SpaceX to challenge major wireless carriers

SpaceX takes on wireless carriers, Microsoft AI tops doctors, and more.

SpaceX to challenge major wireless carriers

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

📡 SpaceX moves to challenge major wireless carriers

🩺 Microsoft AI beats doctors on hard cases

🇺🇸 US to curb AI investment in China

💰 OpenAI nears $30B funding round

🎨 Canva launches its first AI image model

Plus: 🎁 15 more stories you might like, 🧰 6 tools, and 📚 5 papers.

SpaceX moves to challenge major wireless carriers LINK

  • SpaceX is in talks with cellular infrastructure firms to lease at least 80,000 sites across the US, a step toward building a fourth nationwide wireless carrier to rival AT&T, Verizon, and T-Mobile, Bloomberg reported yesterday.
  • The 80,000 sites would match half of the 158,500 cell towers running in the US last year, and SpaceX is buying nationwide low-band spectrum while in talks with tower operator Crown Castle, the report said.
  • TD Cowen estimated a nationwide ground network would cost more than $80 billion and take years, and shares of AT&T, Verizon, and T-Mobile dropped between 4.2% and 5% in extended trading.

Microsoft AI beats doctors on hard cases LINK

  • Microsoft researchers built an AI system that works through hard medical cases step by step, diagnosing them more than twice as accurately as general physicians, according to a study published in Science yesterday.
  • Instead of matching symptoms to a disease in one go, the system uses an orchestration tool to coordinate a simulated panel of specialists, asking questions, ordering tests and revising its possible diagnoses as new evidence arrives.
  • The approach also lowered the cost of diagnostic work-ups, and Microsoft is now teaming with Beth Israel Deaconess Medical Center in Boston to test whether it can help triage emergency department patients in real settings.

🇺🇸 US to curb AI investment in China LINK

  • The US Treasury has proposed rules to restrict American investments in China's artificial intelligence, military technology, and life sciences sectors, acting under a 2025 law meant to tighten checks on money flowing to rival nations.
  • The draft rules cover investments above $50,000 in advanced semiconductors, AI, certain biotech materials, and technologies like brain-computer interfaces, and the Treasury must still review public feedback before finalizing them.
  • Earlier this year, the Treasury fined Amidi LLC $200,000 after a subsidiary failed to report its $92,500 stake in Noematrix, a Shanghai firm working on AI, robotics, and embodied intelligence.

OpenAI nears $30B funding round LINK

  • OpenAI is closing in on a funding round of up to $30 billion, led by investors from the United Arab Emirates and the firm Coatue, and wants signatures by the end of next week.
  • The round sets a $1.25 trillion pre-money valuation, $150 billion below the $1.4 trillion OpenAI pitched to backers including BlackRock and MGX earlier this month, and is framed as a bridge until a public offering.
  • Pitched as moving faster than typical, the deal aims to close before rival Anthropic's expected mid-November listing, and OpenAI is also negotiating a separate commercial deal to secure more computing capacity from the UAE.

Canva launches its first AI image model LINK

  • Canva has released Craft, its first AI image model, built from scratch by Canva Research specifically for design work rather than photography, with the aim of understanding layouts and matching what users picture.
  • Craft makes 2K images at any aspect ratio with editable results, trained on millions of human-made templates from Canva's library like posters, social posts, presentations, infographics, and menus, and it placed fourth on DesignArena behind three OpenAI models.
  • The model is available now in the Canva app with a paid subscription and to developers through a public API, with the Canva MCP Server coming soon, plus two cheaper variants, Craft-fast and Craft-ultrafast.

Other news you might like

  • Greek parliament greenlights under-15 social media ban LINK
  • ‘Pure arrogance’: anger in Finland as Google illegally fells 250,000 trees for datacentre LINK
  • Left alone with nothing to do, AI agents started playing LINK
  • Trump asked Musk to lift Starlink slowdown, WSJ reports LINK
  • Australia is forcing Apple to loosen its grip on iPhone app distribution LINK
  • Switch 2 security takes a hit as hackers unlock homebrew access LINK
  • Illegal Nvidia Blackwell Ultra B300 GPUs found inside state-owned China Telecom's network LINK
  • Musk invests $1B in CBS and CNN owner Skydance LINK
  • Oura sues rival Ultrahuman again, threatening its U.S. smart ring comeback LINK
  • Claude and Grok among AI agents used to automate attacks on Latin American institutions LINK
  • OpenAI makes its Jev competitor widely available as decision model competition heats up LINK
  • a16z Drops Crypto From Its Name, Widens Bets to AI and Frontier Tech LINK
  • Warhound raises $200M to engineer war dogs and arm them with Anduril gear LINK
  • Amazon Rebrands Kindle Unlimited To Kindle+ LINK
  • Arc browser sunsets December 22 as The Browser Company doubles down on Dia LINK

Trending tools

Doneas: helps solo builders plan weekly, join short daily accountability calls with five peers, and track shipped work on a public progress page. LINK

heal.​dev: runs your app in a production-like sandbox to catch runtime bugs, performance regressions, and vulnerabilities, reproducing each defect before reporting it. LINK

Product Hunt Visual Launch Editor: lets makers design and customize their Product Hunt launch pages visually, helping products stand out to the tech community. LINK

Resend Inboxes: create and manage email inboxes programmatically, letting developers build transactional email flows and reach users reliably instead of spam folders. LINK

Latch: blocks distracting apps until you scan a physical NFC or QR key or walk a set number of steps, with no accounts, ads, or tracking. LINK

DoOne: a single-task focus timer that runs issue actions like labeling and assignment in Git, Linear, Sentry, and Todoist when you finish. LINK

Trending papers & reports

Tool-using chatbots often take harmful actions even for requests they would refuse to answer in plain conversation, failing to carry over their safety limits in 10 to 52 percent of cases. LINK

AI coding checkers separate the agent that writes software from an independent one that verifies it, catching more of 46 planted bugs than the builder's own tests and leaving the final release call to a human. LINK

Content moderation tools judge messages by their words instead of what they actually do, wrongly flagging dialects and minority voices while missing real harm, with 51 of 55 system-language pairs falling short of native-speaker accuracy. LINK

Gender gap in AI jobs emerges as women see slower employment growth than men in occupations whose tasks AI can do or assist with, a Pew Research Center analysis of government data finds. LINK

Near-universal internet use now reaches 97% of US adults, up from about half when Pew Research Center began tracking in early 2000, per a survey of 5,511 adults. LINK

See you tomorrow for a new dose of ☕️ Techpresso!

More from the archive