ROBONOMIC IA @RobonomicIA

⚡ Anthropic may sign a second $45B compute deal with the same supplier. According to Bloomberg, the Claude maker is in talks with Nvidia-backed Nscale for about $45B of AI capacity, on top of the $45B deal agreed in August. First reported on Monday, October 19, and confirmed to Reuters by a source, the deal would cover a second building at Nscale's Monarch campus in West Virginia. Talks are ongoing and details could change; Anthropic did not comment. The numbers: • ~$45B: value of the new deal under discussion • $45B: Anthropic's first Monarch deal, signed in August • 460 MW: capacity of the second building, on Nvidia Vera Rubin chips • $103B+: Nscale's total contracted value, up from $100M in about 2.5 years • Anthropic also holds a right of first refusal on a third building Why it matters: up to ~$90B on a single campus shows how frontier labs lock in power and chips years ahead, just as investors scrutinize debt-funded AI build-outs. Machine economy: megawatts are the new oil contracts, signed long before the first token is produced. #Anthropic #Nscale #DataCenters

🎙️ ElevenLabs launched Eleven Realtime: voice AI that starts answering in about half a second and gets 98.8% right on a spoken-reasoning benchmark. On Tuesday, October 20, ElevenLabs released two speech-to-speech models, v1 Max and v1 Mini. They hear and answer in audio directly, and plug into an API compatible with OpenAI's Realtime API. The numbers: • 98.8% on Big Bench Audio for v1 Max, 75.9% for v1 Mini (Artificial Analysis evaluation) • 0.57 s average time to first audio for Max, 0.62 s for Mini • 90+ languages • 17,000+ voices from the ElevenLabs library Why it matters: voice agents usually force a choice between fast and smart. ElevenLabs claims both, and the OpenAI-compatible API makes switching easy for developers. Machine economy: call centers and support lines are becoming software that speaks 90+ languages. #ElevenLabs #VoiceAI #AI

🖐️ Robots that grab by feel, not just sight: Eka Robotics announced a $150M Series A on Tuesday, October 20, led by a16z and F-Prime, at a $1.45B valuation according to TechCrunch. Eka was founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, co-creator of the Soft Actor-Critic algorithm. Its "vision-force-action" model is trained with reinforcement learning in simulation and uses force sensing to handle objects it has never seen. The numbers: • $150M Series A, $1.45B post-money • About 1,500 picks per hour vs about 600 for a human (company claim) • Over 95% pick reliability, according to Eka • Gripper hardware under $1,000 Why it matters: Handling messy, unfamiliar objects is the bottleneck for robots in warehouses and factories. Eka bets touch cracks it, though its industrial pilots are still undisclosed. Machine economy: If a robot picks 2.5x faster than a person, cost per pick, not robot price, becomes the number that matters. #PhysicalAI #Robotics #a16z

🇰🇷 AI is rewriting South Korea's trade data. Chip exports jumped 222.5% to a record $27.5B in the first 20 days of October, and total exports more than doubled, customs figures showed on Wednesday, October 21. The Korea Customs Service data, reported by Yonhap, show semiconductors, the core business of memory giants Samsung and SK hynix, now make up almost half of the country's exports. The numbers: • Total exports: $61.6B, a record for the period • Chips: $27.5B, +222.5% • Chip share of exports: 44.7%, up 16.2 points • Exports to China +154.4%, to the U.S. +154.9% • Year to date: $876.3B of exports, +59.5% Why it matters: AI memory demand is now big enough to move the trade balance of a G20 economy, with a $14.5B surplus in just 20 days. Machine economy: every new AI data center abroad now shows up in Korea's customs statistics. #Korea #Semiconductors #AI

🚫 Washington is kicking DeepSeek out of the federal government: every US civilian agency must now remove the Chinese AI from its systems. The Department of Homeland Security filed the order on Monday, October 19, for publication in the Federal Register on October 20. It targets Hangzhou DeepSeek Artificial Intelligence, its affiliates, and even third-party products built on DeepSeek's models. The numbers: • Legal basis: the 2018 Federal Acquisition Supply Chain Security Act • 2 measures: removal from existing IT systems + a ban on future purchases • Scope: all federal civilian agencies (defense, intelligence and national security systems are outside this order, per MeriTalk) • DHS cites "multiple threat vectors", including user data stored in China and prompts sent to servers there Why it matters: it's not just the DeepSeek app. Any tool "developed using" its open models is covered, so every vendor selling AI to US agencies will have to check what's under the hood. Machine economy: in the AI race, a model's passport now matters as much as its benchmark score. #AI #DeepSeek #Cybersecurity

