TL;DR
Money is clustering around two AI poles: Anthropic as the second model superpower and Musk’s SpaceX/xAI stack as the would‑be GPU landlord, both financed on the assumption that compute will stay scarce and priceless. At the same time, the first big enterprise adopters are finding that AI is raising their opex and political heat faster than it is cleaning up their P&Ls.
The open question is whether this is just the messy middle of an S‑curve or an early sign that today’s AI economics and social license don’t actually pencil out.
Key Events
Report
Capital is slamming into two opposite walls at once: record‑size AI infra lock‑ups around Anthropic and SpaceX, and early enterprise deployments that are torching token budgets and jobs.
The gap between the keynote narrative of effortless AI efficiency and the first real P&Ls just got wider this month.
Anthropic is now the second AI super‑magnet: it hired Andrej Karpathy, has Claude reportedly ahead of ChatGPT in business adoption and annualized revenue, and is guiding to first profitability by Q2 2026 at around $500M. It has raised on the order of $65B from Google and Amazon and agreed a separate $30B funding deal at a $900B valuation, putting it in the same capital orbit as the hyperscalers it runs on.
On the infra side it is locked into paying SpaceX about $1.25B a month for compute through 2029, roughly $45B in total, effectively pre‑booking a slice of the global GPU supply chain.
Downstream, large customers are clustering around it: Salesforce is planning about $300M of Anthropic token spend this year with AI already handling 30–50 percent of internal workload, and Uber has put $3.4B into an Anthropic‑based AI initiative.
Cloudflare has wired Claude Managed Agents into its autonomous code‑delivery system and Anthropic just acquired Stainless, a dev platform used by OpenAI, Google and Cloudflare, deepening its grip on the services and tooling layer as Nvidia systems climb toward $7.8M per AI build with memory costs up 485 percent and AMD’s MI355 prices roughly 40 percent below Nvidia’s B200 for single‑node serving.
SpaceX is coming public around a $1.75T valuation while targeting up to $75B in IPO proceeds. The underlying business printed about $18B of revenue and a $5B operating loss, so investors are effectively underwriting future AI and connectivity cash flows rather than current profitability.
Management pitches a $28.5T total addressable market with $1.6T attributed just to Starlink connectivity. Starlink already generates roughly $11B and more than half of SpaceX’s revenue, helped by airline deals and rural broadband, but it is drawing FCC concern over potential dominance.
On the AI side, Anthropic’s agreement to pay about $1.25B a month for compute through 2029, roughly $45B in total, turns SpaceX into one of the largest external GPU landlords overnight while xAI is reportedly burning around $1B a month with GPU utilization near 11 percent across some 500,000 GPUs.
Early large‑scale deployments show how messy AI economics look before governance catches up: Salesforce says AI now handles 30–50 percent of its workload, is cutting 4,000 customer‑service roles, has paused new software‑engineer hiring, and plans about $300M of Anthropic token spend this year.
Customers report paying 83 percent more than last year as they drop tools like Notion, yet describe Salesforce’s AI features as low‑quality and workflow‑breaking rather than simplifying.
Uber put $3.4B into an Anthropic‑based AI initiative and then blew through its entire 2026 AI token budget in four months, with the CTO admitting that 73 percent of reads were redundant.
Intuit is laying off more than 3,000 people, around 17 percent of its workforce, in an AI‑framed restructuring that many former employees say they do not trust or understand.
Across the sector over 110,000 tech jobs have been cut in five months tied to AI adoption, while frontline reports from Amazon describe AI tools being pushed into workflows in ways that add busywork rather than value.
Google, Microsoft, and Meta are all pivoting core products to AI‑first monetization while visibly straining user trust and regulatory patience.
Google has overhauled its 25‑year‑old search box, is injecting ads into AI Mode results, and is pushing Gemini 3.5 Flash and Gemini Omni as the default interface even as users complain about declining relevance, ad bloat, and a bug that deleted a large customer’s account affecting 647,000 users.
Microsoft is rolling out Copilot Cowork and Azure Linux 4.0 while forcing a shift from SMS codes to passkeys, and its AI chief is talking about full white‑collar automation within 18 months amid user fears about invasive Copilot behavior, alleged BitLocker backdoors, and long‑standing reliability problems.
Meta is getting a $3.3B tax break for a $10B Louisiana data center and posted $26B in Q1 net income while still firing about 8,000 people and forcibly reassigning 7,000 into AI roles, all while facing censorship accusations and rolling back end‑to‑end encryption for Instagram DMs as it markets private AI chats.
At the same time regulators are rewriting rules: the White House will require pre‑launch safety checks for frontier models from OpenAI and Anthropic, the U.S. DOJ has ordered Apple and Google to unmask over 100,000 users of a car‑tinkering app, and the EU’s top court says Meta must pay Italian publishers for content use, while Gen Z audiences are literally booing Eric Schmidt’s AI boosterism at graduations.
What This Means
Capital and regulation are both concentrating around a tiny set of AI labs and infra providers at the exact moment that enterprises are discovering how hard it is to turn models into clean margin. The live decision is whether to treat this as a temporary digestion phase on the way to real productivity, or as evidence that AI economics and legitimacy will stay structurally volatile for longer than the hype cycle implies.
On Watch
Interesting
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
Sources
Key Events
On Watch
Interesting