TechLook LogoTechLook
LoginSign Up

SpaceX's $70B Compute Deal — Plus New York Freezes the Boom — Techlook Daily, June 8, 2026

SIsivaguru·
SpaceX's $70B Compute Deal — Plus New York Freezes the Boom — Techlook Daily, June 8, 2026
✨Summarize with AI
ChatGPTPerplexityClaudeGrokGoogle AI

Two stories landed this weekend that should reframe how every founder plans the next 18 months. SpaceX quietly became the largest AI compute landlord in the Western hemisphere, locking in more than $70 billion of cloud contracts days before a record-setting $75 billion IPO filing. On the same day, New York's legislature passed the first statewide freeze on new data centers — a one-year moratorium aimed at hyperscale builds. The AI capex story is still accelerating, but the politics of where it lands, and who profits, are catching up fast. The rest of this week's news — bots outnumbering humans online, a $1 trillion chip rout, OpenAI's superapp play, and Sriram Krishnan's exit from the White House — sits in the same arc.

If you build for a living, today is a good day to reread our May 22 piece on SpaceX's $1.7T IPO and its $1.25B monthly compute dependency — and to remember, as we wrote back in May, that the AI compute war is now about power, not just GPUs.

SpaceX Just Became the Biggest AI Landlord on Earth

Three days before its IPO filing, SpaceX signed a multi-year cloud deal with Google that dwarfs anything in the consumer-internet era: $920 million a month, from October 2026 through June 2029, for roughly 110,000 Nvidia GPUs, per a June 5 CNBC report on the S-1 filing. A parallel deal with Anthropic — announced May 6 on xAI's newsroom and covered by CNBC the same day — already consumed every processor at the Colossus 1 site in Memphis, over 220,000 chips and roughly 300 MW of capacity, at a reported $1.25B per month through May 2029.

Together the two contracts total more than $70 billion over the term — about $26 billion a year — against a company whose existing revenue is a small fraction of that. The target IPO raise is $75B, with Reuters reporting a planned share price of $135 that would value SpaceX at $1.77T.

Here's everything you need to know:

  • Google pays $920M/month from October 2026 through June 2029 for ~110,000 Nvidia GPUs
  • Anthropic's deal consumed the full 220,000+ processor capacity of SpaceX's Colossus 1 site in Memphis, at $1.25B/month through May 2029
  • Combined contracts are worth $70B+ over the term — about $26B/year
  • SpaceX is targeting a $75B raise at a $135 share price, valuing the company at ~$1.77T
  • The pricing model is unusual: SpaceX is selling compute as a bundled story (launch + power + Nvidia allocation), not as a commodity cloud product
  • This is a direct play on the gigawatt-constrained bottleneck — most of the deal value is the right to plug into power and space, not the chips themselves

This is not a cloud business. It is a real-estate business with a launchpad attached. SpaceX owns the power purchase agreements, the grid interconnects, and the regulatory relationships that take most data-center developers three to five years to build. Founders building AI products should be planning around the assumption that the cheapest compute in 2027 will not come from AWS or Azure — it will come from SpaceX-flavored capacity, priced by the gigawatt and bundled with launch rights. The teams that lock in 2026 capacity now will be in a fundamentally different cost position than the ones waiting for hyperscaler prices to fall.

New York Froze the Data Centers — For a Year

New York's legislature passed a one-year moratorium on new large data centers last week, the first statewide ban of its kind, reported Politico affiliate Penn Capital-Star and E&E News. The bill, S9144A on the New York Senate site, pauses permits for any data center drawing 20+ megawatts and tasks the state Department of Environmental Conservation with a study of electricity, water, and land use during the freeze. Post-freeze, builders must fund a public hearing three months before any approval. Governor Kathy Hochul has not yet signed the bill into law, with signing expected later this year.

Here's everything you need to know:

  • One-year pause on approvals for any data center drawing 20+ MW
  • State environmental agency must report on electricity, water, and land impact during the freeze
  • Post-freeze, a mandatory three-month public hearing is required before approval
  • Awaiting Governor Hochul's signature
  • The 20 MW threshold captures nearly every frontier-scale AI training facility
  • Aimed at hyperscalers, but mid-size builders will get caught in the same paperwork

This is the first real signal that U.S. states can move faster than the federal government on AI infrastructure. A 20 MW site is roughly the floor for a serious training cluster — under that you are doing fine-tuning, not pretraining. The "why" matters more than the "what": New York is not killing data centers, it is buying twelve months for the political backlash to organize into durable zoning law. Expect California, Massachusetts, and Virginia to watch the rollout closely. If New York's report lands before November, expect copycat legislation in three to five states by Q1 2027.