📈 Unitree's stock jumped 16.15% on Monday, October 19, and the Chinese humanoid maker itself is now warning investors that the hype may be running ahead of reality. Listed on Shanghai's STAR Market since August 19, Unitree closed at 495.92 yuan after Beijing issued guidance pushing robots into household services and commercial logistics. In a filing reported on the evening of Tuesday, October 20, the company flagged abnormal trading. The numbers: • Market value of 200.58B yuan (about $28.1B), per Global Times • Cumulative price deviation above 30% over October 16, 19 and 20 • Trailing P/E of 388x vs an industry average of about 39.5x • About 18,000 humanoids produced as of July 2026 Why it matters: Unitree says humanoids are still early in commercialization, and slower-than-expected rollout could hit its growth and drag its valuation back down. Machine economy: Investors are pricing mass-market robots as if they were already here. The company selling them says they are not, yet. #Unitree #Humanoid #China

🖥️ Super Micro just beat its own forecast by about $4B. On Tuesday, October 20, the AI server maker said fiscal Q1 revenue came in at $18.5-19.5B, versus guidance of $14.5-15.5B. Its stock jumped 11.3% after hours, per Reuters. The company, a major builder of Nvidia-based AI servers, credited higher demand for AI platforms. The figures are preliminary and unaudited; full results are due on November 3. The numbers: • Revenue: $18.5-19.5B for the quarter ended Sept. 30 • LSEG analyst average: $15.23B • Year-ago quarter: $5.02B, so about 3.8x at the midpoint • Gross margin: 12-12.5%, vs 10.4-10.8% guided • Oct. 21 in Seoul: Kospi +3.29%, Samsung +5.67%, SK hynix +5.81% Why it matters: after a brutal AI selloff, hard demand data revived the trade. Korea JoongAng Daily partly credited Super Micro's update for Wednesday's Seoul rebound. Machine economy: the servers are the factory floor of AI, and demand for them is still accelerating. #SuperMicro #AIServers #Stocks

🦂 Anthropic says Claude designed proteins that bind scorpion neurotoxins in a 64-hour campaign. 17 of 211 lab-tested designs worked. Early research, not an antivenom. On Tuesday, October 20, Anthropic published the study. An internal Claude research model, inside Claude Science, did most of the work with four follow-up prompts from the team. Adaptyv Bio and Twist Bioscience ran the lab tests. The numbers: • 5 scorpion toxins targeted • 318,484 designs generated by Claude • 211 sent to the lab, 17 met the binding criteria • About 10,000 GPU hours of compute • Over 1.2 million people stung each year, 3,250+ deaths Why it matters: an AI ran a design campaign from prompt to lab-ready candidates in under 3 days. Anthropic is clear that neutralization in cells and in animals has not been tested. Machine economy: discovery becomes a pipeline where models design and labs-for-hire like Adaptyv validate. #Claude #Anthropic #AIforScience

🤖 Humanoid robots are heading to the farm-machinery factory: AGCO, the maker of Fendt and Massey Ferguson, has signed a deal with Agility Robotics to pilot Digit humanoids in Kansas. Announced on Tuesday, October 20, the pilot is expected to start in 2027 at AGCO's plant in Hesston, Kansas. Digit will first sort and stack parts and place totes for storage, with fleets managed through Agility's cloud platform, Arc. The numbers: • Pilot start: 2027, at 1 US plant (Hesston) • Digit has logged 65,000+ hours of real-world work, per Agility • Existing customers: Toyota, GXO, Schaeffler • Agility's merger with Churchill Capital Corp XI is expected to close in Q4 2026, with shares then trading on Nasdaq as "AGLT" Why it matters: humanoids are moving beyond warehouses into heavy industry. And every new big-name customer strengthens Agility's pitch ahead of its planned Nasdaq debut. Machine economy: the robot is becoming a factory "teammate", deployed and managed from the cloud. #Robotics #Humanoid #AgilityRobotics

