MinuteMoola (@MinuteMoola) header
MinuteMoola (@MinuteMoola) profile picture

Research on Space, Defense & AI. $RKLB @ $5, $NBIS @ $70, $EOS.AX @ $6.30 $PNG.V $MDA $SPCX Minute Moola on YT/IG. NFA

Joined March 2014
Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)Photo by MinuteMoola (@MinuteMoola)
A comprehensive view of the next incoming tech giant, Nebius. youtu.be/_Hi73CaVhBw?si=nhPV… $NBIS
2
5
1
12
4,507
MinuteMoola retweeted
Oh my god: Anthropic mathematician Levent Alpöge (@__alpoge__) says he has proven the Riemann Hypothesis, open for 167 years. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." He says the method also covers the Generalized Riemann Hypothesis. The proof is out as Lean code (a proof language where a computer checks every step), so experts should be able to check it quickly, Scientific American reports. Same Alpöge who was behind this year's Jacobian conjecture result and the Navier-Stokes work with Tristan Buckmaster. I assume we will get an official statement very soon from him!!
234
551
229
5,923
571,338
This about sums up how institutions are looking at the market right now. They’re uncertain on where value will accrue and what happens next. I’m dipping in and buying more of my favourites here and possibly at even better prices in the next couple weeks/months. $NBIS $RKLB $PNG.V / $KRKNF $SPCX $EOS.AX
7
399
$PNG.V / $KRKNF at C$2.93 = ~C$1.1B for a subsea defence platform that's announced C$457M of orders in 2026, 60%+ from navies and defence primes. That's ~10x my conservative 2027 EBITDA estimate (btw FY27 EBITDA will probably be around FY25 Revenue…) The market is valuing Kraken like it'll never grow again. Not loading the boats before Nov 24 though. H1 was C$49M of revenue and they’ve guided to C$290-320M for the full year. Hopefully Q3 is near ~C$100M or the low end gets a little dicey, and Kraken did miss its 2025 guide.
3
2
47
2,326
$NBIS continues to be thrown in the generic neocloud bucket. On OpenRouter this week, Nebius cut DeepSeek V4.1 Flash 40% and daily tokens on it went from 12B Monday to 610B Wednesday. 100% uptime, ~250 tok/s, and still priced above most providers on the model. Token factory is the growth engine the long-term bulls are betting on. Stock getting beat up while the real thesis is coming to life.
The story of $NBIS Token Factory suddenly pacing towards ~700 Billion tokens today (Most of that from Deepseek V4.1 Flash) is not just a price discount. It is the serving performance holding up through the volume explosion even with the discount holding token pricing significantly above most of the inference players on OpenRouter. This to be honest is the first time I was genuinely impressed at $NBIS Token Factory as a Developer and a Shareholder. After the 40% cut for the model, $NBIS pricing for DeepSeek V4.1 Flash 1017 is still above 66% of the inference providers on Open Router. So for people who aren't familiar with Token Prices this is not a "Crazy Discount" At 0.18 input and $0.72 output per million tokens, this is not at all traffic chasing the lowest price. The impressive part is availability maintained at a 100% throughout the Token Explosion, with remarkably high throughput and low latency on OpenRouter’s most popular model and arguably the most popular open weight in the market right now. $NBIS is delivering ~250 tokens/sec at P50. Above every regular offering slightly above even Morph's Fast (Premium) Tier at 246 and behind only Baseten Fast at 475 which are Premium Offerings with significantly high Token Prices. Both Fast tiers charge $0.60 input and $2.40 output per million tokens. Over 3x the $NBIS pricing. At the same time, $NBIS is maintaining ~0.6s latency, roughly a 93% cache hit rate and a displayed 100.0% uptime on DeepSeek. Maintaining that combination while absorbing this much traffic is an achivement most will fail to comprehent. The volume charts I'm tracking put $NBIS Token Factory on pace for ~700 billion tokens today. Most of the incremental volume looks like DeepSeek token share shifting over from other providers. They took on this sudden surge while maintaining uptime and throughput all in a matter of hours which is more than worth my glaze here. And the growth is not confined to the discounted model. I'm seeing nearly doubled token volumes and roughly doubled serving capacity on other popular open weights without a single $NBIS discount. MiniMax M3.1 Flash, Kimi K3 and Kimi K3.1 Pro, which I'd currently call the most powerful open weight LLMs in the market, all at regular market pricing (No discounts), with strong throughput across the broader lineup which is also suddenly multiplying the traffic $NBIS was receiving to these models prior to this week. As someone who works in inference, this is the first time I've been genuinely impressed by $NBIS Token Factory and if anyone ever wants to point to an inflection point for Token Factory I believe it will be the last 2 weeks. A discount can attract traffic. Scaling capacity and serving that traffic reliably while staying at the top of the chain for quality (Token Throughput in this case) is what makes this worth paying attention to.
