Lukas

@babyfolio I wish I could get into ALAB. Alternatively, NAND with Kioxia or SNDK. Vera Rubin comes with a monster memory rack and memory hierarchy - not just HBM. But, I'm worried about macro: rates (Thu jobs report) and hyperscalers potentially guiding reduced capex at earnings.

Memory looks good here. I've been adding a little. What are you guys buying?

Fully agree! Value accrues to the layer in the value chain that commoditizes the other layers. Memory: fungible GPU/TPU: competitive and the CUDA moat is gone. Semi fabs: potentially a moat but geopolitically not desirable. Data center power and land: commodities. AI models: open weights is only 6 month behind. What remains: the application software + data + customer relationship.

I now admit $IREN bulls were partly right when they said AI cloud software will commoditize. As always, they were thinking one chess move too short. When software gets commoditized, basically if or when they build 80–90% of the $NBIS stack, the advantage moves to whoever owns the most proprietary data linked to many different workloads running on that software stack. Basically, we move from peak AI cloud value accruing to megawatts in the early game, then to the best inference providers in the mid game, and finally to proprietary cloud data and AI models in the late game. Science fiction stuff like reprogramming how hardware runs in real time to make inference faster, cheaper, and more reliable. Revisit this in 5 years and tell me I was wrong. Cheers. (Not investment advice.)

@daniel_koss We'll likely see an open weights Fable/Mytos class model still in 2026.

@daniel_koss GLM 5.2 is arguably on the same level as Opus & GPT 5.5. It's available under the most open open source license. Pair this with Ant and OpenAI models now being a business risk for enterprises b/c the US govt can just pull the plug. Very bullish news for AI compute stocks

With both OpenAI and Anthropic now essentially not giving billions of potential users access to their latest models, how obvious has it become that the world needs frontier level open source? Or at the very least a new, non-US player to force competition for international users? Right now the frontier labs are so far ahead, that even their 2. newest models are still better than the frontier alternatives, so there's no pressure to release the best ones.

@intelfabs @AtlasShrug1 @ExponentialView Report values AI infra off blended $7B/GW revenue - way too low, look at SPX deals with Google or Ant at $50B/GW. It assumes falling token prices - opposite of what's happening - and misses that inference-heavy revenue has very different economics.

Neocloud won’t end well For the full report: follow @ExponentialView Very interesting and it is full for charts from my smooth brain to understand

@daniel_koss Curious how you approach point 2)

@AtlasShrug1 @babyfolio You're arguments are just correlations to other bubbles. That's why Sep timing is likely wrong. By 2000, only 7% of fibers were lit. Today, every GPU and GB of memory is sold out and utilized. We're still far away from the AI bubble popping.

You know the technology, I know the markets. Let’s revisit this convo in September. No sense in arguing, if you can’t understand the consequences of all of the chart/cycle/macro/market history things I post when you synthesize them together, you’re gonna drive the 🚌 off the cliff w/ everyone else. This isnt about $NBIS or any particular stock, it lies a layer above individual names, in bubble and liquidity trap, and the evidence that points to the fact that we are nearing the end of the road for this secular bull market.

I think the neoclouds are going to get a massive de-rating…soon and abruptly.

@JoshTradeOption @DollarCostAvg Sundar Pichai: the risk of under-investing is dramatic, and the risk of over-investing is far smaller

@DollarCostAvg has it right! $MU

@babyfolio Wait the key to predicting markets is just copy-pasting curves!

What do we even call these types of posts? And more importantly... how do people build massive followings posting stuff like this? The guy knows next to nothing about stocks, got liquidated multiple times in crypto, and now spends his time selling fear on X. This has to be a joke, right?

@InvestNorthwise CapEx projection seems too low. Industry estimates are ranging from $40-80B per GW. 5GW capacity would require $200-400B Capex.

Our full $NBIS model is officially here! Check out the full report in the first comment. Since our first model Nebius received a significant $NVDA investment, a very underrated $META backstop as a financing lynchpin, and is nearing the 20 site mark globally. We continue to build our Nebius model from the ground up using a site by site energization method that we now have visibility to run out through 2030. Thanks to interviews and events with @daniel_koss @mvcinvesting @romanchernin Tom Blackwell and many other Nebius members and contributors, we gained immense insight into how Nebius continues to execute faster that even our aggressive bull case could image. While we were directionally correct that mix shift would continue to weigh more towards ai cloud contracts with enterprise and ai natives through the end of the decade, we couldn't imagine that it would be near achieved in 2026. Our 2030 Base case below illustrates just how forward our expectations for growth have shifted at this rate of execution. Capacity — Connected MW (base case) 2026: 905 2027: 2,142 2028: 3,964 2029: 4,646 2030: 5,200 Undisclosed data center expansion bucket — Connected MW (base case) 2027: 175 2028: 425 2029: 600 2030: 739 ARR per MW (M, base case) 2026: 9.9 2027: 11.3 2028: 12.8 2029: 13.8 2030: 14.5 Exit ARR (B, base case) 2026: 9.0 2027: 24.2 2028: 50.7 2029: 64.1 2030: 75.4 Recognized revenue (B, base case) 2026: 3.4 2027: 15.8 2028: 36.1 2029: 58.1 2030: 70.3 Gross CapEx (B, base case) 2026: 25.0 2027: 39.4 2028: 59.6 2029: 26.2 2030: 25.2 Cumulative 2026–2030: ~$175B Funding assumptions (base case) 1. Prepayments, % of CapEx: 55% 2. Core OCF, % of EBITDA: 70% 3. External gap, debt/equity: 85/15 4. Blended interest cost: 5.5% Funding outcomes (B, cumulative 2026–2030, base case) 1. Prepayments: ~95 2. Core OCF ex-prepayments: ~49 3. Debt raised: ~34 4. Equity raised: ~6 5. Ending debt: ~43 6. Ending cash: ~20 Adjusted EBITDA margin (base case) 2026: 40% 2027: 42% 2028: 44% 2029: 45% 2030: 45% Implied 2030 adj. EBITDA: ~$32B D&A (B, base case) 2026: 2.9 2027: 8.1 2028: 16.2 2029: 23.1 2030: 27.3 Share count (base case) Ending diluted shares: ~339M Base case scenario probability weight: 55% Thank you to our premium members for your massive support in bringing this refresh so quickly. Price targets, our portfolio allocation, present value calculations, and our buy/hold/trim/sell zones are now live.

@optionscjp the pets.com of AI

Perplexity available to invest at $20B valuation Worth it?

@babyfolio NVDA even made neoclouds a separate item in their latest earnings report. I don't think it's bearish: likely significant compute efficiency gains (tokes per watt) to be gained from system-level integration and optimization. Like Apple vs Windows

Not surprising. Companies like $CRWV, $NBIS, $IREN and others are heavily dependent on access to GPUs, so they don't exactly have much leverage in these situations. At the end of the day, Jensen is going to do what's best for $NVDA and its shareholders, that's his job. The good news is that NVIDIA's GPUs remain the gold standard. As long as they continue to offer the best GPUs on the market, it's nothing to worry about.

In CPO (according to GS), value of Fiber+FAUs is x2 the value of Laser content! 😱 The Optical Engine is the major part of the value, but now it's not in the Transceiver! $LITE $GLW $AAOI