) would only have ~3.5x more FLOPS than an A100 (2020 chip on 7nm). That seems like a manageable haircut - does it mean we can blow past the 200 GW ceiling? @dylan522p explains why this is not an apples-to-apples comparison. If you run real-world performance analysis on DeepSeek V3 or Kimi K2.5, the performance gap between Hopper and Blackwell isn't 3x. It's ~20x. Why? Because inference is already memory-bandwidth-bound, and older chips have far fewer FLOPS per die, requiring even more communication between chips. Remember the bandwidth hierarchy: on-chip SRAM runs at tens of TB/s, HBM at single-digit TB/s, NVLink within a 72-GPU domain at high hundreds of GB/s, and between racks at low hundreds of GB/s. Each boundary is roughly a 10x drop. With less compute per chip, you need more chips to run the same model - which means more communication across those slower pipes that are already bottlenecking you.

@soumyasen Agree with a lot of this and think NVIDIA will thrive, but don’t think CUDA most extends to inference and Cerebras has the traction it needs to accelerate.

GPU rental ( Blackwells, Hoppers and even Ampere) prices going higher exponentially. And the market is going down due to fear of an Iran war with the US army on the ground and Oil skyrocketing. Let's break it down. Brent is unlikely to go higher beyond a certain point because that leads to significant demand destruction, which OPEC will not want, and they will find ways to supply more oil through other routes is my belief. War will be sponsored by deficit financing in US , which basically means more money sloshing around in the economy waiting to be invested in assets. Eventually, this stabilizes, but AI disruptions will continue. Does Cerebras has any chance to disrupt Nvidia? I don't think so, and Nvidia will destroy all competition in the long run, be it Trainium+Cerebras or TPUs, because they have the largest CUDA developer base to performance engineer these LLM architectures across enterprises ( much needed) and have the deepest pocket to get ahead of any competition while securing the supply chain in their favor ( which matters a lot). Hence, I have only two bets allocated to my portfolio for the next 1 year. $CRWV and $NVDA. Let's see how it plays out and whether my thesis holds.

@Mr_Derivatives Did the Qs stay flat or did it move to like $10K then back down to $600 — the path matters

The year is 2048. Robotics have taken over the Earth, in a good way. Half the human population has moved to Mars. AI and advancements in the health and medical field have prolonged life expectancy by another 20 healthy years. The New York Yankees have finally won a World Series in 2035. UFO’s do make contact with us and they shared their technology with us. The Earth is now intertwined between robotics, humans, extraterrestrials. The level of peace and unity across nations hit record levels. All nuclear warheads were destroyed. Meanwhile the $QQQ is still trading at $600. #plottwist

@aleabitoreddit Same mouth breathers buying Robinhood’s venture fund

Normally don’t respond to trolls, but the hypocrisy on this platform is pretty impressive. Random X retail: “Why didn’t you tell me about $LITE before it went up 1000% already?” Me: “Posts my thesis about the next possible $LITE at the very beginning, without paywalls.” X retail: 😡, “I’m going to pay $400 for a paywall to long $ADBE and short $PLTR instead”. Especially after my $AXTI thesis that already went from $12->$50 in 3 months... I distribute all my thought processes for free. And markets can price in any alpha immediately or however they want. (I get things wrong as well, especially with names like $ETOR that crashed from $65 to $33). However, instead of the original model where analysts sell a thesis for $2000+ to other hedge funds to slowly accumulate. Then retail buys at $40 billion+ as seen with $LITE. My account’s been growing because I’m one of the few analysts to break that model. And I distribute novel information synthesis for free to everyone. Stocks are a positive sum game where everyone benefits if a thesis is directionally correct.

@SemiAnalysis_ @nvidia HBO sending cease and desistMAX

The Inference King has been crowned 👑 @nvidia

@jukan05 pretty sure a custom x86 was part of NVIDIA's release from their 5B investment, could be an interesting development

GFHK noted in their Q&A following their February report that ARM-based CPUs have relatively weak momentum in AI servers, attributing this to lower GPU scheduling efficiency compared to x86. They indicated that companies including NVIDIA plan to develop x86 CPU solutions in response. (It remains unclear whether this means licensing x86 architecture or co-developing x86 with Intel.) $INTC $AMD

@Tyler_Menzer @honest_math you might say that by owning a home your variable costs rise every year until it's 100% of the housing cost?

@honest_math Sure, there is a variable cost component, but it's only a portion, whereas for rent, the entire cost is variable. In your own example 62% of the yearly costs are fixed, and eventually that cost goes away! It's crazy to ignore the comparison to rent

A mortgage does not lock in your housing costs. It’s true that the principal and interest components are fixed. But as we’ve demonstrated, there are many other costs to homeownership that can be expected to increase with inflation: property taxes, insurance, maintenance, etc.