1
1
22
1,136
$NBIS Sovereign AI is here to stay and Nebius can play a crucial role in this with token factory. $PANW is paying $4.5B in stock for Applied Compute, an 18-month-old startup that post-trains open-weight models on a company's own data and runs them in production. Palo Alto's stated view is that enterprise AI goes hybrid, with a large share running on open models tuned to each company's data, cost and control needs. The same team will also train Palo Alto's in-house models. Incumbent software can't rent its edge from the frontier labs forever. Every SaaS company that follows needs researchers first, then a place to do the work: GPUs for post-training, RL tooling, and dedicated inference for the finished model. Nebius bought Eigen AI this year for $643M to build that layer into Token Factory. Eigen does model-, kernel- and system-level optimization across post-training and inference. Post-training and serving an owned model is recurring compute spend, and it will likely stay where the model was trained. More than just a neocloud.
More of our customers want models adapted to their own data, tasks and workflows. Our new guide walks through four ways to adapt open-weight models: supervised fine-tuning, preference tuning, RL and distillation 👇
2
11
1,103
Kraken Robotics offers a comprehensive suite of technologies for Uncrewed Underwater Vehicles (UUVs), bringing together advanced sensing, navigation, communications, and subsea power solutions to support a wide range of mission requirements. From high-resolution seabed mapping and obstacle avoidance to precise navigation, underwater communications, and extended mission endurance, Kraken’s combined technology portfolio can be tailored to meet the operational needs of different UUV platforms. Our UUV technology offering includes Kraken Synthetic Aperture Sonar (SAS), Solstice Multi-Aperture Sonar (MAS), Vigilant Forward Looking Sonar, high-resolution optical imaging cameras, BlueComm Optical Communications, AvTrak 6 Acoustic Communications, SPRINT-Nav, and SeaPower subsea batteries. Explore our interactive UUV model to discover how these technologies can be integrated to support your mission requirements: ow.ly/r7r55107MZn #KrakenRobotics #UUV #UncrewedSystems #UnderwaterTechnology #MarineRobotics #SubseaTechnology
4
40
7
291
10,568
MinuteMoola retweeted
Go Team Go 🚀 @nebiustf @nebiusai
14
16
4
242
13,666
$NVDA $NBIS $CRWV $IREN Nvidia cleared a lot of fears: neoclouds and open models are a core part of their plan. Straight from tonight’s call: Interim CEO Colette Kress: “Neoclouds remain an increasingly important part of the global AI infrastructure ecosystem.” The segment that includes neoclouds did $47B, nearly 3x YoY. What drove the growth from last quarter? “Vera Rubin shipments to top leading neocloud customers.” Neoclouds are getting Nvidia’s newest racks first. Utilization: “NVIDIA Compute remains fully utilized across every hyperscaler and neocloud we serve.” Older Hopper GPUs have been re-rented at up to 2x their original price, and B300 rental rates are up 60%+. GPU rentals are getting more expensive. Open models are the next demand engine. Weekly token volume on OpenRouter is up 35x YoY. Nvidia: “A rapidly expanding ecosystem of inference providers is bringing these models to companies everywhere.” With Hugging Face under Nvidia's umbrella, they are “committed to fostering a vibrant open model ecosystem so every country, company, and industry can build its own intelligence.” Somebody has to host all that compute, and that’s exactly how clouds like $NBIS grab more margin with token factories serving open models. Sovereign AI revenue more than tripled YoY. Kress credited neoclouds with “supporting many of the different countries around the world, particularly on the sovereign perspective.”, recall that Nvidia laid out a blueprint for sovereign AI with $PLTR. The same Palantir that chose Nebius as their sovereign AI infra partner. Nvidia says it’ll be “disciplined and selective” with future financing. I see this as a positive for $NBIS, you've seen how fond Nvidia is of Nebius as they've shouted Nebius out at practically every event this year. More concentration of support to the top tier players. Bullish signals across the board for the competent clouds.
