The AI industry's financial stakes reached new heights this week, while a generational reckoning over jobs played out at graduation ceremonies across the country. From a historic math breakthrough to the politics of data centers, here's what moved the needle today.
---
Anthropic's Breakout Quarter — and Its Unusual Computing Deal
Anthropic is on track to post its first profitable quarter, projecting $10.9 billion in Q2 revenue — a 130% increase from the previous quarter — with an estimated $559 million in operating profit. The milestone reflects surging enterprise adoption of Claude and validates the company's sky-high $900 billion projected valuation.
But the story behind the numbers is striking: Anthropic is reportedly paying xAI $1.25 billion per month through May 2029 for computing power, effectively leasing the entire output of xAI's Colossus 1 data center. Deleted disclosures from SpaceX's S-1 filing suggest that Colossus II infrastructure was built at approximately $2.7 million per megawatt — meaning the Anthropic contract recoups SpaceX's capital expenditure in under a month. It's a window into the raw economics of frontier AI infrastructure.
The IPO Wave: OpenAI and SpaceX Aim for the Public Markets
OpenAI is preparing to file a confidential S-1 with the SEC, targeting a September 2026 listing at a valuation between $852 billion and $1 trillion. The filing would be one of the largest in history and would mark a defining moment for the AI era.
Meanwhile, SpaceX filed for a $75 billion capital raise at a valuation exceeding $1.75 trillion. The filing made a striking argument: 93% of the company's $28.5 trillion total addressable market is attributed to AI — a framing that asks public investors to think of SpaceX primarily as an AI infrastructure company. That claim raised eyebrows given that SpaceX's AI segment reported a $14 billion cash loss in 2025. Enterprise SaaS broadly is attracting record capital in this environment, with VC deal value hitting $173 billion in Q1 2026 as investors shift toward AI-native, outcome-based platforms.
OpenAI Breaks an 80-Year-Old Math Problem
An internal OpenAI model disproved the Erdős unit distance conjecture — an unsolved geometry problem dating back roughly 80 years — by identifying weakly super-linear scaling through possibilities human researchers would likely have overlooked. The result was significant enough to generate coverage beyond the AI press, with researchers suggesting the finding represents a qualitative shift: AI may now be capable of originating mathematics, not just assisting with it. Separately, GPT-5.5 running Codex achieved 25% accuracy on the FutureSim benchmark, which tests forecasting of real-world events 90 days out — outperforming crowd prediction markets.
Google I/O 2026: From Answering to Acting
Google's annual developer conference this week was organized around a single thesis: AI is shifting from tools that answer questions to agents that perform actions on your behalf. On the product side, Gemini is integrating Adobe, Canva, and CapCut, giving users the ability to generate and edit images, designs, and short videos directly in chat, with "Magic Layers" enabling further professional customization. Gemini Omni, a conversational video editing tool, is now live on YouTube Shorts and YouTube Create, with deliberate limits on audio manipulation to reduce deepfake risk. Google CEO Sundar Pichai also took an unusual step at a commencement ceremony where AI protesters heckled his speech — he acknowledged the legitimacy of public concerns about jobs and admitted he is uncertain whether current AI is on an absolute path to AGI.
AI Meets Finance and Design
OpenAI launched a personal finance feature in ChatGPT that connects to more than 12,000 financial institutions via Plaid, allowing Pro users to ask natural-language questions about their spending, debt, and taxes. The feature brings AI into territory that has historically been either left to licensed professionals or ignored entirely by tech platforms.
Figma launched a canvas-native AI agent that generates designs, executes bulk edits, and automates repetitive visual tasks using plain-language prompts — moving the AI design assistant from a sidebar or plugin into the core creative workspace.
The Workforce Backlash Gets Louder
The cultural undercurrent of this AI cycle surfaced visibly at graduation ceremonies this season. Former Google CEO Eric Schmidt was heckled at the University of Arizona after mentioning AI; elsewhere, students booed AI-boosting commencement speakers at multiple institutions. The protests aren't purely symbolic: data suggests 30% to 40% of the class of 2026 are currently unemployed, with many attributing their job-search struggles to AI-driven displacement.
Jeff Bezos pushed back in a public statement, calling AI job doomsayers "dead wrong" and arguing that generative AI will shift workers to higher-value tasks while boosting developer productivity. He also warned against early regulation that could slow the transition. John Doerr published a contrasting op-ed in the Wall Street Journal arguing that the US must invest urgently in workforce retraining alongside AI infrastructure — framing the current moment as a generational opportunity that requires active policy intervention.
AI Infrastructure Under Pressure
A Gallup poll released this week found that 70% of Americans oppose local data center construction — a significant headwind for the industry's aggressive expansion plans. Data centers are increasingly facing neighborhood opposition over energy consumption, noise, and land use, even as AI companies race to secure compute capacity.
The White House quietly approved $9 billion in classified funding for intelligence agencies to purchase Nvidia Blackwell chips, acknowledging that classified AI networks are constrained by the ongoing chip shortage. The move suggests that government AI programs are competing directly with commercial demand for frontier hardware.
Meta, meanwhile, drew attention for deploying internal software to track employee mouse movements, clicks, and screen activity. The company framed the monitoring as training data collection for AI agents; analysts interpreted it more bluntly as performance surveillance.
---
Quick Takes
Chinese video models take the lead. ByteDance's Seedance 2.0 and Kuaishou's Kling now top independent video model leaderboards, with TikTok's training data providing a significant structural advantage. Chinese open-weight models more broadly now account for 61% of top-model token consumption on OpenRouter, driven by costs approximately 9x lower than premium US models for bulk agent workloads.
Southwest bans robots on flights. Southwest Airlines prohibited humanoid and animal-like robots from cabins and checked baggage, citing lithium-ion battery fire risk following a viral incident involving a robot passenger.
Spotify licenses AI music covers. Spotify reached a deal with Universal Music Group allowing Premium subscribers to create AI-generated covers and remixes of songs, with participating artists receiving a revenue share — one of the first major label agreements to explicitly sanction fan-generated AI music.
AI companions may be backfiring. New research suggests that while AI companionship apps offer short-term emotional relief, they correlate with increased distress markers over time. The findings add complexity to the $37 billion companion AI market.
SpaceX Starship Flight 12 succeeds. Starship's twelfth flight successfully demonstrated orbital docking ports required for propellant transfer — a critical milestone for NASA's Artemis lunar program.
Colossal hatches chicks from synthetic eggs. Biotech firm Colossal Biosciences successfully hatched 26 live chicks from fully artificial eggs, advancing its de-extinction research program.
Huawei Watch adds diabetes risk scoring. The Huawei GT 6 Pro now uses up to 14 days of biometric data — including heart rate variability, sleep patterns, and activity levels — to assign users a Low, Medium, or High diabetes risk rating and prompt medical follow-up.
Data leaks persist despite AI-driven device destruction. A new Blancco report found that 38% of organizations suffered data leaks last year even as AI compliance requirements are forcing widespread device disposal, suggesting that data sanitization practices have not kept pace with disposal volumes.