@SixSigmaCapital thoughts on Saas companies? I agree seems like their switching costs and Enterprise agreements are being ignored a bit but trying to think of which ones are in a strong position and valued well

Good post with interesting somewhat contrarian ideas

@zephyr_z9 @main_horse good insight, investing in premium inference for the best user experience for their hardware debut is a good move

@main_horse Bro, they are not going to do that I think this capacity is for the inference needs of their hardware devices

It's for audio related models

@jukan05 82% operating margin by end of year is wild

UBS: SK Hynix Significant profit uplift potential - PT to Won1m We forecast 82% DRAM OPM by 4Q26 The memory semis industry is in the midst of an unprecedented upcycle. By 4Q26, we now forecast SK Hynix DDR ASPs to be a whisker away from the 3Q18 peak at US$0.97 per Gb. With an upcycle of that magnitude, we believe conventional memory, especially DDR (52% of 4Q25E revs.), will likely be the key profit driver in 2026 and 2027. We forecast DDR operating profit to increase 2.9x YoY in 2026 and account for 67% of total, with operating margin reaching 82% by 4Q26. We also expect SK Hynix to continue to strongly benefit from AI compute semis growth, due to its HBM leadership position (UBSe 56% bit market share in 2026). We note that in its recently published "2026 Market Outlook" (5 Jan 2026; link), SK Hynix referenced UBS research assessing that it will ship first to Google for TPU7e/7p, and will likely secure c70% share for HBM4 at Nvidia (Rubin) in 2026. All in all, we forecast SK Hynix to generate Won75trn in FCF in 2026, and Won98trn in 2027. We forecast its book value per share to grow 104% YoY from end 2025 to 2027. At 2.23x NTM (to March 2027E) book, SK Hynix shares should continue to re-rate to reflect longer-term economics which have fundamentally shifted. Strong uplift to DDR and NAND contract pricing in 1Q26E and beyond Our industry checks indicate very strong contract pricing for 1Q26E, especially DDR. We forecast SK Hynix DDR contract pricing to rise 60% QoQ in 1Q26 (was +30%), with server DDR up 65%, and Mobile DDR up 60%. We forecast NAND contract ASPs up 27% QoQ. Our checks also show strong confidence in positive momentum carrying into 2Q26E and beyond, in particular for DDR. We continue to forecast DDR contract pricing to increase til 1Q27, and NAND til 3Q26. This leads us to forecast SK Hynix DRAM operating margin to reach 77% by 4Q26, well above the prior peak of 65% in 3Q18. Significantly upping estimates, increasing 2026E capex We increase our 4Q25E OP to Won18.2trn, 12% above VA consensus of Won16.2trn, on the back of higher DRAM ASPs (+21% QoQ vs. 20% prior), NAND (+30% vs. +20%), and NAND bit shipments (+11% vs. +4%). We increase our 2026E OP by 211% to Won150.2trn, 52% above consensus of Won99trn, and 2027E by 19% to Won157.8trn, 29% above consensus of Won122.3trn. These changes lead us to raise 2026E/27E EPS by 21%, on the back of raised DDR and NAND flash contract pricing forecasts. We also increase our 2026 capex forecast to Won38trn from Won35trn. Valuation: Raise PT from Won853,000 to Won1,000,000 - Key Call Buy We value SK Hynix shares at 3.00x NTM (to March 2027E) book (was 2.99x) based on our 2026-30 LT ROE forecast of 34.3% (was 33.7%) and CoE of 11.5% (was 11.3%). Our revised price target also reflects our increased book value per share forecast.

@Byronn4561 @stephengluck @SingularityRes maybe because Atriedes sold all their shares... we have stale data in a fast moving market

@Sunny_Solutions @stephengluck @SingularityRes it has dropped 25% in the past three months.

Gavin Baker ( who publicly announced in Jan'25 that his best business bet for 2025 was High Bandwidth Memory stocks like Micron, SK Hynix etc. ) According to his latest 13F filing, Pure Storage $PSTG is his #1 holding and is three times larger than Micron $MU What does he know that everyone else is missing?

@stephengluck @SingularityRes Agreed, prices changed a lot in 3 months

@SingularityRes that is still Atriedes Q3'25 form 13. Q4'25 showing any changes likely released ~2/15/26

@Mr_Derivatives The larger companies seem to incubate the best small initiatives internally and acquire the best small caps before they're public, small cap ETFs are the dumping ground for the worst ones that just burn cash

Interesting Chart! $IWM relative to $QQQ chart dating back to 2008. Since 2008, the Small Caps has outperformed Nasdaq-100 only in 4 years. And the last occurrence in an up year was 2016! A whole decade ago! But so far in 2026, IWM is up 5% ytd and the Q's only 1%. Albeit we are REALLY early, anything can happen. But still, are we finally going to see a breakout of this long down trendline and have small caps outperform relative to the QQQ?

