2026-06-16
By Vadym · Generated with AI, curated by me
The AI infrastructure bet went institutional this week — $35 billion from Apollo and Blackstone for Broadcom’s compute platform, $1.4 billion for German humanoid robotics. At the tools layer, open-source is winning on pricing: 7.5 million developers moved to a free model-agnostic agent as subscription tools repriced toward metered usage. And inside Meta, the human cost of the AI transformation surfaced loudly.
German humanoid robotics company Neura Robotics closed a $1.4 billion Series C on June 10, backed by NVIDIA, Amazon, Qualcomm, Tether, Bosch, and Schaeffler. The round values Neura at approximately $7 billion and is described as the largest single investment into a full-stack robotics company to date. Capital will fund production capacity expansion and next-generation physical AI systems, with Neura targeting millions of deployed robots by 2030 across manufacturing, logistics, healthcare, and consumer markets. The investor syndicate reads as a vertical supply chain for humanoid robotics: NVIDIA and Qualcomm supply the silicon, Amazon provides cloud and logistics infrastructure, and Bosch and Schaeffler bring industrial manufacturing depth. When semiconductor manufacturers, cloud providers, and industrial groups co-invest in the same humanoid robotics company, it’s not a venture bet on a demo — it’s a commitment to treating physical AI as infrastructure. The $7B valuation and a path to millions of deployed units by 2030 make Neura the first humanoid robotics company being capitalized at industrial-deployment scale.
SpaceX debuted on the Nasdaq Stock Market on Friday, June 12, under ticker SPCX, opening at $150 per share — 11% above its $135 IPO price — and closing the day up 19% at approximately $160. The company raised $75 billion in the offering, making it the largest IPO in recorded history by capital raised. The implied market capitalization of approximately $1.77 trillion positions SpaceX above most public technology companies. The investment thesis extends well beyond the launch business: Starlink’s satellite internet network serves 7 million+ subscribers across 100+ countries, and AI-assisted guidance, propulsion, and satellite operations systems are core to the company’s operations. SpaceX’s $1.77T debut establishes a new market frame for how institutional capital values frontier technology infrastructure — $75 billion raised in one offering, in a week when AI-adjacent companies are also entering the public market pipeline, signals that investors are comfortable pricing trillion-dollar bets on platforms that are years from peak revenue.
On June 12, a Meta employee seized the microphone during a livestreamed internal presentation watched by thousands of colleagues to publicly criticize the company’s AI leadership. Hours later, Zuckerberg sent a company-wide memo acknowledging Meta had made “mistakes” in its AI restructuring and promised no further company-wide layoffs in 2026. The incident surfaced frustration inside Meta’s Applied AI organization — a 6,500-person unit under CTO Andrew Bosworth, created in a May overhaul that cut 8,000 jobs and shifted 7,000 more engineers into AI roles. Engineers in the unit are calling it “the gulag”: their assigned work is generating puzzles and coding problems to train AI agents on how humans complete computer tasks. They have no choice but to join or leave the company. More than 1,600 Meta employees separately signed a petition opposing a program that monitors workers’ clicks and keystrokes to generate AI training data. This is what the AI transformation looks like from inside — experienced engineers being converted into training data generators for the systems that will eventually replace them, with no credible path back to product work.
On June 9, Apollo Global Management led an initial $35 billion capital solution for Broadcom’s AI XPV Platform, partnered with Blackstone and a syndicate of global banks. The platform is designed to deliver more than 20 gigawatts of AI compute capacity using Broadcom’s custom XPUs and networking silicon — targeting frontier model training and inference through 2028. Broadcom’s XPU architecture competes with GPU-based infrastructure by offering lower cost and higher efficiency for specific workload profiles. The $35 billion financing structure brings institutional-scale private credit and equity into AI infrastructure at a size larger than most sovereign wealth fund AI commitments. The 20GW target matters in context: global data center capacity today is roughly 100GW, meaning this single platform commits to adding 20% of that in purpose-built AI compute. Power availability is already the binding constraint on AI expansion — gas turbines are booked through 2028 and at least 86 local moratoriums have been proposed across the US. Apollo and Blackstone managing this capital means AI infrastructure is no longer a technology sector asset class; it’s institutional real estate.
Today’s stories share a through-line: the cost structure of AI is being renegotiated at every layer simultaneously. At the infrastructure level, $35 billion from Apollo and Blackstone says AI compute is now institutional-grade real estate. At the hardware layer, Neura’s $1.4B round says humanoid robotics gets its own industrial funding stack. At the tools layer, both Cursor and GitHub Copilot moved away from flat pricing this month — and the 7.5 million developers on OpenCode are voting with their API keys for the alternative. And inside the companies building all of this, Meta’s revolt shows what the human cost looks like when you convert experienced software engineers into AI data generators without a credible reason or an exit path. The capital is moving. The talent is restless. The tools are repricing. The next six months will depend mostly on which infrastructure bets actually close out — and whether the power grid can keep up with the compute commitments already on paper.
— Boba
Curated by Vadym