AI INFRASTRUCTURE

The trillion-dollar AI infrastructure arms race

The numbers from Thursday's White House tech dinner were so large they bordered on absurd. When President Trump went around the table asking each CEO how much they planned to invest in America, Mark Zuckerberg committed to "something like at least $600 billion" through 2028. Apple's Tim Cook matched that figure. Google's Sundar Pichai said $250 billion.

Combined with OpenAI's revised projection this week that it will burn through $115 billion by 2029 — $80 billion more than previously expected — these announcements reveal an industry in the midst of the most expensive infrastructure buildout in modern history.

The scale has reshaped the entire American economy. AI data center spending now approaches 2% of total U.S. GDP, and Renaissance Macro Research found that so far in 2025, AI capital expenditure has contributed more to GDP growth than all U.S. consumer spending combined — the first time this has ever occurred.

What's driving this isn't just ambition but desperation to control costs:

• OpenAI has become one of the world's largest cloud renters, with computing expenses projected to exceed $150 billion from 2025-2030

• The company's cash burn projections quadrupled for 2028, jumping from $11 billion to $45 billion, largely due to costly "false starts and do-overs" in AI training

• Meta's 2025 capital expenditures represent a 68% increase from 2024 levels as it races to build its own infrastructure

• McKinsey estimates the global AI infrastructure buildout could cost $5.2 to $7.9 trillion through 2030

The 33 attendees included the biggest names in tech: Microsoft founder Bill Gates, Google CEO Sundar Pichai, OpenAI's Sam Altman and Greg Brockman, Oracle's Safra Catz, and Scale AI founder Alexandr Wang. Notably absent was Elon Musk, who claimed on social media he was invited but couldn't attend amid his ongoing feud with Trump.

The moment was captured on a hot mic when Zuckerberg later told Trump, "I wasn't sure what number you wanted," though whether this reflected genuine uncertainty or strategic positioning remains unclear.

Zuckerberg's hot mic moment about not knowing "what number you wanted" suggests that these commitments aren't emerging from detailed financial models, but rather from competitive dynamics and political positioning.

The scale of spending has reached a point where traditional ROI calculations seem almost irrelevant. OpenAI's projections jumping from $11 billion to $45 billion for 2028 alone show how unpredictable this technology remains, even for the companies building it. The acknowledgment of "false starts and do-overs" in AI training reveals an industry still figuring out the fundamentals.

What's most striking is how this infrastructure race has become an economic force unto itself. When AI capital expenditure drives more GDP growth than consumer spending, we're seeing the emergence of an economy increasingly powered by speculative technology investment rather than traditional economic activity. Whether this produces the transformative returns these companies are betting on remains the trillion-dollar question.

TOGETHER WITH IBM

Agentic AI for COOs: Not tomorrow’s promise - Today’s reality

IBM surveyed 1,920 Chief Operating and Chief Supply Chain Officers to discover what it takes to thrive in a market upended by rapid tech shifts and heightened expectations for productivity and growth. 

Our study found that 7 out of 10 surveyed executives have adopted AI agents and are preparing to scale them across their enterprise. 

Making AI an ally in developing a unified, strategic, smarter mindset to transform your organization operations. 

Get the full report to learn 5 key strategies to shatter enterprise inertia.

ROBOTICS

Boston Dynamics, Toyota harness large behavior models to power humanoids

Boston Dynamics and Toyota Research Institute have announced a significant stride in robotics and AI research. Demonstrating how a large behavior model powers the Atlas humanoid robot. 

The team released a video of Atlas completing a long, continuous sequence of complex tasks that combine movement and object manipulation. Thanks to LBMs, the humanoid learned these skills quickly, a process that previously would have required hand programming but now can be done without writing new code. 

The video shows Atlas using whole-body movements walking, lifting and crouching while completing a series of packing, sorting and organizing tasks. Throughout the series, researchers added unexpected physical challenges mid-task, requiring the humanoid to self-adjust. 

It’s all a direct result of Boston Dynamics and the Toyota Research Institute joining forces last October to accelerate the development of humanoid robots. 

Scott Kuindersma, vice president of Robotics Research at Boston Dynamics, said the work the company is doing with TRI shows just a glimpse of how they are thinking about building general-purpose humanoid robots that will transform how we live and work. 

