Together AI Just Raised $800M on $1.15B+ Annual Bookings: Here's Where the Real AI Money Flows (Spoiler: Not App Layer)
Together AI announced an $800 million Series C led by Saudi Arabia's Aramco Ventures, with NVIDIA, Vista Equity Partners, and General Catalyst also participating. The company's run-rate annual bookings topped $1.15 billion in Q2 2026. That implies an $8–10 billion+ valuation. In 36 months, Together went from nonexistent to a $10B venture-backed company with infrastructure revenue.
It's also not an app. Not a SaaS wrapper. Not "we'll integrate Claude for your enterprise." It's model serving infrastructure: the compute layer between a model lab and an API consumer. The fact that infrastructure is pulling $800M rounds while solo operators are doing $40K/month consulting gigs is not a coincidence. It's a market signal about where leverage actually sits.
The capital pattern: infrastructure eats everything else
Together AI is one data point in a broader pattern. In the last 60 days:
- Together AI: $800M Series C on $1.15B+ bookings. Model serving and inference optimization.
- Crusoe Energy: Raising $3 billion to triple its valuation to ~$18 billion. AI compute infrastructure powered by stranded energy.
- Kling AI: $2 billion raise at $18 billion valuation. Video generation (hardware-adjacent: GPUs, inference).
- Even Realities: $150 million pre-Series B from Meituan and Tencent. Camera-free smart glasses (hardware + AI).
The pattern: capital is flowing to infrastructure, hardware, and physical-world autonomy. The "I'll build a SaaS wrapper on Claude" category is getting seed and Series A rounds ($2–5M), not Series C mega-rounds ($800M).
The difference is 100x. Mega-round capital doesn't go to application layer anymore. It goes to the layers that infrastructure builders depend on.
Why infrastructure pulls bigger rounds than apps
Infrastructure has defensible TAMs because it's not 0-sum.
When you build "sales automation with Claude," you're competing for a finite budget in your customer's IT department. You win a $40K/month deal, an Anthropic services partner loses a $40K/month deal. It's zero-sum. Your TAM is the slice of "enterprise AI consulting spend" that you can capture, and that market is crowded.
When you build "inference optimization," you're not competing for budget. You're expanding budget. A customer that runs inference on your infrastructure is a customer that might run 2x more models, 3x more inference jobs, because your stack made it cheaper and faster. Your TAM is "all AI inference done by enterprises over 10 years." That's infinite relative to any single company's budget.
This is why infrastructure raises hundred-million-dollar rounds and apps raise seed rounds. Infrastructure creates value for everyone downstream. Apps compete for scraps.
The uncomfortable implication for you
If you're a solo consultant doing Claude integration work, you're in the wrong layer of the stack.
I'm not saying this to be harsh. I'm saying this because capital allocation reveals the truth. Aramco Ventures doesn't put $800M into something that'll do $40K/month deals forever. They're betting that infrastructure will grow to billions in annual revenue. Meanwhile, the "I'll consult on Claude" business maxes out at $500K–$1M ARR for most solo operators.
This isn't failure. It's structural. The economics of the layers are different.
If you're making $40K–$80K/month on Claude integration work, you're in the mid-market segment that can't afford $2–3M annual infrastructure spend. But you're also underutilizing infrastructure capabilities that could make your work more defensible.
Here's the honest question: Are you building on infrastructure, or are you competing with it?
If you're competing (trying to be cheaper or better at "Claude integration" than the platforms themselves offer), you're playing a game you'll lose. The platforms have more capital, more engineers, and better distribution.
If you're building on infrastructure (using Together AI's model serving, Crusoe's compute, Anthropic's platform layer), you're layering services on top of something that has defensible economics. That's a viable position.
How to reposition
This takes strategy work. But here are three paths:
Path 1: Specialize vertically and own the customer problem. Stop saying "I'll integrate Claude into your CRM." Start saying "I build AI agents for pharma sales workflows" or "I optimize LLM spend for 50–500 person SaaS companies." Vertical specialization creates defensible TAM because you own the domain, not the technology. Infrastructure is still your platform, but you're selling to a specific customer segment and bundling domain expertise with the technology stack.
Path 2: Move closer to infrastructure. Become the sales and integration arm for a company like Together AI or Crusoe. You're not competing on technology; you're competing on sales and customer success. This means leaving solo operation, but it's a higher-growth path.
Path 3: Own the observability layer. Infrastructure is growing fast, but nobody's giving customers visibility into their inference spend, model performance, and cost optimization. Build tools that sit on top of Together AI's infrastructure and give customers control. This is app-layer positioning on infrastructure, and it has defensible economics.
I'd pick Path 1 if I had a consulting business today. Find a vertical where you have unfair domain advantage (worked in pharma? telecom? insurance?), build deep expertise, and use Claude/Fable as your tooling layer. You're not selling Claude. You're selling pharma expertise plus Claude.
The honest counter-take
Infrastructure is capital-intensive and slow. Building an $800M company in infrastructure takes 5–7 years and requires massive investment. Most solo operators don't want that. Console integration consulting is faster to cash, higher autonomy, and less capital-intensive.
If you're doing $40K/month and you're happy with that revenue, velocity matters more than layer positioning. The infrastructure story is for people trying to build a $10M+ ARR business.
But if you're trying to grow beyond $1M ARR or you're burned out by the commodity consulting grind, then layer positioning matters. And the data says: build on infrastructure, don't compete with it.
What I'd actually do
I'd audit my current client base and map each one to two buckets:
- Vertical specialization potential: Does this client segment have unfair domain requirements that I can own? (E.g., "healthcare AI compliance" is more defensible than "Claude integration.")
- Infrastructure dependency: Which parts of my work depend on raw inference cost? Where do I waste money because I'm not optimized?
Then I'd pick the vertical with the highest defensibility potential and start positioning myself as the expert, not just the Claude integrator. I'd also audit my infrastructure spend and potentially shift to a company like Together AI if it saves me 20%+ on inference costs.
This is a 2–3 week project. It pays for itself if it clarifies your positioning.
Author
Lukas
@lukcombinator