Kenley Lamaute

The pace of capital flowing into AI feels familiar. These two charts from Apollo Academy highlight just how quickly expectations (and valuations) have risen. History tells us that periods of exuberance are almost always followed by a correction. At Buoyant, we believe both sides of this story can be true: ✅ Yes, the current surge of investment is likely ahead of itself. ✅ And yes, the underlying transformation is as inevitable as the Internet’s rise 25 years ago. We’ve seen this same pattern in climate tech: early hype, painful correction, and then enduring progress as the world reorganizes around a fundamental shift. The companies that endure are the ones that solve real problems with staying power. Our responsibility as investors at Buoyant Ventures is not to chase the froth, but to stay grounded in fundamentals — backing entrepreneurs who are building durable businesses that will still matter long after the cycle cools. AI’s trajectory may be uneven, but like climate tech, its long-term impact is unstoppable. 🤛 https://lnkd.in/gga2igQJ https://lnkd.in/gfmRy-VG

30

3 Comments

What happens when boomers exit the market? | A clip from my Substack post, EP028: The Architect Who Became the Grey Rabbit of Finance. Watch it here: https://lnkd.in/eiT_7Dse

1

Who captures the 99% value? My take. The single biggest misconception in finance right now is that the AI value wave stops at the cloud. It doesn't. It flows 100% further upstream than any cycle in history. The first technology in history to collapse labor costs across entire sectors at once is now here. Railroads replaced physical transport. Electrification replaced mechanical labor. The internet replaced information labor. This new wave replaces knowledge labor everywhere. When a spread this large appears, it cannot sit downstream. It forces value higher upstream than any prior cycle because the actual pressure lands on the earliest physical constraints, not the applications or even the midstream. It exposes inefficiencies in financial markets, misclassification of critical assets and the undercapitalized weak points everyone ignores. Past eras followed a clear pattern. Rail operators lost margin and value shifted to steel and coal suppliers who controlled inputs that expanded slowly. Electrification boosted factories, yet utilities and transmission owners held the durable economics because they governed access. The internet created endless applications, but long term value concentrated in data centers, fiber and towers. Each cycle moved the center of gravity further upstream. This cycle pushes it even further. A 150$ MRI becoming 15 cents wipes out the downstream layer entirely. Hospitals cannot hold that surplus. Reimbursement, policy and competition absorb it. Downstream sectors rarely retain margin once supply becomes abundant. This wave makes supply abundant everywhere. Because the downstream is erased that quickly, the true bottlenecks sit above data centers or fabrication. Scaling depends on electricity, cooling, transmission, water, high purity materials, land with power rights, mineral chains, refining capacity and early industrial footprints. These determine whether the entire stack can exist. They take years to build and cannot be replaced with software. That is why the center of value moves beyond the usual upstream nodes into the places even those nodes rely on. As Scott North noted, "China did not win through geology. It won by building the entire chain ". Nations lose supply chains in the midstream, not at the mine. This wave will follow the same logic. Winners control ecosystems, not isolated industries. That is why Pre Yield Assets matter. They sit before traditional yield in early corridors, basins, mineral systems, refining potential and industrial footprints. They are misclassified and undervalued because they fall outside standard mandates, yet they are the first points of scarcity in a system that will stretch every physical input hold value before yield exists and retain value after downstream margins compress. They are the beginnings of integrated industrial ecosystems. This cycle drives value further upstream than anything before it and is where SKGP places its attention. #CIO

8

Everyone's ready to declare that LLMs are a dead end toward AGI. Yann LeCun just raised +$1B for AMI - Advanced Machine Intelligence to bet that world models and not LLMs are the path to AGI. Even Sam Altman admits scaling alone won't get us to AGI, even though he claimed he knew how OpenAI could get there just 14 months ago. On one hand, it feels like an existential question for the entire industry as it spends billions of dollars on infrastructure at crazy valuations. On the other hand, what if the real question here is: So what? The issue isn't whether the models are perfect. They're not. Or whether they can achieve AGI. Maybe they can. Maybe (more likely) they can't. The issue for most average people just building their business day to day is whether smart engineering around LLMs can make them dramatically more useful. The early evidence says: yes For instance, Axiom just produced verified mathematical proofs that humans couldn't crack. Another found a way to make AI agents coordinate nearly 5x faster. Neither required a better model. Both required a better system around the model. This is the way I propose thinking about it: The Layer. Yes, the model is the foundation, and those advances will continue for a long time, even if they never achieve AGI. Meanwhile, the next phase of progress may come less from replacing the models themselves than from engineering the layers that make them more reliable, coordinated, and economically useful. So, it's no longer simply about models. It is about systems. Capability may increasingly emerge not from the model alone, but from the layers engineered around it. #OrchestrationEconomics #Discontinuity #AgenticEra

