Scale & Strategy
together with
This is Scale & Strategy, the newsletter serving up “oooh, I get it” moments, daily.
Here’s what we’ve got for you today:
- Google and SpaceX inch toward orbital AI infrastructure
- The gains are real. So is the invoice.
- Celonis thinks enterprise AI’s real problem is context, not intelligence
Google and SpaceX inch toward orbital AI infrastructure
The AI compute race is getting weird fast.
Google is reportedly exploring a launch partnership with SpaceX to support orbital data center infrastructure, pushing the idea of space-based AI compute from sci-fi side quest into something major tech companies are actively preparing for.
And no, this isn’t just a random collaboration between two giants with too much capital and insufficient hobbies.
Google already owns a 6.1% stake in SpaceX from its $900M investment back in 2015, and Google VP Don Harrison still sits on SpaceX’s board. Now the relationship appears to be evolving from strategic investment into infrastructure alignment.
At the center of this is Google’s internal “Project Suncatcher,” a moonshot effort aimed at deploying prototype compute satellites by 2027. Satellite imaging company Planet Labs is reportedly helping build the first hardware.
Meanwhile, SpaceX is making orbital compute a major part of its long-term infrastructure pitch ahead of its eventual IPO. The company has already filed approvals for up to one million satellites. Which is either the foundation of a planetary compute network or the opening chapter of humanity accidentally recreating traffic congestion in low Earth orbit.
The timing also lines up with another recent development: Anthropic finalized a compute partnership with SpaceX last week and reportedly expressed interest in securing multiple gigawatts of orbital AI compute capacity.
That’s the important shift here.
This isn’t being framed as experimental research anymore. Frontier AI labs are actively looking for future compute expansion paths beyond terrestrial data centers because everyone sees the same wall coming: power, cooling, permitting, and grid constraints are becoming bottlenecks faster than model demand is slowing down.
Space changes some of that math.
Orbital systems potentially offer near-limitless solar energy, passive cooling, and physical scaling unconstrained by local infrastructure politics. The tradeoff, obviously, is that you now have to launch your servers into orbit and somehow maintain them without turning the atmosphere into an Nvidia graveyard.
OpenAI CEO Sam Altman publicly dismissed the concept recently in New Delhi, calling orbital compute “ridiculous” and saying it won’t matter at scale this decade.
Maybe he’s right. The economics are brutal today.
But it’s also worth remembering that many ideas involving reusable rockets, satellite internet, EV dominance, or private space infrastructure sounded ridiculous right before Elon Musk made them deeply annoying to bet against.
The strategic logic for both sides is pretty straightforward:
- Google gets launch infrastructure without building rockets
- SpaceX gets validation from one of the world’s largest AI and cloud players
- Both companies position themselves early if terrestrial compute constraints worsen dramatically later this decade
And there’s another layer underneath all this.
Google and SpaceX may eventually become rivals in AI infrastructure even while cooperating today. SpaceX increasingly looks like it wants to evolve from aerospace company into vertically integrated compute and connectivity infrastructure provider. Google already dominates cloud infrastructure. Their incentives overlap now because the market is still expanding fast enough for everyone to sprint in the same direction.
That tends not to last forever.
Still, the bigger signal here is that frontier AI infrastructure conversations are no longer confined to bigger data centers, nuclear power deals, and GPU supply chains. The industry is starting to think in planetary-scale terms now. Which is either visionary long-term planning or the natural endpoint of an industry that looked at “more servers” and somehow arrived at “space.”
Risk and regulations are ramping up, and customers expect proof of security just to do business. That’s why Vanta is a game changer.
Vanta automates your compliance and security processes in one AI-powered platform.
Whether you’re prepping for SOC 2, or running an enterprise GRC program, Vanta keeps you secure and your deals moving.
Companies like Ramp and Writer spend 82% less time on audits with Vanta. Faster compliance—and more time for growth.
Check out the on-demand demo to learn more. – trusted by top startups like Cursor, Linear, and Replit.
The gains are real. So is the invoice.
SoftBank’s OpenAI bet is starting to look brilliant on paper. The problem is the paper is financed with mountains of debt. Human civilization’s favorite game: lever up into a euphoric private market and pray the music keeps playing long enough for an IPO.
SoftBank is expected to post roughly $1.5B in net profit for the January-to-March quarter, driven largely by the exploding value of its OpenAI stake. After OpenAI’s latest raise reportedly pushed its valuation to around $840B, TD Cowen estimates SoftBank’s 11% position is now worth roughly $80B, up from $54B just three months earlier.
That kind of mark-up changes the mood fast. Investors love visionaries right up until the refinancing starts.
