• Ascend Analytics power-market forecaster Brent Nelson says large-load queues are poor proxies for actual construction: ERCOT's queue is about 400 GW versus a roughly 90-GW system peak. He estimates plausible builds at only around one-tenth to one-fifth of raw queue totals, which are inflated by normal project duplication and speculative entrants; forecasts already discounted for plausibility should not be discounted again. TechCrunch
• Nelson expects political slowdowns to delay data-center development by roughly one to two years, affecting 2027 more than 2032, while access to power is a separate bottleneck. Long-lead gas projects also face political and permitting risk. TechCrunch Solar and storage are poorly matched to long overnight winter reliability needs, helping make gas the default despite constrained generation equipment, pipeline challenges, and fuel-delivery risks during winter storms. Nelson argues that a curtailable grid connection paired with behind-the-meter batteries could be a comparable, potentially cleaner, cheaper, and faster alternative to gas, while noting this is not a consensus view. TechCrunch Paying for grid upgrades is necessary but may not secure community support, given concerns about affordability, water, noise, emissions, and AI more broadly. TechCrunch
• AI-infrastructure demand depends on timing and customer durability: hyperscalers reportedly struggle to find capacity for 2027, but around 2030 many data centers may compete for tenants, while some AI startups may fail or be acquired, changing who needs compute. TechCrunch Banks generally require a creditworthy tenant lease to finance a data center, though the adequacy of lender due diligence is an open question; the interview cites a Meta-leased data center that struggled to close without substantial concessions amid a tightening debt market. TechCrunch A strong neocloud anchor tenant can make a project more reliable but creates concentration risk, so investors should assess who will pay the rent rather than relying on capacity plans alone. TechCrunch

• Theranos’s pitch was that a single drop of blood could support central-lab testing in settings from homes to battlefields Y Combinator. Schultz says there were no tests validated when the Walgreens product launched, patient samples instead went to other labs or third-party equipment, and only seven tests were validated by the time he left Y Combinator. He also recounts that a syphilis test initially showed 65% sensitivity; after that data was deleted and the study repeated, sensitivity was 80%, while more than 20% of assumed-negative colleagues tested positive before a doctor signed off Y Combinator.
• Schultz describes an 11-year culture of fear in which employees who questioned the CEO were fired and sometimes sued; he says people who knew the device was inaccurate stayed silent Y Combinator. He says employees might have raised concerns with the board if they had trusted that they could do so safely and confidentially Y Combinator.
• Schultz recounts that Khosla’s investment process led the firm to pass, and says he heard similar accounts from Sequoia and Google Ventures; Tim Draper was an exception, making a $500,000 angel investment Y Combinator.

• One Password’s CEO warned that AI agents’ need for secrets is colliding with growing exposure from long-lived credentials scattered across devices, while the cost of finding and exploiting them is rapidly falling. Lightspeed Venture Partners The company said it had built a zero-knowledge framework for agent access, announced One Password for Agents, and saw “explosive adoption” on platforms where it was released. Lightspeed Venture Partners
• The interview reports that One Password protects over 200,000 businesses and recently crossed $500 million in ARR. Lightspeed Venture Partners Its CEO said Q3 net-new ARR growth was about 90%, with AI-oriented use cases driving growth rather than token resale. Lightspeed Venture Partners He attributed the business’s ability to fund product and distribution investment while remaining profitable to strong retention and a profitable core consumer business. Lightspeed Venture Partners
• For early-stage diligence, the CEO recommends testing funnel progression and real customer pull before scaling, then examining gross/net retention and expansion across customer segments; he cautioned that sales through founder, portfolio-company, or board networks can give a false signal. Lightspeed Venture Partners

• The interview presents Nuance Labs’ $50 million Series A as led by Lightspeed. Lightspeed Venture Partners Founders said the team had reached 27 after initially planning for about eight hires over six to eight months; they attributed the expanded scope to building custom inference and orchestration systems as well as the model and its personality and emotional-intelligence capabilities. They said the round would accelerate work on talent, data and compute. Lightspeed Venture Partners Fong Chong Ma described prior Apple work on Vision Pro digital personas, and Karen Yang cited audiovisual research experience at Adobe, Meta and Apple. Lightspeed Venture Partners
• Nuance Labs is building a unified foundation model for real-time understanding and generation across text, speech and video. Lightspeed Venture Partners The founders’ proposed differentiator is full-duplex operation—attending to a user while generating the system’s response—rather than a turn-taking pipeline; they argue that existing cascaded systems lose expressive cues when they are reduced to text and incur latency. Lightspeed Venture Partners
• The founders describe natural conversational dynamics as a data bottleneck: they say suitable data is not available at internet scale, and podcasts or interviews often differ from direct conversations because they are edited or audience-facing. They also say simulation is unsuitable for this problem, leaving data gathering as a core challenge. Lightspeed Venture Partners
• The founders said their immediate focus was releasing a research demo, with a consumer-facing interface as a path to enterprise use. At company launch, they reported inbound interest in customer support, user and market studies, and training and coaching; this was reported interest, not evidence of deployed customers. Lightspeed Venture Partners