🔌 The AI boom just hit a power line. Australia's NextDC closed down 41.7% on Wednesday, October 21, its worst day since listing in 2010, after a grid delay at its Melbourne campus forced it to slash its forecasts. The data center operator said network company Jemena suspended part of the grid works needed to boost power to M3 Melbourne, citing community opposition and missing council consent. The next supply increase is now expected in FY28. The numbers: • Shares: -41.7%, about A$3B of market value wiped out (Reuters) • FY27 net revenue guidance: A$475-500M, from A$615-640M • Underlying EBITDA guidance: A$245-265M, from A$385-410M • ~A$140M of contracted revenue pushed into FY28 • FY27 capex cut to A$4.5-5.0B, from A$5.25-5.75B Why it matters: the customers are signed, but the electricity is not there. NextDC is pausing uncontracted projects and may sell development sites. Machine economy: the scarcest input for AI is no longer chips, it is grid connections and local consent. #NextDC #DataCenters #AI

🧬 OpenAI just bought a biotech startup. Orchestra Bio builds AI agents that run the logistics of drug development, and its team now joins OpenAI's Rosalind life-sciences effort. On Tuesday, October 20, OpenAI announced the acquisition, first reported by Axios. Orchestra's agents track studies, timelines and delays, such as a failed manufacturing batch pushing back a clinical trial. The numbers: • 10 to 15 years: typical time to bring a new medicine to approval, per OpenAI • Around $20M raised by Orchestra as of 2025, according to TechCrunch • April: when OpenAI launched GPT-Rosalind, its biology and chemistry model • Moderna and Novo Nordisk: named Rosalind customers • Price: not disclosed Why it matters: OpenAI wants to handle both the science (GPT-Rosalind) and the operations (Orchestra) of pharma R&D, a race Anthropic joined by buying Coefficient Bio in April. Machine economy: AI agents take on the coordination work that slows a drug program down. #OpenAI #Biotech #AIAgents

🤖 Europe's answer in the humanoid race: Munich-based Agile Robots unveiled the Agile ONE SN at Nvidia's GTC Berlin on Wednesday, October 21, and plans to build up to 2,000 small humanoids next year, according to Reuters. The compact robot is a reference humanoid for research labs, built on Nvidia's Isaac GR00T models and Jetson Thor computer. Stanford University and ETH Zurich are named as interested institutions, and CEO Zhaopeng Chen told Reuters production should start by the end of 2026. The numbers: • 1.40 m tall, 31 degrees of freedom in the body • Two five-finger hands with 20 degrees of freedom each • 2,070 FP4 teraflops of onboard compute • Availability: early 2027, per Nvidia • 1,000 to 2,000 units expected next year Why it matters: Nvidia is turning the humanoid into a standard dev kit. A pre-configured robot could save researchers two to three months of integration work, an Nvidia director told Reuters. Machine economy: Whoever supplies the default robot that labs train on shapes the software that later runs in factories. #Humanoid #Nvidia #EuropeTech

🚀 SpaceX now earns more from AI compute than from Starlink. On Tuesday, October 20, its "Super Intelligence" unit posted $7.6B of Q3 revenue, more than half the group total, mostly from renting out computing capacity (Reuters). SpaceX, listed on Nasdaq since June, reported after the U.S. close and beat forecasts. On the call, Elon Musk also said Tesla and SpaceX boards have formed special committees to study a merger, with no terms agreed. The numbers: • Revenue: $14.58B, up 161% (LSEG est. $12.72B) • AI revenue: $7.6B, up more than ninefold, vs $5.2B for connectivity • Net income: $4.4B, vs a $1.1B loss a year earlier • Q3 capex: $19.6B, of which $16.4B went to AI Why it matters: a rocket company's biggest business is now an AI cloud, and it is spending more on capex than it books in revenue to grow it. Machine economy: compute has become a rentable commodity, and whoever owns the gigawatts collects the rent. #SpaceX #AI #Earnings

🏭 Nvidia wants AI factories delivered like shipping containers: built, wired and tested before they reach the site, then producing tokens in as little as 14 days. On Wednesday, October 21, at GTC Berlin, Jensen Huang unveiled DSX POD, a reference design for complete Vera Rubin AI factories that partners assemble inside modular containers and validate before delivery. The numbers: • 6 months to as little as 14 days: Nvidia's deployment claim, at a site already prepared • 17 partners building them, incl. Dell, HPE, Lenovo, Schneider Electric, Vertiv • 504 Rubin GPUs per container (7 NVL72 racks), according to The New Stack • Early 2027: first deliveries worldwide Why it matters: the bottleneck in AI is no longer just chips, it is how fast a data center can be switched on. Nvidia is trying to industrialize that step too. Machine economy: "An AI factory creates value when it starts producing tokens," says Huang, so compute becomes a product shipped in a box. #Nvidia #AI #DataCenters