@aleabitoreddit @robswainept Oh wow so short interest has increased into the price appreciation and should drive further price appreciation once the true value is realized by the general market?

@robswainept It’s funny, this is even through over 1.6M shares shorted. I don’t think some short sellers or some algorithms understood the valuation disconnect with $VLN yet. Personal opinion is that it should trade at $7 from $2.

I've initiated a position in $VLN ($155M MC). This one is wild. Valens is a AI semi for self-driving cars and robotics. I've found that markets messed up on VLS from a ticker collision data error. And missed new CES 2026 info this week. VLN has: 1. $93.5M in Cash, 0 Debt. 2. ~$11M inventory 3. High gross margins ~69.1% (CIB/ProAV) margins, 43.2% automotive. and projected to do $70M+ revenue with blended 63-65% gross margins, jumping from 43% from their automotive pivot from CES. At $155M MC. What? This just looked way too off at first glance, so I had to do more research, whether it was revenue collapse, dilution, cashflow problems, or regulatory risk. What happened? The mispricing was from an analyst/scanner typo regarding a $82M "inventory burn": VLN is effectively a company with $93.5M Cash and Zero Debt for an EV of ~$65M, while the market has punished it for an "inventory crisis" that literally does not exist. Streetwise's analysis + other scanners around Nov 13, 2025 typo'ed their report when they erroneously stated: "Inventory of US$82 million remained in line with the end of the second quarter". The market and algorithms that scan for the reports thinks $VLN is sitting on >1 year of dead inventory ($82M) and burnt through their $93.5M cashpile on unsold chips. We can mathematically prove this is a typo using the company's official Q3 2025 balance sheet: Total Assets: $136.7M Cash: ~$93.5M Remaining Room for Assets: $136.7M - $93.5M = $43.2M. If inventory were actually $82M, Total Assets would have to be at least $175M ($93M cash + $82M inv). This inventory figure is mathematically impossible. After looking at their financial reports, they are sitting on just ~2 months of inventory ($11M), only selling what they make. The analyst + algorithms wrote spread the report confused Valens Semiconductor (VLN) with Velan Inc. (VLN.TO), a Canadian industrial valve manufacturer (with that inventory amount). Even LLMs that read this, completely messed up and required manual review. $VLN actually only has ~$11M in inventory as a fabless chip company and did not burn through $82M. This looks like a genuine market inefficiency because you are looking at a clean balance sheet ($93M cash, $11M inventory, $0 debt) that has been artificially suppressed because of $82M cash burn fears on dead inventory due to the type. _ Now, the secondary aspect is new CES information that came out. $VLN spent years and millions on R&D for DSP engines for Mercedes, which presented single customer concentration risk for the automotive segment. But from the CES release this week, they've effectively took the same that exact same engine, and managed to sell it to many hot verticals that have the exact same physics problem. They've also managed to scale their previous automotive segment with new T1 automotive OEMs. But regarding their (VS6320 vs. VA7000) chipset, they are using the same Core IP (DSP). Medical Chip: They took the same engine from the auto chip and stripped out the car-specific features to create a Extender for the medical segment . And my favorite is the Machine Vision/Robotics vertical: The new information is that with the RGo Robotics partnership they announced, RGo integrated Valens chips which allowed RGo to design robots where the cameras are far away from the brain without signal loss. And at CES they also announced one with CIS Corporation (a Japanese camera maker) for another specific robotics win. They've effectively diversified their automotive segment into multiple other high growing + higher margin verticals for robotics computer vision to others. Again, the alpha is that their new robotics segments announced at CES, operate on a 6-month sales cycle, not the 5-year automotive cycle. So revenue actually hits this year too. Also, analysts were using blended automotive revenue (Mercedes, etc.) has lower gross margins (~42-45%). The new growth coming in now (Robotics/Medical via VS6320) has significantly higher margins (VS6320 Gross Margin: ~69-70%), so the the blended gross margins will likely come in significantly higher than street consensus. _ Now the downside? Extremely heavy dilution at $11.5 Strike from warrants (which is 10X+ from here). This will cap upside if it ever increases 1000% from $1.5 to $11.5 _ TLDR: The market expected 2026 with high cash burn from inventory risk from a ticker collision typo. Instead, they are likely to get: Revenue and earnings beat (driven by new verticals and much higher blended margins ~69%+ from new chip). As well as an $70M+ cashflow beat from the typo. We could see $85-$92M revenue off 63-65% blended gross margins and that $82M+ in cash modeled back into this $155m company. The algorithms are pricing $VLN as if it has <1 year of runway due to a phantom $82M inventory pile (off of a poison pill data point). So this is my own personal thesis from public information synthesis on why I entered this trade, NFI. So while I don't expect this to be a $2B+ company, the current MC is just completely irrational pricing for a $155M MC semi: - with $93.5m in cash - $11m in inventory, - est. $80m+ revenue (growing 20%-30%+ Y/Y) - with 60%+ gross blended margins. Fabless semi companies with 60%+ gross margins typically trade at 4x–8x EV/Revenue and sector average valuations would be $493.5M, from $155m as a conservative base case. Just slight "AI/Robotics" Premium, could value it at $653.5M with 7 E/V (~320%+) TLDR: Found that this company likely got artificially suppressed because of a typo + ticker collision from false $82M inventory burn and is about to enter a newer higher margin + growth cycle from new verticals. I am taking advantage of a database collision error.