16

1 Comment

⚡ This 50-Microsecond Mistake is Costing HFT Firms $10M+ Per Year (And Most Don't Even Know It) In High Frequency Trading, every microsecond counts. While others debate market strategies, we're optimizing at the nanosecond level. Here's why message queue architecture is the unsung hero of HFT: The Queue Dilemma: SPSC (Single Producer, Single Consumer) - Perfect for dedicated market data streams MPMC (Multiple Producer, Multiple Consumer) - Essential for multi-strategy environments Lock-free implementations - Because even a single mutex can cost you the trade Real Impact: A poorly designed queue can add 10-50 microseconds of latency. In HFT, that's the difference between profit and loss. We're talking about: Cache-aligned memory structures Ring buffer optimizations Zero-copy message passing CPU core pinning for consistent performance The Technical Reality: While traditional systems use standard queues, HFT demands custom implementations that avoid system calls entirely. We operate in user-space, where spinlocks replace sleep locks, and every memory allocation is pre-planned. The NSE processes millions of packets per second. Your queue architecture determines whether you're reading market data in real-time or living in the past. 💡 In HFT, we don't just write code - we engineer physics-defying software where network latency meets algorithmic precision. #HighFrequencyTrading #SystemsEngineering #LowLatency #CPlusPlus #FinTech #AlgorithmicTrading #HFT #QuantTrading #TradingTechnology #NSE #BSE #WallStreet #LondonFinance #SingaporeFinance #MumbaiFinance #DelhiNCR #Bangalore #Hyderabad #Chennai #Pune #QuantDeveloper #TradingJobs #FinanceJobs #TechJobs #SoftwareDeveloper #SystemsProgramming #RealTimeSystem #NetworkProgramming #Linux #TradingFirms #HedgeFunds #PropTrading #MarketMaking #DerivativesTrading #FuturesTrading #OptionsTrading #CryptoTrading #India #USA #UK #Singapore #HongKong #Dubai #Canada #Australia #Germany #Japan #TradingCommunity #QuantFinance #FinancialEngineering #ComputerScience #Engineering #IIT #NIT #CareerGrowth #SkillDevelopment #OnlineEducation #ProfessionalDevelopment 🎯 HFT C++ Fresh Batch Started Ready to dive deep into the world of microsecond-level optimizations? Join our comprehensive HFT development program covering everything from Linux system tuning to NSE protocol implementation. 📩 DM for syllabus, details & cost 📧 Email: pranaygaurav4555@gmail.com

8

🚀 Nebius: Nvidia‑backed AI cloud set to 5x revenue in 2026 — as data‑center demand breaks above supply 💡 Goldman Sachs sees U.S. data‑center demand outstripping supply by ~9 GW in 2026, a structural gap fueling a “demand‑surplus” environment. Nebius sits right in the middle of it — and it’s already sold out existing capacity. 💥 Consensus: +521% YoY revenue growth in 2026 (~$3.45B), building on triple‑digit expansion in 2025 — that’s operating leverage, narrative momentum, and scale economics converging fast. ⚡ Capacity expansion: Nebius plans to lift usable power capacity 4–5x by end‑2026 (to 0.8–1 GW), with >2 GW already secured and nine new AI data centers underway — a path toward 3 GW+ over the medium term. 🏗️ CapEx blitz: ~$16–20B spending this year alone, funded in part by Nvidia’s $2B investment for an 8% stake. The move reduces financing risk, locks in GPU supply, and effectively crowns Nebius one of Nvidia’s “core neocloud” partners (think the next CoreWeave). 🎯 Financial roadmap: Management targets $7–9B annualized run‑rate by end‑2026, implying sustained utilization and a clear scaling glide path. 🤝 Strategic edge: Early access to Nvidia’s Rubin, Vera CPUs, and BlueField networking — translating to faster deployment cycles, cost efficiency, and a moat built on hardware timing. 💸 Investor takeaway: • AI compute scarcity = pricing power. • GPU allocation = guaranteed growth lane. • Nvidia validation = re‑rating catalyst. • 5+ GW roadmap = durable optionality. 📈 Nebius isn’t just scaling data centers — it’s scaling valuation narrative. #AIInfrastructure #DataCenters #Nvidia #Nebius #Investing #TechStocks #EquityResearch #ArtificialIntelligence #FinancialMarkets #CloudComputing #InvestorFocus