2
7
756
Give this man his company back $NVDA
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen
1
8
229
MinuteMoola retweeted
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen
61,013
121,716
24,489
1,174,061
58,636,721
MinuteMoola retweeted
If you believe that fundamental truths about reality and the universe exist, one might argue that through our proud history of scientific discovery it is becoming harder and harder to discover new fundamental truths. I would agree with that. The low hanging fruits in AI powered discovery will most likely evolve around bridging scientific disciplines, rather then creating net new ones. It’s been a long time ago since one scientist could be at the frontier in multiple disciplines and AI will have/has far less difficulty with consolidating all this scientific data, understanding it and making new cross disciplinary connections. A few notes: - science is mostly not practised as one shot discovery but rather being incrementally less wrong each experiment. More compute is more hypothesises and more hypothesises compound into being less and less wrong. Hence compute is the name of the game! - we might need bigger context windows to hold enough context in memory to start bridging those scientific disciplines - todays human “body of collective knowledge” is more like a labyrint then a holistic map. It’s just too much information for the human brain to keep making cross disciplinairy connections, heck I already have this with AI discipline alone. Through luck or intuition we humans might find them once in awhile, but AI can probably brute force its way into it - not every scientific discipline has the same amounts of data, levels of verifiability, feedback speed, simulation complexity or physical constraints so its not like we will have these advances across disciplines all in the same timeframe - AI research, mathematics, biology, chemistry, software engineering, finance and medical diagnostics are to benefit first. (Not investment advice 😂) - note that scientific breakthroughs in one discipline can be a catalyst for the other. Fe. If we have breakthroughs in material science we might be able to create better chips which could create advances in AI research which could allow for breakthroughs in other disciplines. - its kinda funny that we have built artificial intelligence by drawing inspiration from the neural nets in our brains, the architecture of human intelligence, yet AI greatest early scientific breakthroughs will occur in domains furthest removed from understanding human intelligence itself. We’re much further away from breakthroughs in disciplines like cognitive science or neuroscience. - this all strengths my believe that we humans are so much more intelligently designed, so much more advanced that fearing that AI is becoming conscious or super human is short-sighted. It will become super human in some domains, but fail drastically in others. We’re at the dawn of the scientific renaissance, and Im fired up folks 🙌
1
2
1
18
2,057
$NBIS Huge open-source growth in just 5 months. 0% -> 30% of token share. I believe open models will see a revenue ramp similar to the ones the frontier models had as open models get capable enough to handle more tasks. Why not own the intelligence and run models cheaper rather than giving away your “secret sauce” and paying a mark-up on the service? I still believe @nebiustf token factory will be the best positioned to capitalize on the open model usage ramp up. The developer community loves Nebius, their acquisitions help them squeeze the most value out of every GPU, and they own the full stack from infra to serving layer. The inference market has a longgg way to go and open models will be majority of the usage.
Replying to @pwendell
3/ Rise of open source: OSS models are quickly increasing in token share amongst enterprises, growing from nearly zero to 30% in the 5 month period. Anecdotally, we see those models used mostly to reduce costs on high volume workloads (large batch workloads, coding, etc).