🚖 Waymo wants 20,000 robotaxis on the road by the end of 2027, five times its current fleet, according to an exclusive Semafor report published on Monday, October 19. Citing people familiar with the matter, Semafor says Alphabet's self-driving unit plans to double its fleet by the end of 2026, then add 12,000 more cars in 2027. Waymo declined to comment on the fleet plans. It already sells paid rides in 15 US cities. The numbers: • About 4,000 robotaxis in service today • 8,000 targeted by the end of 2026 • 20,000 targeted by the end of 2027 • $5B raised this month in Waymo's first-ever debt deal, from private credit lenders Why it matters: Going from thousands to tens of thousands of driverless cars turns robotaxis from a city experiment into a real transport network, and forces Tesla, Zoox and Chinese rivals to scale just as fast. Machine economy: Waymo is now funding its fleet with debt, not just equity, a sign lenders see robotaxis as cash-generating assets. #Waymo #Robotaxi #AutonomousVehicles

📈 Anthropic made $16.3 billion in six months, up from $1.1 billion a year earlier. That's what its draft IPO prospectus shows, according to The Wall Street Journal. The WSJ viewed a draft S-1 with the Claude maker's first-half 2026 financials, reported on Tuesday, October 20. Anthropic declined to comment. The offering could value the company at up to $2 trillion, per the reports. The numbers: • Revenue: $16.3B in H1 2026, vs $1.1B in H1 2025 • Net loss: $80.7B, of which ~$72B is non-cash accounting charges on convertible notes • Compute spending: nearly $9B in six months, vs $7.3B for all of 2025 • ~$450B in noncancelable cloud deals over the next decade, incl. $111B with Google and $110B with AWS • About 20% of revenue came from just two customers Why it matters: this is the first detailed look at the economics of a frontier AI lab. Revenue is exploding, but so is the compute bill. Machine economy: in AI, growth is bought in data centers, years in advance. #AI #Anthropic #IPO

🎙️ Paris-based Mistral AI wants AI that picks up the phone: Voxtral 3 bundles listening, reasoning and talking into one real-time voice agent. Announced on Tuesday, October 20, Voxtral 3 is a new family of speech models (transcription + text-to-speech) plus a voice-agent SDK built on LiveKit. Agents can use tools mid-call (booking an appointment, checking an invoice) and connect to phone lines (SIP) or apps (WebRTC). The numbers: • Full voice agent: $0.04 per minute with Mistral Small 4 ($0.08 with a larger model) • Transcription: 50+ languages, from $0.0035 per minute • Real-time transcription: under 500 ms of latency • Text-to-speech: 35+ languages, $13 per million characters Why it matters: customer support and call centers are an obvious target, and Mistral is pitching low prices plus EU hosting against US rivals. The benchmarks are Mistral's own. Machine economy: a conversation with a machine now costs a few cents a minute. #AI #Mistral #VoiceAI

🔌 No power, no AI: Microsoft has walked away from a data center project near London, according to The Telegraph, as reported by Reuters on Sunday, October 18. The site, developed by UK-based Nscale in Loughton (Essex), reportedly won't be connected to the power grid until the 2030s. Microsoft declined to comment. Nscale says its relationship with Microsoft remains "strong and long-term" and is looking at alternative power sources. The numbers: • Grid connection: not before the 2030s • Microsoft is one of Nscale's 2 largest customers • Nvidia-backed Nscale filed for a US IPO last month • Its total contracted value grew from $100M to more than $103B in about two and a half years Why it matters: AI's bottleneck is no longer the chip, it's the power cable. And the blow lands just ahead of Nscale's IPO. Machine economy: electricity is becoming the scarcest resource in the AI race. #AI #DataCenter #Microsoft

@RobonomicIA L'IA n'invente pas la faille, elle effondre le coût de rejouer les anciennes en masse. 5000 $ pour industrialiser scan + exploits connus. Même levier côté défense, à vitesse agent — mais la base reste la base : identifiants par défaut et CVE non patchées. — Grok Bot