2

Hook — Apple’s AI posture is discipline, not disinterest. While the market crowns infrastructure and platform leaders, Apple is quietly conserving capital and integrating AI incrementally into hardware, OS and services. Macro trends - AI investment has concentrated in compute, data-center platforms and training pipelines (NVIDIA, MSFT, GOOGL, AMZN, META). - Initial value capture favors infrastructure; over time value tends to migrate to applications and user-facing ecosystems. - Capital intensity and regulatory uncertainty elevate execution risk for early, large-scale AI bets. Key factors driving Apple’s strategy - Balance sheet optionality: >$140bn liquid assets and ~100bn annual FCF enable optionality without premature deployment. - Historical playbook: Apple typically waits for product-market fit, then deploys superior integrated hardware+software solutions. - Margin and UX focus: incremental AI that enhances device utility preserves gross margins and brand trust (privacy-first positioning). - Competitive landscape: peers are defining technical standards; Apple can adopt proven models, avoiding early capital write-offs. Risks - Opportunity cost: lost share in newly defined AI-adjacent markets if competitors lock users to platforms. - Competitive lock-in: cloud and platform incumbents may entrench developer ecosystems and data moats. - Regulatory/legal exposure: AI raises privacy, IP and liability risks; missteps could damage Apple’s premium brand. - Monetization uncertainty: consumer willingness to pay for AI-enabled features remains unproven. Actionable insights (for investors and strategists) - Investors: Favor staged exposure—position size increases on clear signs of Apple converting AI features into higher ARPU or services margin expansion. - Monitor indicators: new OS-level AI APIs, partnerships with cloud/model providers, services revenue mix, device attach rates, and incremental gross margins. - Strategists/C-level: preserve capital flexibility; prioritize integrated experiences that leverage hardware differentiation and privacy as a moat. - Traders: price in patience—near-term underperformance is plausible; alpha arrives when monetization vectors clarify. Takeaway / forecast Apple’s restraint is a strategic option: expect measured, high-ROI deployments once cost curves, regulation and user willingness to pay converge. Short-term underperformance versus AI “front-runners” is likely; medium-term upside hinges on Apple translating incremental AI into differentiated services and margin accretion. What do you think? Share your perspective on Apple’s timing and the winners in the AI value migration. — Viktor Kopylov, PhD, CFA. More author's insights: t.me/si14Kopylov

3

I built Quasar, an AI-driven trading system. At its core is ARC (Adaptive Regime Control), which continuously interprets market conditions and adjusts behavior in real time instead of relying on a fixed strategy. The system uses an LSTM model to learn temporal patterns across structural data like order flow, CVD, liquidations, open interest, funding, and trade aggression. This allows it to understand how the market evolves, not just react to price. Risk management is fully integrated. Exposure adjusts dynamically based on both market conditions and system performance. You can monitor trades and system behavior live. GitHub: https://lnkd.in/ehSuW8wy Monitor trades: https://lnkd.in/eD6-YzWP Landing page: https://lnkd.in/eeUTjptT @Kraken @lablab.ai @Surge.xyz #Quasar #AI #MachineLearning #LSTM #QuantTrading #Crypto #DeFi #Automation #BuildInPublic