• Loyal founder Celine Halioua was introduced as having raised over $200 million; she described Loyal as a Series C company pursuing the first FDA-approved lifespan-extension drug, beginning with dogs. At the time of the talk, she said its first drug had cleared efficacy and safety approvals, with manufacturing approval remaining. Stanford eCorner
• Loyal’s dogs-first thesis uses shorter dog lifespans and cash-pay owners to avoid a human-development bottleneck: proving lifespan extension in people could take decades, nearly matching a 20-year patent term. The company’s “save the dog, save the world” thesis is that dog-longevity work could establish regulatory, social and biological precedent for humans. Stanford eCorner
• Loyal abandoned its original gene-therapy approach about six months in after its team cited safety, cost-of-goods and technical risks, switching to another way of targeting the biology. Halioua described a portfolio of about five drugs at different stages; one had recently been dropped over a theoretical safety concern, illustrating the company’s effort to avoid dependence on any single drug. Stanford eCorner
• Halioua sees opportunities for AI in clinical development and operations, and cautions that some AI-for-bio work may focus on areas that become commoditized without enough understanding of drug-development bottlenecks. Loyal’s veterinary trial team has built “vibe-coded” interface skins over difficult existing study software to reduce user errors, which she cited as an example of unmet workflow needs. Stanford eCorner

• Ro reported growing from about $250M to more than $3B in run rate over four years while employing fewer people than in 2021; GLP-1s were its fastest-growing category. Its CEO credited a major GLP-1 tailwind but said credit-card data showed Ro signing up roughly 2–3 times as many patients as peers. Invest Like The Best
• The 2022 refounding followed a 9% growth year after a $500M raise at a $5B valuation; the CEO said the company had tried to do too much and hire too broadly. The reset included moving on from seven VPs and about 20% of staff, unifying brands under Ro, and clarifying the company culture; Ro later reported a 10x larger business with about 30% fewer people. The CEO also described prioritizing sequential execution and dropping long-term roadmaps in fast-changing markets. Invest Like The Best
• Ro said it routes every patient message through an LLM to analyze the issue; reported side effects become structured tasks for a healthcare provider with an SLA, and the company claimed 24/7, sub-30-minute response times. Invest Like The Best
• The CEO sees healthcare-AI opportunities in combining newly available symptom, lab, genetic, and wearable data with AI and new treatments; he pointed to variable GLP-1 responses as a potential personalization use case. He also cautioned that medicine is difficult for AI because much of its data is unstructured and outcomes such as heart disease take a long time to observe. Invest Like The Best

• a16z frames the B2A economy as an emerging distribution shift: as agents move from answering questions to completing tasks, they may choose products, subscriptions, and purchases for users. The business models enabled by this shift are mostly not yet visible, making this an early thesis rather than a mature market. @omooretweets
• Early signals include Shopify-reported 8x year-over-year growth in AI-driven store traffic in Q1 2026 and nearly 13x growth in orders from AI-powered searches; Human Security also saw agentic-browser traffic grow 8x from August to October, with Muse driving 60% of activity and OpenAI products 30%. @omooretweets The article says AI-referred consumer traffic has higher engagement and conversion and lower bounce rates than non-AI traffic, and estimates that 11% of U.S. shopper journeys visited AI before a transactional site. @omooretweets Coding agents offer a separate adoption precedent: as of September 2026, upwards of 40% of GitHub PRs started with an agent draft, versus 17% six months earlier. @omooretweets
• a16z identifies potential opportunity areas in agent models and harnesses; agent-enabled businesses such as self-running SMBs, AI-intermediated marketplaces or network products, and auto-optimizers; incumbents adapting for agent-led product evaluation and purchasing; and agent infrastructure. Payments, fraud, identity, and communications are already experiencing strain at relatively low agent volumes. @omooretweets