1
10
794
MinuteMoola retweeted
Too many people treat AI as a zero-sum race. Open-source is winning by doing the opposite: collaboration over competition! This is why we’re excited to release today the first part of our partnership with @OpenRouter to make open models easier to use for all! Hugging Face for storage and collaboration. OpenRouter for inference. Connected so you can: 🤗 Run OpenRouter inference directly on Hugging Face 🔓 Access Open model weights directly from OpenRouter 🤖 Store your OpenRouter agent traces on HF to inspect them, train better models or share them with the world! More collaboration. Fewer zero-sum games. More power to AI builders. Let’s build! 💛💛💛
93
137
21
1,170
45,225
MinuteMoola retweeted
LIFT-OFF for the 100th time!
78
393
63
4,281
168,266
We started shipping the Nvidia H100 in 2022. A year ago it rented for $1.96/hour. Today: $3.10. +58%, on a chip that was supposed to be obsolete by now. So much for GPU depreciation.
323
1,010
207
9,747
1,079,097
MinuteMoola retweeted
The standwithjensen.com/#letter is currently overwhelmed with requests to sign it. Need to upgrade those servers!
279
517
149
3,636
702,023
MinuteMoola retweeted
Days before Christmas 2023, I emailed Jensen after an investor pulled out last minute. We didn’t have cash for payroll, and I was talking with bankruptcy lawyers. Jensen called me at 6am. Later that day, I was having dinner with his team to bridge the company. Four months later, we rang the NASDAQ bell for our IPO.
In June @NaderLikeLadder emailed Jensen about a crazy idea we had over dinner to make Local AI 10x better on NVIDIA hardware. Jensen replied in 7 minutes. He saw this as important work, and approved a war room for us to work on this at NVIDIA HQ. For a month, NVIDIA HQ was the @exolabs satellite office. We were overwhelmed by the support we got. We were back to back the first week with smart NVIDIANs coming in to our war room to help us. Creators of CUDA, Dynamo, people who built NVIDIA. Long story short with NVIDIA we managed to improve their Local AI playbooks by 9.6x (almost the 10x goal we set out with). We’ll be publishing a whitepaper on this work. We were just a small startup with a crazy idea. Jensen took a bet on us. Jensen cares deeply about Open Source AI and Local AI. I don't believe Jensen is replaceable, he's a 1 of 1 human, but I hope NVIDIA will continue to aggressively support Open Source AI and Local AI, and continue to take bets on small startups with crazy ideas. Thank you Jensen
120
702
89
12,267
2,056,774
$NBIS Q3 thoughts: Weak quarter but strong outlook. Headline revenue was $981M but ~$325M of that was a one-time, point-in-time revenue recognition from asset light. Core was ~$656M, up only 13% QoQ. The $6.4B ARR is just September x12 with that $325M asset light payment sitting inside it, so the real recurring run rate is more like $2.5-3B, basically where it was in June. No surprise they're killing the ARR metric, arguably a bullish sign of them anticipating more material asset light lumpy revenue that distorts the recurring figures? Flat recurring revenue in a quarter where customers committed $12B+ of contracts at >$25M/MW means the execution wasn't there. Reasoning provided was that colo partners slipped (Big one was that DataOne wound down the temporary power at Vineland before the Bloom fuel cells were ready, so the rest of Microsoft moves to Q2'27) while Nebius' own sites are on time or early. Most of the 2027 GW is on their own sites, and they say they've already secured all the power and components for it, which matters a lot with memory and power this tight so that's a great sign for next year. The deal terms are the best part of this report. Prepayments on Q3 deals covered more than 100% of the capex, including the data center and networking, up from 50-60% in Q2 and still rising in Q4. Prepaid share of contract value went up ~20 percentage points in one quarter (As crazy as it sounds I suspect this to increase again in this environment). The main risk outside of AI demand going away is financing the build out, so it's actually very encouraging to see how much of the buildout customers are willing to back. Payback on new deals dropped to 1 year 7 months covering both capex and opex. So customers are funding the build and Nebius gets its money back in under 2 years on 1-5 year contracts. That's barely any capital at risk on new capacity. Pricing is moving the right way too: GPU-hour prices up 20%+ QoQ, reserve auctions clearing 15%+ above reserve, spot at times