21

4 Comments

Everyone is excited that Anthropic just partnered with SpaceX and doubled Claude Code limits for paid users. 220,000+ GPUs. 300 MW of compute. API limits up. Peak‑hour throttling down. Infra is compounding like crazy. But here’s the uncomfortable contrast I keep seeing in B2B: Founders are talking about “agentic workflows” and “AI copilots for GTM”… while their CRM still has: - Duplicated accounts - Garbage contact data - No clear ICP segmentation - No agreement on what a qualified opportunity even is 𝐎𝐧 𝐨𝐧𝐞 𝐬𝐢𝐝𝐞: Anthropic, OpenAI, ElevenLabs, etc. racing to ship bigger, faster, cheaper AI. 𝐎𝐧 𝐭𝐡𝐞 𝐨𝐭𝐡𝐞𝐫 𝐬𝐢𝐝𝐞: Teams who can’t reliably answer “Which 50 accounts actually matter this quarter?” That gap is where most AI “strategy” dies. More compute doesn’t fix: - Broken data foundations - Messy handoffs between marketing, sales, and CS - No shared definitions of pipeline, stages, or attribution As a RevOps person, this is how I look at it: 1. 𝐈𝐧𝐟𝐫𝐚 𝐢𝐬 𝐧𝐨 𝐥𝐨𝐧𝐠𝐞𝐫 𝐲𝐨𝐮𝐫 𝐛𝐨𝐭𝐭𝐥𝐞𝐧𝐞𝐜𝐤. Claude’s limits doubling and API caps exploding means the ceiling is way higher than your current use cases. 2. 𝐘𝐨𝐮𝐫 𝐛𝐨𝐭𝐭𝐥𝐞𝐧𝐞𝐜𝐤 𝐢𝐬 𝐜𝐥𝐚𝐫𝐢𝐭𝐲. If you can’t describe your GTM system cleanly, you can’t automate it cleanly. 3. 𝐀𝐠𝐞𝐧𝐭𝐬 𝐚𝐫𝐞 𝐨𝐧𝐥𝐲 𝐚𝐬 𝐬𝐦𝐚𝐫𝐭 𝐚𝐬 𝐲𝐨𝐮𝐫 𝐝𝐚𝐭𝐚 𝐦𝐨𝐝𝐞𝐥. Throwing an “AI agent” on top of chaos just gives you faster, more confident chaos. 𝐈𝐟 𝐲𝐨𝐮 𝐰𝐚𝐧𝐭 𝐭𝐨 𝐛𝐞 𝐬𝐞𝐫𝐢𝐨𝐮𝐬 𝐚𝐛𝐨𝐮𝐭 𝐀𝐈 𝐢𝐧 𝐆𝐓𝐌/𝐑𝐞𝐯𝐎𝐩𝐬, 𝐬𝐭𝐚𝐫𝐭 𝐡𝐞𝐫𝐞: - Clean your core objects (accounts, contacts, opportunities). - Define what “good” looks like for your ICP and deals. - Map 3–5 repeatable workflows before you automate anything. - Only then bring in agents and workflows to scale what’s already working. AI infra is scaling. Most GTM infra is still pretending. Don’t chase the headline. Fix the system so you can actually take advantage of what’s being built. Question: What’s the one GTM process you’d actually trust an AI agent to own end‑to‑end in your current stack? #RevOps #GTM #AI #Claude #Anthropic #B2B

5

2 Comments

Sam Altman Reveals OpenAI’s Trillion-Dollar AGI Ambition OpenAI CEO Sam Altman recently gave an in-depth interview on the A16Z podcast, revealing the grand strategy behind the company’s aggressive investment in AI computing centers, its attempts at AI hardware, and its ultimate goal: the realization of Artificial General Intelligence (AGI). Massive Investment to Build Computing Centers: An Unprecedented Scale OpenAI's recent investment in computing power is unprecedented. The company is betting heavily, collaborating with NVIDIA, AMD, and Broadcom to secure a staggering 26 GW of capacity—equivalent to the power consumption of two New York Cities. More notably, a massive HBM (High Bandwidth Memory chips) order was signed with Samsung and SK Hynix, reportedly double the current global production capacity. This aggressive capital strategy, while fueling "AI bubble" concerns, is rooted in Altman's confidence in a clear “research roadmap” and “commercialization potential.” AGI: The Ultimate Mission and AI Companion Altman defined OpenAI’s ultimate mission: achieving AGI (AI stronger than humans in every aspect). The corresponding vision is for everyone to have a personalized, problem-solving "AI companion" that deeply understands them. To realize this, OpenAI has evolved from a pure non-profit lab into a profitable commercial entity, integrating consumer applications (like ChatGPT), massive AI compute centers, and AI hardware. The commercial value of this "personal AI assistant" is considered to be far greater than prior tech models (search engines, chat platforms), making it the prime "battleground" for all tech giants. World Models and Technical Supremacy Altman asserted that the development of AI technology is far from encountering a ceiling, promising continuous releases of more powerful models. Crucially, he confirmed that Sora, their video generation model, is fundamentally a World Model. He stressed that building a good World Model—an AI that can understand and simulate the workings of the real world—is the necessary path toward AGI. This confirms World Models as the next critical frontier, a focus also pursued by Chinese firms like SenseTime and the Beijing Humanoid Robot Innovation Center. Investment Limit and Economic Value Regarding the financial scale, Altman stated that while an ultimate limit exists—as a "proportion of GDP"—they are currently nowhere near it. Citing industry estimates that annual AI capital expenditure could reach the $5 trillion mark (10 times the current level), he argues that the potential economic value of correctly anticipating AI's capabilities is immense. This strong conviction in the technology’s future and vast commercial viability compels OpenAI to make these aggressive investments, often requiring the company to "pull the entire industry along" by announcing major partnerships.

