Siemens Mobility has completed its acquisition of key businesses of Italian rail technology group MERMEC, strengthening its global railway diagnostics, asset intelligence and signaling portfolio. https://t.co/RifWUTeZWq
Anthropic’s updated S-1/A IPO filing gives us the clearest look yet at the real economics of frontier AI. And it is a massive validation for Google Cloud and custom silicon.
Out of Anthropic's $895.4B in total long-term compute commitments, Google’s TPU architecture sits right at the center:
• $121.3B minimum commitment to Google Cloud (April 2026 to July 2033). It is non-cancelable with true-up provisions. Anthropic pays regardless of actual usage.
• $125B to $150B in estimated additional payments to Google for extra TPU capacity between 2028 and 2031.
• $161.2B in 5-year master equipment leases with Broadcom specifically for TPU-based compute infrastructure in partner facilities.
That is over $400B of frontier AI compute anchored around Google’s TPU ecosystem.
Compare that structure to the rest of the filing:
• Amazon (AWS): $110.0B minimum over a longer 10-year window (through April 2036).
• SpaceX AI Unit: Up to $120.5B for Nvidia-based compute (cancelable on 90 days' notice after early 2027).
• Microsoft (Azure): $44.9B minimum through June 2033.
Why does this matter so much for Google?
Silicon Margin Capture When a cloud provider rents out third-party GPUs, Nvidia takes the silicon margin. When Google rents out TPUs, it captures the economics across both the custom silicon and the cloud stack.
Capex De-Risking Critics have questioned whether hyperscaler AI capex will ever see a real return. Locking in a $121.3B take-or-pay backlog, plus a potential $125B to $150B expansion, gives Google multi-year visibility to underwrite data center, power, and chip R&D investments.
Breaking the GPU Monopoly A top frontier lab isn't just using custom chips for minor inference tasks. It is betting hundreds of billions of dollars of its core roadmap through 2033 on TPUs.
Yes, there is reciprocal financing across the industry. Google, Amazon, and Microsoft are both major equity investors and compute suppliers. And Anthropic still has to execute to service a nearly $900B multi-year bill.
But one takeaway from the S-1/A is undeniable: Google's decade-long bet on custom AI chips is now translating into locked-in, multi-decade cloud demand.
$Google solana:7yBMbdLB26sgvC1NgJHgKs3fmuG65vHhPtTinvXtRnjq
‼️ Proton CEO Andy Yen called a staffer's plea to pentest new code before release "bullshit and loser thinking".
Yen wrote that workers should prompt “better.”
Proton's CTO told staff AI wrote 90% of its code in October, much of it with Anthropic's Claude.
An Accenture contractor allegedly connected a personal Anthropic Claude Code agent to a secure Pentagon network, per WIRED.
It ran for two months and reportedly sent credentials and a network map to a high-risk third-party site.
Accenture disputes the allegations.
A French appeals court rejected on Thursday an attempt by Meta to overturn an order to negotiate in good faith with French media over rights payments for information used on its social networks.
https://t.co/vjD0l5IQMA
É bom ouvir pessoas de esquerda a dizer isto. Pode ser que finalmente se possa falar disto racionalmente… Alerto para isto há muito tempo e sempre que o fiz vinham acusar-me de estar a promover uma guerra entre gerações…
IPO or die. ‘Bold but sane’: What we learned from Anthropic’s 240,000-word IPO filing. The prospectus details the AI lab’s ‘mission’, $800 billion data center commitments, and an extraordinary governance structure.
https://t.co/resDSucxiw