• The guest sees silicon—not power—as the main AI-capacity bottleneck, with fab capacity likely to remain a governing constraint for the next two to three years; he expects demand to become more price-sensitive as buyers route work to cheaper models. a16z
• He expects an economic correction and flagged weak neocloud balance sheets, but believes the underlying compute assets will remain useful. GPU longevity depends partly on ongoing scarcity: abundant newer, more efficient GPUs could make older hardware uneconomic. a16z
• He argues AI infrastructure should be designed as an integrated system, increasingly at rack or row scale; he sees patchwork systems as less efficient. Because AI data-center traffic is highly utilized and more predictable than conventional network traffic, he sees an opening for cheaper, purpose-built networking, although adoption requires coordination across the system and its buyers. a16z
• For startup exit decisions, he emphasizes the founding team's appetite to continue and whether the business could plausibly grow 10×, rather than merely double. He advises founders to maintain regular conversations with large companies and competitors for market intelligence and relationship-building, not necessarily to signal a sale. a16z

• Index Industries emerged from stealth with a $50 million seed round to build a platform for index assets. Its first product, USHP, is a token tracking broad U.S. home-price indices and backed by real-estate-related and liquid assets; an income-bearing variant, USHPI/USHP Income, is designed to pay holders yield generated by reserve assets. Lightspeed Venture Partners
• CEO and cofounder Josh Rosenthal previously led at Polychain Capital and joined the company after investing; cofounder Brian Hardeker brings residential real-estate acquisition and operating expertise through a firm described as having deployed about $2 billion in U.S. residential assets. Lightspeed Venture Partners The execution caveat is substantial: Rosenthal says tokenized real estate had yet to produce a breakout product by 2026, citing both challenging tokenization UX and the difficulty of scaled real-estate asset management. Lightspeed Venture Partners
• Rosenthal's timing thesis is that tokenization technology is now usable, large financial institutions including BlackRock and Fidelity are adopting tokenized assets, and housing's rising barriers to ownership create demand for easier investment access. Lightspeed Venture Partners The company says the seed capital will fund legal/regulatory work and product structuring, a functional launch product, and hiring; it reported about 10 employees and hiring across core functions. Lightspeed Venture Partners

• Robert Greene says public fear and anger over AI are widening the gap between the public and those controlling the technology; he sees demands for AI laws and data-center restrictions as a potential source of major change. My First Million
• He warns that AI could enable dangerous misuse, citing rising phishing and scams on phones and speculating about access to information for making biological or other weapons. He also fears that reliance on AI could weaken young people’s ability to write, read deeply, and think independently, while acknowledging possible benefits such as cancer treatment and climate solutions. My First Million

• Reducto’s sponsored launch of r-1 mini offers document parsing from $1 per 1,000 pages, preserving reading order, tables, and layout; Reducto says it processes more than a billion pages monthly for AI teams at Scale AI, Harvey, Toast, and Vanta. Not Boring by Packy McCormick
• The essay argues that AI’s utility, revenue growth, and lab profitability do not rule out a bubble: those conditions can draw in capital and competitors, leading to overbuilding and price collapses even as consumers benefit. Not Boring by Packy McCormick The scale of the compute bets is substantial: Anthropic reported $900.9 billion in commitments through 2045, about half expected before 2030 ends; OpenAI’s compute-spending target for 2030 was about $600 billion, down from a previously touted $1.4 trillion. Not Boring by Packy McCormick Anthropic had operating profit in Q2 and, according to the essay, would have had it in Q3 excluding roughly $2 billion in noncash donation-matching charges. Not Boring by Packy McCormick
• Packy’s investment lens is to favor durable competitive advantages, scale and cost advantages, and companies able to adopt better technologies rather than be displaced; he warns AI may accelerate both imitation and advantage-building, citing Claude’s weakening of NVIDIA’s CUDA moat. Not Boring by Packy McCormick A cited Nordhaus study estimated that producers captured about 2.2% of innovation’s social surplus in 1948–2001, versus roughly 97–98% for consumers—an example of transformative technology creating value without equivalent producer returns. Not Boring by Packy McCormick

• Andrew Ng says software engineering is among the roles most affected by AI, but software-engineering job postings are up as AI makes engineers more productive and companies want to hire more of them. He says his teams would not hire an engineer who cannot use a coding agent because their productivity is much lower; he also criticizes universities for graduating computer-science students without experience using coding agents or calling AI APIs. ET Digital
• Ng distinguishes AI’s ability to get work done from its ability to support learning: he says students using AI for homework can earn better homework grades but learn less and perform worse on closed-book finals. He says Learn Vector, backed by Coursera, is working on AI that helps people learn and scalable learning experiences. ET Digital
• On AGI, Ng argues that human-level AI capable of any intellectual task is not here yet and may be decades away; he notes that current AI still cannot perform some tasks humans can, while broader definitions of AGI vary. ET Digital

