Monday, October 5, 2026·Live edition·Updated 11:56 PM ET
Tech and AI
Chips, artificial intelligence and the big tech companies. 23 stories in the last three days.
AI and chips
Nokia CEO says AI data center buildout is still early
Illustration for the Tech desk
Nokia CEO Justin Hotard says customers would build data centers twice as fast if memory and power constraints eased. He views the AI infrastructure expansion as still in its early stages.
The Wikimedia Foundation says its May service disruption may have been caused by OpenAI agents. The group is investigating whether automated tools from the artificial-intelligence firm triggered the problem.
Microsoft reduced its expenditure on Anthropic's Claude by over 33 percent. Meta reported that employee adoption of the tool dropped to 30,000 from 60,000.
A VentureBeat survey found Google is the primary AI platform for 48% of production users, ahead of OpenAI at 22% and Microsoft at 13%. Most respondents report low GPU utilization and plan to add or replace platforms within a year.
Morgan Stanley and Wells Fargo raise SpaceX price targets on Starship progress
Morgan Stanley set a $300 target for SpaceX, while Wells Fargo set a $212 target, citing future rocket launches and space-based computing plans. Deutsche Bank sees the space sector correction nearing a trough.
Michael Burry questions Nvidia’s AI hardware residual value methodology
Michael Burry argues Nvidia’s residual value estimates rely on modeled rental income rather than resale prices. He notes that changing useful-life assumptions significantly lowers the calculated value for older chips.
Researchers spot AI agents querying Alibaba map service on Tencent infrastructure
Researchers found a fleet of AI agents running on Tencent infrastructure repeatedly querying Alibaba's Amap service for directions. The agents bypassed normal API restrictions but showed no signs of coordination or malicious intent.
Meta’s Muse AI assistant tracks user connections and personal details
Meta’s Muse AI assistant creates detailed records of individuals in a user’s life by merging public data with information from messages, financial accounts and health services. The system monitors relationship patterns and requires human confirmation before executing significant tasks.
Goldman Sachs sees U.S. data-center capacity reaching 64 gigawatts by year-end 2026
Goldman Sachs forecasts U.S. data-center capacity will hit 64 gigawatts by the end of 2026 and 90 gigawatts by late 2027. The bank notes electricity demand will grow 38 percent annually, though local opposition could complicate future permits.
Telus and Ast SpaceMobile complete first satellite-to-smartphone network test
Telus and Ast SpaceMobile finished their initial integration test for a satellite-to-smartphone network. Telus expects to offer the service to customers within a year to cover remote Canadian areas.
Clockwork.io secures $31 million for AI resilience tools
Clockwork.io secured $31 million to grow its AI resilience tools, bringing total capital to $73 million. The startup's technology helps LinkedIn and Together AI avoid GPU downtime.
Navitas and Microchip introduce new power setup for AI data centers
Navitas Semiconductor and Microchip Technology launched a power conversion platform for AI facilities. The system converts 800V to 6V DC, aiming for 96% efficiency and 2,100 W/in³ density, with materials ready for the October 12 event in San Jose.
TeraWulf doubles contracted power at Muskie campus to 1GW
TeraWulf amended its agreement with Kentucky Power to raise contracted capacity at its Muskie site from 500MW to 1GW. The second 500MW phase is now targeted for 2029 instead of 2030, pending regulatory approval.
D-Wave partners with University of Arkansas on quantum supply-chain research
D-Wave and the University of Arkansas launched a quantum supply-chain initiative to study logistics optimization in retail, transportation and defense. The program includes a new support fund for independent research and education.
Companies struggle to predict AI costs as token usage varies widely
Only 11% of surveyed organizations forecast AI spending within 10% of actual costs, with cheaper models sometimes costing more than pricier alternatives.
Aletheia sees AMD server CPU supply constrained through 2027 before 2028 expansion
Aletheia projects AMD server CPU revenue of $40 billion and 13.5 million shipments in 2027, but notes supply limits. The firm expects manufacturing capacity to expand in 2028, lifting revenue 70 percent and shipments 30 percent.
Rosenblatt sees Amazon Web Services revenue reaching $334 billion by 2028
Rosenblatt analysts project Amazon Web Services revenue will grow from $129 billion in 2025 to $334 billion by 2028. The firm cites a $496 billion backlog and more than $200 billion in Trainium chip commitments as drivers of the forecast.
SpaceX to rename AI unit to SpaceXSI following Trump directive
Elon Musk confirmed SpaceX will rename its artificial intelligence division from SpaceXAI to SpaceXSI. The change follows a directive from President Trump to replace artificial intelligence with super intelligence in official government communications.
Micron results support J.P. Morgan view of memory shortage through 2028
Micron reported that high-bandwidth memory supply for 2027 is largely sold out at higher prices. J.P. Morgan sees demand growth of 63 percent for that product and notes customer commitments worth $32 billion extending into 2031.
OpenAI Chief argues AI gains outweigh risks despite some harm
Sam Altman contends that widespread access to artificial intelligence produces more value than damage. The firm now supports stricter safety measures, aligning closer to Anthropic on oversight.
Airbnb CEO says consumer AI needs system-level architecture to replace apps
Airbnb CEO Brian Chesky argues that artificial intelligence requires richer interfaces and system-level design before replacing current applications. The company plans to introduce voice tools for finding trips and handling support inquiries this autumn.
Reflection prepares Nvidia-backed AI model to rival Chinese systems
Reflection is developing an open-weight AI model backed by Nvidia to compete with leading Chinese systems. The startup signed server deals with Nebius and SpaceX and is testing localized AI setups with South Korea's Shinsegae Group.
Musk confirms talks with TSMC over dedicated chipmaking partnership
Elon Musk said discussions are ongoing with TSMC about a chipmaking partnership that would supply Tesla, SpaceX and xAI. Intel is currently the sole other partner for the project.