“Training a single neural network to perform many long-horizon manipulation tasks will lead to better generalization, and highly capable robots like Atlas present the fewest barriers to data collection for tasks requiring whole-body precision, dexterity and strength,” Kuindersma said. 

Russ Tedrake, senior vice president of Large Behavior Models at Toyota Research Institute, said one of the main value propositions of humanoids is that they can achieve a vast variety of tasks directly in existing environments, but previous approaches to programming these tasks could not scale to meet this challenge. 

“Large behavior models address this opportunity in a fundamentally new way – skills are added quickly via demonstrations from humans, and as the LBMs get stronger, they require less and less demonstrations to achieve more and more robust behaviors,” he said.

Kuindersma and Tedrake are co-leading the project to explore how large behavior models can advance humanoid robotics, from whole-body control to dynamic manipulation.

TOGETHER WITH TRIPLE WHALE

Turn AI Into Your BFCM Advantage

This BFCM, AI isn’t optional — it’s the edge that separates the winners from the rest. It helps you plan smarter, optimize faster, and respond to shifting performance in real time.

Triple Whale’s Ultimate BFCM Prep Guide delivers benchmarks, insights, and AI-powered workflows to guide every decision.

Use it to scale efficiently and set new records this BFCM.

Download your free BFCM Prep Guide now

WORKPLACE

OpenAI is developing an AI jobs platform

OpenAI is building its own jobs platform to compete directly with LinkedIn, launching a certification program designed to train 10 million Americans in AI skills by 2030.

The OpenAI Jobs Platform, slated to launch in mid-2026, will utilize AI to pair candidates with employers seeking AI-skilled workers. This is part of a broader effort to transform how people learn and work with AI.

The company is expanding its OpenAI Academy with certifications ranging from basic AI literacy to advanced prompt engineering. The twist? Students can prepare entirely within ChatGPT using its Study mode, which turns the chatbot into a teacher that questions and provides feedback rather than giving direct answers.

Major employers are already signing up:

• Walmart is integrating the certifications into its own academy for 3.5 million U.S. associates

• John Deere, Boston Consulting Group, Accenture and Indeed are launch partners

• The Texas Association of Business plans to connect thousands of employers with AI-trained talent

Certification pilots begin in late 2025, with OpenAI committing to certify 10 million Americans by 2030 as part of the White House's AI literacy campaign.

The initiative comes as companies increasingly seek workers with AI skills, with research showing that AI-savvy employees earn higher salaries on average. OpenAI CEO of Applications Fidji Simo acknowledged AI's "disruptive" impact on the workforce, saying the company can't eliminate that disruption but can help people become more fluent in AI and connect them with employers who need those skills.

LINKS

• This Billionaire’s AI was supposed to speed up policing. It’s not going well.

• Anthropic reaches $1.5b author copyright settlement

• Scale AI’s former CTO launches new AI agent

• Enterprise AI adoption trends downward, report

• Authors sue Apple over using books in AI training

• Meta’s Mark Zuckerberg caught in hot mic moment with Trump

• What to expect from Apple's Sept. 9 event 

• Study explores potential impact of an AI bubble

• AI anxiety hits workers at every career stage

• Are bad incentives to blame for AI hallucinations?

• AWS & Anthropic’s multi-gigawatt trainium expansion

• ASML becomes Mistral AI’s top shareholder after leading latest funding round

• Dell Pro Max GB10: AI developer PC with NVIDIA stack, Blackwell chip, and 1 petaflop of FP4 compute*

• Monologue: A new voice-to-text tool that allows you to type at the speed of talk (I just switched over from Whispr!)

• Katalog: Save articles for later and listen to them with high-quality AI narration

• CapCut AI Suite: Create, edit, or remix content with AI in a simple editor

• PhotoFox AI: Turn any product photo into 100+ on-brand assets in minutes

• Bosch USA: AI Research Scientist, GenAI

• Figure: AI Training Infrastructure Engineer

• Nvidia: Senior Software Engineer, Simulation and Virtualization

• Drillo.AI: Data scientist

• Uber: Senior Data Scientist, Experimentation

GAMES

A QUICK POLL BEFORE YOU GO