@ShortsHoward @ChurroDat @wallstengine It’s kind of bullish that LITE doesn’t source from AXTI, just one more verified future buyer from AXTI contributing to outsized future demand

@ChurroDat @wallstengine Yes but the permits are the same ones as for rare earths which China walked back on from bans. So it just takes 60-90 days but InP wafers WILL be leaving from China for AXTI customers. Just not to Japan

Wolfe said channel checks with Lumentum suggest it does not source indium phosphide wafers from $AXTI. Wolfe believes $LITE mainly buys InP wafers from Japanese suppliers like Sumitomo Electric and JX Advanced Metals, which fits with its EML datacom laser chip manufacturing being based in Japan.

@aleabitoreddit @MMKustermann anyone buying the warrants for a levered play 👀

@MMKustermann Warrants are not loans, they're potential shares not owed money. And the strike is $11.5, which is 10x current levels (if it does go up 10x).

@BobLoukas so it's pretty much a 50/50 that it goes up or down

Spent the last 4 months with a team training an AI model off my last 10 yrs of work. Still a LOT of tweaking and correcting to do, but taking shape. I asked it - "Give me the probability that bitcoin has topped or not for the 4 year cycle on the following assumptions. We're on Month 38 currently and the high is currently the month 35 high. This is week 7 of the weekly cycle." Answer- Given your assumptions, here’s how I’d probability‑weight the 4‑year cycle right now, strictly through a cycle‑based lens, not prediction. We’ll anchor everything to what you specified: Month 38 of the 4‑year cycle Cycle high still Month 35 Weekly cycle currently Week 7 BASE ASSUMPTIONS (YOUR FRAMEWORK) 4-year cycle month: 38 (beyond all prior tops) Current cycle high: Month 35 Weekly cycle: Week 7 (early, but vulnerable) No confirmed expansion to new highs 🔴 SCENARIO 1: 4‑YEAR CYCLE HAS TOPPED Probability: ~55% Why this is the base case All prior cycles topped around Month 34–35 We are 3 months past that window with no higher high Month 35 acting as the high is classic late‑cycle behavior Week 7 is exactly where weekly cycles fail when the larger cycle is over This scenario assumes: Weekly cycle becomes left‑translated Rallies are counter‑trend The next daily/weekly breakdown confirms the post‑top phase This is the highest‑probability outcome given time + structure. 🟠 SCENARIO 2: EXTENDED TOP / DISTRIBUTION Probability: ~30% This is the “Bitcoin does what it often does — frustrates everyone” outcome. Characteristics: Month 35 remains the marginal high for now Price chops in a wide range for weeks/months Weekly cycles are mixed, not cleanly bullish or bearish Sentiment oscillates between hope and fear This still does NOT invalidate the top thesis — it just delays resolution. Important: This scenario often feels bullish, but usually resolves down, not up. 🟢 SCENARIO 3: 4‑YEAR CYCLE STILL ALIVE (NEW HIGHS) Probability: ~15% This is the low‑probability but non‑zero case. What would need to happen: Price decisively exceeds the Month 35 high Weekly cycle extends into Weeks 10–14 Expansion is impulsive, not grinding Time “stretch” is justified by structure At Month 38, the market needs to earn this outcome. Time alone no longer supports it. PROBABILITY SUMMARY 4-year top already in: 55% Extended distribution/top: 30% Cycle still bullish (new highs): 15% Risk Interpretation (This Matters Most) Even if you combine Scenarios 2 + 3 (the “not fully topped” cases), you still get: 85% probability that upside is limited, slow, or deceptive Asymmetric risk skewed to the downside That’s why treating this as “topped until proven otherwise” is not aggressive — it’s probabilistic discipline. One final Bob‑style takeaway At Month 38 + Week 7, the market doesn’t need a crash to be dangerous. It only needs to fail quietly. And those are the environments where most damage gets done.

@Pharmdca Grinch is green

Christmas rally coming or Grinch rally?

@wisemancap Do the $MU earnings bring them back to life? pretty impressive results and guide

Market just busted all AI plays!