• Jason endorsed the post as hard-fought founder advice. @Jason
• Ro says its 2022 refounding followed 2021 growth of ~9% YoY at a $270M run rate while burning after raising ~$800M; by 2026, it reported a $3B+ run rate and revenue growth above 100% CAGR over the prior three years. @ZReitano The company says it nearly 10x’d since the reset while headcount fell from 596 at its May 2022 peak to 406 in October 2026; ARR per employee rose more than 10x and People & Opex fell below 15% of net revenue. @ZReitano
• To narrow its focus, Ro shut down 10+ products, cut ~$50M+ in DTC revenue it described as vanity revenue with upside-down unit economics, and ended B2B and external API efforts. @ZReitano It also consolidated its health offerings under one Ro brand, aiming to provide a single trusted place for patients across health categories. @ZReitano
• The operating reset required business cases and functional-lead, CFO, and founder approval for headcount, ended automatic backfilling, and replaced detailed long-term roadmaps with annual priorities and precise 90-day planning. @ZReitano The account cautions that its lessons come from one company and one founder’s experience, with survivorship bias and no guarantee they generalize. @ZReitano

• a16z’s B2A thesis is that agents are moving from answering questions to completing tasks and choosing products, subscriptions, and purchases; the article says agents already account for most web traffic and production code, while noting that the business models enabled by this shift are still largely unseen. @omooretweets
• Early commerce signals include Shopify-reported AI traffic to stores growing 8× year over year in Q1 2026 and orders from AI-powered searches growing nearly 13×; the article also says AI-referred consumer traffic has higher engagement and conversion and a lower bounce rate than non-AI traffic. @omooretweets
• Coding agents offer an adoption precedent: as of September 2026, more than 40% of GitHub pull requests reportedly started with an agent draft, up from 17% six months earlier; reported infrastructure spillovers include AI tools launching over 60% of new Supabase databases and agents accounting for nearly 40% of Stripe documentation traffic and 70% of CLI requests for API resources. @omooretweets
• The article’s opportunity map includes agent models and harnesses; agent-enabled businesses such as self-running SMBs, AI-intermediated marketplaces, and auto-optimizers; incumbents optimizing beyond discovery for agent-led evaluation and transactions; and agent infrastructure. It flags payments, fraud, identity, and communications as strained even at relatively low agent volumes, while many enterprises still block agent traffic over fraud and chargeback concerns. @omooretweets

YC resurfaced a Social Radars interview recorded December 4, 2024, in which former Theranos employee Tyler Shultz describes discovering that the lab’s reality differed from the company’s claims, identifying failed blood tests and other red flags, facing a culture of fear, trying to raise concerns internally and with regulators, and encountering legal pressure after speaking to The Wall Street Journal—a cautionary case on technology validation and startup governance. @ycombinator

Andrew Chen congratulated Azra Games and its leader Mark Otero on UnGodly, saying he looked forward to playing it on iPhone Duo @andrewchen. The game's account said it made UnGodly after tiring of waiting for God of War on mobile and urged users to download it @UngodlyGame.

• Stebbings now sees AI sovereignty as a strategic imperative and describes Mistral as Europe’s remaining frontier-model hope. @HarryStebbings Mistral says capital—not energy—is its main constraint on adding compute: energy is about 10% of compute’s total cost of ownership, Europe lacks capacity, customers may need thousands of GPUs as use cases scale, and Mistral is investing equity ahead of demand. @HarryStebbings
• Mistral characterizes inference as a commodity-like market with price-sensitive on-demand sales and a premium tier for mature customers seeking SLAs, reserved physical capacity, and latency and generation-speed guarantees; more precisely specified demand helps Mistral price that capacity. @HarryStebbings
• Mistral says ambiguous enterprise problems and human-in-the-loop objectives resist self-improving loops, so it uses forward-deployed engineers (FDEs) to define problems better. @HarryStebbings
• Mistral warns that reliance on two or three centralized model providers could give them significant leverage and let them use customer-generated data to improve systems and compete for those customers’ business; Arthur Mensch reportedly considers that outcome already priced in. @HarryStebbings
• Mistral says models’ ability to carry out cyberattacks makes security of supply for both infrastructure and model IP important to customers. @HarryStebbings
• Mistral argues it can go public in Europe, citing worldwide investor appetite for European private and public equity, and says it is preparing for public-company operating constraints while pursuing commercial milestones. @HarryStebbings

As renewed attention turns to Elizabeth Holmes, Jessica Livingston urges people to hear the story of Tyler Shultz, whom she identifies as the original whistleblower @jesslivingston.

Garry Tan argues that Nvidia’s board would be ill-advised to oust Jensen Huang, saying his leadership is central to the company’s operations and that a transition could cause a massive market-cap decline; this is a hypothetical warning, not a reported leadership change.@garrytan

Standard Capital opened applications for its latest Series A funding cycle, with applications due December 1 and applicants hearing back by December 11 @daltonc. Applications and additional information: https://www.standardcap.com/ @daltonc.