The deeper story is structural: the same week SpaceX is selling gigawatts of new capacity, a major U.S. state is closing the permitting door on the next 12 months of new builds. Founders planning physical infrastructure should treat this as a leading indicator, not a one-off — which is exactly the dynamic we explored in The AI Compute War Is Now About Power, Not Just GPUs.

The U.S. Government Wants a Piece of OpenAI

The White House and OpenAI are in active talks over a U.S. government equity stake — reportedly 1% to 5%, potentially feeding a "Public Wealth Fund" so Americans share in AI upside. CEO Sam Altman met with both Bernie Sanders and Trump officials last week. Trump described the idea as "a partnership with the American public… and that would be a beautiful thing." Former AI czar David Sacks came out against it, warning of "corporate-government fusion."

Here's everything you need to know:

This is a tax-and-ownership question, not a regulatory one. The honest read: 1%–5% is a rounding error on OpenAI's cap table, but it is the precedent that matters. Once the federal government owns equity in one frontier lab, every future crisis — safety, layoffs, outages, IP disputes — becomes a direct conflict of interest for every regulator touching the file. Founders building on top of OpenAI's APIs should not care about this today. Founders building a defense-in-depth stack across providers — Claude, Gemini, in-house models — should care a lot about it by the time the next administration takes office.

Supabase Hit $10.5B — Fueled by Claude Code, Codex, and Cursor

Supabase raised $500 million at a $10.5 billion valuation, announced June 4 and reported by CNBC, with TechCrunch noting the round was "explicitly attributed" to demand from AI coding tools. The Series F, led by GIC with Accel, Felicis, and Y Combinator participating, per Axios Pro, effectively doubles the company's valuation in eight months.

Here's everything you need to know:

  • $500M raise at a $10.5B post-money valuation — double the $5B Series E from late 2025
  • Demand is driven by Claude Code, Codex, and Cursor pipelines building on Supabase
  • The "Postgres + realtime + auth + storage" stack maps cleanly to what AI agents can wire together
  • Supabase is now one of the few dev-tool companies priced like a front-tier cloud platform
  • Implies a roughly 50x step-up from earlier rounds

This is the most important read for SaaS builders: the database layer is back to being the most valuable real estate in software, because AI agents are spinning up backends faster than they spin up frontends. The next wave of "agent-friendly infrastructure" companies is being repriced right now, and Postgres-shaped products are the first ones. If your startup touches data, auth, storage, queues, or any of the adjacent primitives, the same repricing is coming for you.

For more on the agent-coded-everything wave, see Claude Wrote 80% of Its Own Code Last Month and Anthropic Quietly Bought the Keys to the Kingdom on adjacent developer-tool shifts.

Bots Now Outnumber Humans on the Web — 18 Months Early

Cloudflare Radar shows machines now generate 57.5% of HTTP requests to HTML content, with humans at 42.5% — a figure NPR-affiliate NBC News also reported on June 5, 2026. Cloudflare CEO Matthew Prince had forecast the crossover for end of 2027; it landed eighteen months early. The driver is agentic AI: a human shopping for a camera visits five sites, the agent doing it visits 5,000. Agentic traffic grew roughly 8,000% from early 2025 to year-end.

Here's everything you need to know:

This is the most underrated story in tech today. The assumption baked into almost every B2B SaaS funnel — that traffic implies humans, that humans can be retargeted, that a session equals a person — is now wrong on the majority of HTTP traffic. The immediate effect is on analytics vendors, ad networks, and content sites whose CPMs depend on human eyeballs. The deeper effect is on every startup whose growth model assumes a human is on the other end of the request. If your product is hit mostly by agents, your CAC math, your conversion funnels, and your telemetry all need to be re-modeled from scratch.

The $1 Trillion Chip Selloff

Underwhelming chipmaker earnings and fading rate-cut expectations wiped more than $1 trillion off U.S. equity values last week. Broadcom, Marvell, and Micron all dropped double digits — Micron alone lost $127B in market cap, down 11%. The selloff snapped a multi-week S&P 500 winning streak.

Here's everything you need to know:

For founders: the AI capex story is not in trouble, but the assumption that every dollar deployed produces 4x returns is being tested in public. The companies paying for the chips (the hyperscalers, SpaceX, the sovereign AI clouds) are still buying. The question is what the marginal buyer pays. If you are raising on a "we are riding the AI capex wave" pitch, your deck just got harder. If you are raising on a "we make the capex wave work" pitch, your deck just got easier.