SoftBank is reportedly preparing to pour another $30B into OpenAI next year, doubling down on what has effectively become the centerpiece of Masayoshi Son’s AI thesis. If OpenAI manages to IPO in late 2026 or early 2027, the upside could get even more absurd.
But underneath the gains, the balance sheet is starting to groan.
In March, SoftBank secured a $40B bridge loan tied to its OpenAI investment plans, while S&P shifted the company’s credit outlook to negative, warning that its financial flexibility could deteriorate. Translation: the market likes the AI exposure, but lenders are starting to count the chairs in the room.
The concentration risk is also becoming impossible to ignore. More and more of SoftBank’s future is effectively tethered to one private company still operating in the most capital-intensive arms race the tech industry has ever seen.
That naturally brings back memories of WeWork, the ghost permanently haunting every SoftBank earnings call. The difference this time is OpenAI actually has real demand, real revenue growth, and a product people use daily instead of a kombucha dispenser masquerading as infrastructure.
For now, investors are willing to ride the momentum. SoftBank shares have nearly doubled since April and are hovering near record highs. As long as OpenAI keeps appreciating faster than SoftBank accumulates debt, the trade works.
The second that reverses, things could get very interesting very quickly.
Celonis thinks enterprise AI’s real problem is context, not intelligence
Most enterprise AI projects fail for the same reason: the model has no idea how the business actually works.
It can write emails, summarize documents, and generate dashboards all day long. But ask it how procurement approvals interact with inventory risk, customer churn, internal policies, and supply chain delays inside your specific organization, and the whole thing starts hallucinating confidence. Corporate theater powered by tokens.
Celonis is trying to solve that problem with a new system called the Celonis Context Model, or CCM, which essentially acts as a real-time operational map of a business.
The company unveiled CCM this week alongside its acquisition of Ikigai Labs, a MIT-born startup focused on prediction, forecasting, and scenario simulation.
Together, the move signals where enterprise AI is heading next: away from generic copilots and toward systems that deeply understand how a specific business operates internally.
The core idea behind CCM is pretty straightforward.
Instead of feeding AI disconnected chunks of company data, Celonis builds a continuously updated context layer that pulls together structured and unstructured information across systems, logs, workflows, policies, screens, and operational processes.
That creates a shared operational language the AI can reason over.
Celonis co-CEO Alex Rinke put it bluntly: enterprise AI only becomes valuable once it specializes in the business itself.
“You need to teach it how to be good at doing a specific job in that business.”
That sounds obvious, but most enterprise deployments still skip this layer almost entirely. Companies plug frontier models into fragmented systems and expect meaningful automation to magically emerge from organizational chaos.
Usually what emerges instead is a very expensive summarization engine.
What makes the Ikigai acquisition interesting is that it pushes the system beyond real-time visibility into predictive reasoning.
According to Rinke, Ikigai’s modeling systems allow Celonis to move from understanding what’s happening now toward forecasting what could happen next. That includes scenario planning, operational forecasting, pricing simulations, and future-state modeling.
So instead of an AI agent simply processing workflows faster, the system could theoretically evaluate downstream consequences before decisions are made.
That’s a much bigger category.
Think:
- Dynamic pricing decisions
- Supply chain rerouting
- Forecasting procurement bottlenecks
- Predicting customer behavior shifts
- Simulating operational tradeoffs before execution
And unlike a lot of enterprise AI marketing, this actually targets one of the hardest problems in large organizations: fragmented operational context.
Celonis says customers including AstraZeneca, Mercedes-Benz, and Domino Foods saw major improvements in task completion, efficiency, and accuracy when AI agents operated on top of the context layer.
Which makes sense.
Smarter models alone were never enough. Frontier labs spent the last two years racing to improve reasoning, coding, and multimodal capabilities while enterprises quietly discovered the real bottleneck was organizational understanding. AI can’t automate workflows it fundamentally doesn’t comprehend.
In a way, CCM is less a brand-new invention and more the natural evolution of Celonis’ existing process intelligence business. The company already built its reputation helping enterprises map and optimize operational flows. This just turns that operational map into something AI-native.
The timing also feels important.
The market is slowly realizing that generic copilots are becoming commoditized. The durable advantage likely shifts toward whoever owns the deepest operational context inside enterprises. Models matter, but context increasingly determines usefulness.
That’s why every major AI vendor is suddenly obsessed with connectors, memory, workflows, retrieval layers, MCP infrastructure, and enterprise integrations. They’re all chasing the same thing: situational awareness.
Because in enterprise AI, intelligence without context is mostly just very articulate confusion.
Was this email forwarded to you?
That’s it for today and as always It would mean the world to us if you help us grow and share this newsletter with other operators.
Our mission is to help as many business operators as possible, and we would love for you to help us with that mission!