beating on-demand, short-term clusters at 2x+ the reserved rate. Compute is scarce. The asset light piece looks very high margin. They book a chunk up front when they lock in capacity, supply chain and architecture for a partner, then earn over time as that capacity gets sold. Back out the core business at a ~35-40% margin and that $325M probably carried $200M+ of EBITDA, something like 60-70% margin with zero Nebius capex. It's also the reason operating income flipped to +$50M. Without it you're looking at roughly -$150-185M. GAAP EPS was -$0.13 (net loss $37.5M incl a $49.5M charge for getting convert holders to convert early) and adjusted net income was +$91M. This is surprisingly close to profitability this early. One thing that I wish analysts asked: Nebius lent out $348.6M to someone ("advances under notes receivable" in the cash flow statement, first time this line has ever shown up). Awfully close to the asset light revenue booked. Could just be a complete coincidence but would have liked some light shed on this. Margins around 40% EBITDA guide for 2026 implies Q4 around 35-36% on $1.04-1.24B of revenue. Then pre-tax breakeven in Q4'27 while spending $40-45B on capex. A new MW at ~$25M a year with ~55-60% EBITDA (my estimate) is ~$14-15M of EBITDA against ~$6-7M of depreciation on 5 year GPUs, and when customers prepay the capex a lot of that capacity carries barely any interest. That's how they're getting to breakeven this fast in the middle of a buildout. The 2027 guide is $12-13B. Capacity live by YE26 plus the Microsoft and Meta contracts gets close to $10B on its own, so they're giving the entire new GW only $2-3B of revenue. Super sandbagged, and the CFO literally called their timing assumptions "very, very conservative". ~1.7-2GW by YE27 is roughly a $20-25B exit run rate on their timing and ~$35B if the GW is online on schedule. That's the number you'd want to see next to $40-45B of capex next year, which is going to look scary without ARR next to it. Doesn't even include asset light btw. RPO was only +3% to $38.7B but most of that $12B+ was signed after Sept 30 so it shows up next quarter. The 5 year deals were all expansions by existing customers, and AI cloud deals hit multi-billion size in early Q4. Funding: 75-85% of 2027 capex comes from prepayments plus asset backed loans on IG contracts. That leaves ~$6-11B plus the $3.5B Citi bridge due Oct '27, so ~$10-15B to find. My bet is 1-2 more converts, ABF stretched to AI lab contracts, a new ATM, and leaning harder on long term prepaid deals which is their cheapest lever. They also sit on ~$3-7B of stakes my model doesn't touch: ClickHouse ($1.6B on the books, ~$3.9B at the $15B round if they still own ~26%), Avride ~$2.5B, Toloka ~$0.5B with revenue doubling QoQ, TripleTen. Token Factory is ramping hard. Approaching a multi-hundred-million run rate less than a year after launch, contract value up 17x QoQ, day zero launches for Kimi K3, GLM 5.3, Reflection Beam and Meta's Muse Spark Open, top tier on Artificial Analysis. Prosus runs it across apps with ~1B users. Customers like Applied Compute and Cursive start with post-training and move their inference onto it without leaving Nebius, and some go from Token Factory into dedicated clusters. That's the platform flywheel the hyperscaler thesis needs. Plus the first enterprise platforms signed (Databricks, Datadog, Snowflake, Palantir), which means distribution through thousands of their customers. Longer term something on my mind is Jensen being out. He was the biggest force behind open models in the US, and open models are what turn Token Factory into a massive business in the ultra bull case. I believe the market may also be pricing in some fear of whether $NVDA will still backstop neoclouds moving forward. I believe Nvidia continues working on both the open model ecosystem and backing up clouds because it is core to Nvidia's strategy, they want customer breadth in an environment where their biggest customers are beginning to develop their own chips. Execution premium deserved to be shaved a little this quarter. But if they close the 2027 gap and get the GW online, the margin profile from late '27 starts carrying the build. 2027 will be a big test financing wise, and 75-85% of it seems to have a clear path through prepayments and ABF, so I’m optimistic on the outlook. A 5GW ramp up for a company of this size is not going to be hiccup free. Investors will need to find patience and conviction, AI clouds are long-term plays. Bullish $NBIS
7
742