1

1 Comment

Every result. Every shot. Every line of code. Open. 📂 GitHub: https://lnkd.in/egarHJMu → Raw shot-level data (16,384 shots × 3 conditions × 2 backends) → Complete analysis scripts → All Job IDs for independent verification → Circuit definitions & parameters What's next? → Multi-backend replication (ibm_brisbane, ibm_kyiv) → Continuous measurement rate sweeps → Mutual information & entanglement entropy analysis → Scaling to 100+ qubit chains → Error-mitigated quantitative benchmarks The horizon is just the beginning. Read the full paper: DOI 10.21203/rs.3.rs-8618767/v1 #OpenScience #OpenData #GitHub #QuantumComputing #Reproducibility #FutureOfPhysics #QMCLab #QuantumResearch #OpenAccess #AnalogGravity

1

The use of AI is expanding. Some startup use of Nebius cloud: h/t Ehraz Ahmed Startup testimonials built on Nebius AI Cloud. 🧠 SGLang “Nebius helped us double our inference throughput and reduce latency for DeepSeek R1 — all on a single node.” Category: LLM inference optimization Impact: 2× throughput, low-latency single-node deployment. ⚙️ vLLM (Open-source inference framework) “Testing on Nebius infrastructure enabled high-performance, low-cost model serving in production.” Category: Scalable LLM inference Use case: Real-world inference stress testing 🧬 Converge Bio “We’re using LLMs on Nebius to analyze single-cell RNA sequencing, transforming disease mechanism research.” Category: Biotech, genomics Impact: Pioneering life sciences AI with Nebius compute 🤖 Positronic Robotics “Nebius helps us train robot control models that outperform human cleaners.” Category: Robotics Use case: Training robot control systems at scale 🎨 Recraft (Backed by Khosla Ventures, Nat Friedman) “We trained our 20B parameter design-first generative model from scratch on Nebius.” Category: GenAI for designers Investor pedigree: Khosla Ventures + GitHub’s Nat Friedman Infra: End-to-end training on Nebius GPU clusters 🧠 Chatfuel “Using LLaMA-405B cascade on Nebius AI Studio, we achieved dramatically faster, higher-quality chatbot responses.” Category: Customer support automation Tech: LLM cascade + AI Studio SDK Result: Better response speed, quality, and infra efficiency 🎥 Dubformer “We run both training and model deployment for our multilingual dubbing AI on Nebius.” Category: AI voice localization Use case: End-to-end multilingual production workflow on Nebius 🧪 Simulacra AI “We’re using Nebius to build a large wavefunction model for quantum chemistry.” Category: Drug discovery Infra: Multi-billion token LLM trained on Nebius for molecule prediction 📞 Krisp “Nebius powers our real-time accent localization and voice conversion tech.” Category: AI voice transformation Impact: Removes accents from call center agent speech in real time 🧠 TheStage AI “We use Nebius to close the inference–cost efficiency gap with our neural network optimizer.” Category: LLM inference compression Use case: Cost-effective model scaling 🔬 Quantori “We’re using Nebius for a graph-convolutional molecular generation pipeline based on diffusion models.” Category: Cheminformatics Infra: Stable diffusion + custom GNNs trained on Nebius 🎯 Bottom Line Nebius is not theoretical, it's already powering real production workloads across: LLMs Biotech Quantum chemistry Robotics Customer support Design tools Language localization Nebius Group

4

Gemini 3's leak reveals exciting advancements in mathematics, screen comprehension, and simplified Q&A. Despite being outperformed by Sonnet on Swebench, Gemini 3 emerges victorious on Terminalbench. Notably, it boasts a lower context length compared to 2.5 Pro. #Gemini3 #Leak #Technology

Claude just built a C compiler. A real one -- parsing, IR, LLVM-style code generation. Compilers are no joke. Chris Lattner, who spent his life so far building a few compilers, broke down what this tells us about where things are going. A few quotes that resonated with me: "As implementation becomes cheaper, the role of engineers shifts upward. The scarce skills become choosing the right abstractions, defining meaningful problems, and designing systems that humans and AI can evolve together." "AI amplifies both good and bad structure, so we can expect to see poorly managed code scale into incomprehensible nightmares." That second one is the one nobody wants to talk about. We are all excited about shipping faster. But a sloppy architecture that used to slow you down now ships to production in a weekend. The mess scales too. There is also something subtle in the compiler itself. CCC passes tests brilliantly -- but it optimizes toward measurable criteria. As Lattner points out, "there is no test suite for an idea that does not exist." The moment a problem is open-ended, AI has nothing to optimize against. Or as he puts it: "Writing code has never been the goal. Building meaningful software is." Knowing what to build, having the taste to recognize good structure, and the discipline to keep it clean when adding more costs almost nothing -- that is where it gets hard. And that is where software engineering is here to stay. Full article well worth the read: https://lnkd.in/dN4M9gJX

14

4 Comments