Sriram Krishnan Is Leaving the White House

Sriram Krishnan is stepping down from his role as White House AI advisor at the end of June, Reuters and The Information confirmed on June 6. The exit removes a key bridge between Silicon Valley and federal AI policy at exactly the moment the OpenAI equity-stake talks are heating up.

Here's everything you need to know:

  • Krishnan was one of the most connected figures in the administration's AI policy shop
  • His exit lands during the OpenAI equity-stake negotiation
  • Reopens a position that will shape how the White House handles the next 18 months of AI policy
  • Signals the administration is still recalibrating on AI, not settling in

Worth tracking but not actionable yet. The replacement matters more than the departure — and the timing, sandwiched between a $70B compute deal and a government-equity negotiation, is hard to call a coincidence.

⚡ Quick Hits

  • OpenAI superapp: The company is reportedly planning its biggest ChatGPT revamp yet — an agent-and-coding "superapp" that prioritizes paid conversion ahead of its IPO. Codex users grew 6x to over 5M since February, and a senior OAI member told the FT "Chat is dead."
  • OpenAI Lockdown Mode: New setting disables live browsing, agent mode, and deep research to reduce prompt-injection exposure. Rolling out to ChatGPT Business and some personal accounts. Risk reduction, not a clean fix.
  • Anthropic Mythos: The model started showing up in Dev Mode and across social media this week, with public release expected imminently. (See our May 30 piece on the Erdos-proof results for context.)
  • Anthropic chip build: Anthropic poached OpenAI chip designer Clive Chen — an early hire on OpenAI's chip team — adding to the rumor mill that Anthropic is building in-house silicon.
  • Claude Code "Looping": The creator of Claude Code says he no longer prompts the agent directly — he uses a looping feature that lets the AI prompt itself. Power users are now delegating the prompt itself. (See our June 5 piece on Claude writing 80% of its own code for the full picture.)
  • Anthropic doubled Claude Cowork usage for paid plans through July 5. Cheapest way to test heavy agentic workloads this month.
  • Nvidia + LG: Joint work on humanoid robotics, motor systems, and data center architecture. The "physical AI" stack is consolidating fast.
  • Vibe coding's distribution wall: AI now builds apps in hours, but overall app usage has fallen since 2025. When code becomes abundant, attention becomes scarce.
  • Apple "top secret" AI meeting (early 2025): A previously unreported session led to Tim Cook personally taking over the AI overhaul. A reminder that the public Siri story is the trailing edge of a longer internal one.
  • Gemini Live: Real-time AI edits through your phone camera — point, prompt, edit. Visual agents are now shipping in consumer products.
  • Ideogram 4.0: 9.3B open-weight image model generating native 2K images, already topping leaderboards.
  • Nvidia Nemotron 3 Ultra: 550B-parameter open-weight model under permissive license, scoring 47.7 on the Artificial Analysis Intelligence Index.
  • AWS Summit NYC: June 17 — 200+ technical sessions, with a keynote from VP of Agentic AI Dr. Swami Sivasubramanian.

FAQ

Has the New York data center moratorium been signed into law? Not yet. The legislature passed the bill on June 5, and it awaits Governor Hochul's signature. Once signed, the 12-month pause on permits for any 20+ MW data center takes effect immediately. Until then, the framework is best understood as a proposed moratorium.

Who is reporting the $70 billion SpaceX compute deal figure? The $70B total is reconstructed from the two largest disclosed contracts: Google's $920M/month deal reported by CNBC on June 5 ($11.04B/year over 32 months ≈ $35.2B) and Anthropic's $1.25B/month deal reported by TechCrunch on May 20 ($15B/year through May 2029 ≈ $45B). The $70B headline is the rough sum of these and adjacent infrastructure agreements cited in the S-1 and IPO filings.

Does the 20 MW threshold apply to existing data centers? No. The moratorium, per the New York Senate bill S9144A, only halts new permit issuance. Operating facilities, projects already permitted before the freeze, and expansions of existing builds are not affected. The 20 MW cutoff, however, is broad: it captures essentially every frontier-scale training cluster.

When does the SpaceX IPO price? Reuters reported on June 3 that SpaceX is targeting a $135 share price and a $75B raise. Law360's coverage notes that on a typical schedule, the offering could price by mid-to-late June 2026.

How are AI agents driving the bot-traffic crossover? Per Cloudflare and HUMAN Security data, agentic browsers — bots that actually shop, add to cart, and complete transactions on a human's behalf — grew roughly 8,000% in 2025. A human comparing cameras visits five sites; an agent doing the same task may visit thousands. The math is no longer "humans + a few bots." The web is now parser-first by traffic share.

Techlook — AI & tech signal for founders and builders.

AI IndustryDev & Infrastructure

Related Posts