☕️ Big Tech capex hits $725 billion in 2026

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

💰 Big Tech capex hits $725 billion in 2026

🛑 White House blocks Anthropic Mythos expansion

🕶️ Apple reportedly abandons Vision Pro

🤖 SoftBank is creating a robotics company that builds data centers

🏨 Uber enters the hotel booking business

🎵 Spotify introduces verified artist badges to help distinguish humans from AI

Plus: 🎁 18 other news you might like, 🧰 6 tools, and 📚 3 papers.

💰 Big Tech capex hits $725 billion in 2026 LINK

Google, Amazon, Microsoft, and Meta plan to spend a combined $725 billion on capex in 2026, a 77% jump from last year's record $410 billion, driven by AI infrastructure demand and climbing memory chip prices.

Microsoft set 2026 capex at $190 billion, with CFO Amy Hood blaming $25 billion of that on memory chip and component costs, while warning the company will stay capacity-constrained through at least 2026.

Alphabet matched Microsoft's $190 billion capex guidance after Google Cloud revenue grew 63% to $20 billion and its contract backlog doubled to $460 billion, pushing shares up 7% toward a $4.3 trillion valuation.

🛑 White House blocks Anthropic Mythos expansion LINK

The White House has pushed back on Anthropic's plan to widen access to its Mythos AI model to around 70 companies and organizations, according to an administration official who spoke anonymously on Wednesday night.

US officials worry Anthropic lacks the computing power to serve more Mythos users without hurting the government's own use of the model, which the company says is strong enough to enable dangerous cyberattacks.

Mythos, unveiled in early April, can reportedly detect and exploit vulnerabilities in critical software, and a small group of unauthorized users on a private online forum gained access the same day Anthropic announced its limited release plan.

🕶️ Apple reportedly abandons Vision Pro LINK

Apple has reportedly stopped work on the Vision Pro after weak sales of the M5 chip model released in October, which kept the $3,499 price tag and added a more comfortable head strap, according to MacRumors.

The product engineering team is being moved to other projects across the company, with a near-term focus on AR glasses to compete with Meta and longer-term work on a cheaper Pro-style successor.

Apple is also shifting engineering resources toward Siri and Apple Intelligence ahead of WWDC in June, as recent delays to its AI work have hurt the company's reputation with users and developers.

🤖 SoftBank is creating a robotics company that builds data centers LINK

SoftBank is planning to launch and list a standalone AI and robotics company in the U.S. called "Roze," which will build data centers and use robotics to make AI infrastructure construction more efficient, the Financial Times reported Thursday.

Masayoshi Son is leading the push, with executives targeting a roughly $100 billion valuation and an IPO as soon as this year, though the timeline could shift partly due to uncertainties from the conflict in the Middle East.

Roze could bundle existing energy, land and infrastructure assets from SoftBank's portfolio along with ABB Robotics, which SoftBank agreed to buy last year, and the listing may help offset its $30 billion-plus commitment to OpenAI.

🏨 Uber enters the hotel booking business LINK

Uber has launched hotel bookings inside its app for US customers, giving access to over 700,000 hotels worldwide through a partnership with Expedia Group, with Vrbo vacation rentals set to join later this year.

Uber One subscribers will get 20% off a rolling list of 10,000 hotels booked through the app, plus 10% back in Uber Credits on all bookings, as the company pitches the subscription harder.

CTO Praveen Neppalli Naga said agentic AI tools like Cursor cut the hotel booking feature's build time in half, alongside other launches including travel mode, a room service hub in Uber Eats, and Eats for the Way.

🎵 Spotify introduces verified artist badges to help distinguish humans from AI LINK

Spotify is rolling out a new "Verified by Spotify" badge that marks human artists in good standing, shutting out AI-generated or AI-persona profiles as streaming sites deal with a flood of machine-made tracks clogging their platforms.

To qualify for the light green checkmark, artists must show consistent listener engagement, follow platform policies, and display "signals of a real artist," with over 99 percent of actively sought-out artists verified at launch.

Spotify is also testing a new context section on artist profiles, described as "nutrition facts," showing career milestones, release activity, and touring activity in the About section on mobile over the coming weeks.

Other news you might like

• Google CEO says Pichai says people "love" AI Overviews and keep coming back to search moreLINK

• Gemini app can now generate Google Docs, PDF, Word, and other filesLINK

• Meta quietly rolls out stablecoin payments four years after demise of controversial Libra projectLINK

• Anthropic reviewing investor offers that would value the company at over $900 billionLINK

• Meta's AI bill keeps growing as spending forecast jumps to $145 billionLINK

• Claude AI agent’s confession after deleting a firm’s entire database: ‘I violated every principle I was given’LINK

• Visa expands stablecoin settlement network as volume hits $7 billion run rateLINK

• The nicest AI in the room is the one you should actually worry aboutLINK

• House committees probe Cursor parent, Airbnb over Chinese AILINK

• Google to sell TPU chips to 'select' customers in latest shot at NvidiaLINK

• Amazon chips no longer just a side dish, they're a $20B bizLINK

• Drone strikes on data centers spook Big Tech, halting Middle East projectsLINK

• OpenAI has effectively abandoned first-party Stargate data centers in favor of more flexible dealsLINK

• Instagram says it doesn’t want your tweet round upsLINK

• SpaceX's theorized data centers in space face 'significant technical complexity and unproven technologies,' and the 'unpredictable environment of space' means they may not be commercially viableLINK

• The Zig project's rationale for their firm anti-AI contribution policyLINK

🧰 Trending tools

VideoOS by Jupitrr AI: an end-to-end video marketing platform that handles scripting, recording, auto-editing, and social publishing in one workflow.LINK

Hera Launch: an AI motion designer that generates editable, code-based animations in seconds, eliminating the need for manual animation work.LINK

File Generation in Gemini: lets you generate downloadable files directly from AI interactions, saving time by skipping manual copy-paste into separate documents.LINK

Invite Only: a WhatsApp-native event platform letting hosts collect real RSVPs and UPI payments through a shareable link, no app download required.LINK

Rova AI: autonomously tests web and mobile apps by reading issue tickets from Jira or Linear, executing workflows, and reporting results without manual test scripts.LINK

Mistral Medium 3.5: a self-hostable 128B model balancing strong coding, reasoning, and instruction-following with 256k context and adjustable reasoning effort.LINK

📚 Trending papers & reports

Scientific replication studies cannot reliably distinguish "replicable" from "not replicable" results, because standard data structures systematically understate uncertainty and make different replicability rates appear separable when they are not.LINK

AI document delegation reveals that current large language models, even top frontier models, corrupt an average of 25% of document content during long editing workflows across 52 professional domains.LINK

Flavor knowledge in AI is already hidden inside a food ingredient database, and researchers found they could reliably recover at least fifteen distinct dimensions, including taste, texture, geography, and culture, without any human tasting.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!