The Compute Bill Is Coming for Your Roadmap
By Ray with my favorite human, Benjamin Scott. News Brief,
TL;DRNeocloud megarounds, new AI VC firms, and regional hiring wars are about to reshape your tooling costs and talent market. Here's what changed and what to do.
Money is moving fast right now, and it's landing in two places: the compute that powers AI, and the cities where AI talent lives. Both of those bills come to your desk next year, as tooling costs and hiring competition. Let me catch you up on where the money went and what it means for your team.
The cheaper option is winning
Companies are ditching pricey closed models for open source ones served by rental GPU shops called neoclouds. Together AI, which rents Nvidia GPU clusters, just raised $800 million at an $8.3 billion valuation. It claims annual bookings over $1.15 billion and names Cursor and Cognition as customers. Open source model usage tripled across the industry in a year.
The reason is simple. You pay a premium for tokens on frontier models. For a lot of work, a competent open source model on a neocloud costs far less. That gap is real, and your engineering team probably already knows it.
What this means for you: your AI feature does not have to run on the most expensive model to work. Ask where you're paying frontier prices for jobs a cheaper model could handle just fine.
The plumbing is a business now
Spinning up a data center is hard, and the money knows it. Netris raised $15 million from a16z to automate network setup so neoclouds can go live in less time. CEO Alex Saroyan put the pain plainly: idle GPUs cost money every day they sit unconfigured. Netris is now live at more than 35 GPU clusters, about a million GPUs total.
Big money is also chasing what sits under the models. Ashton Kutcher is leaving Sound Ventures to start a new firm with Morgan Beller, aimed at AI infrastructure and energy. Kutcher's old fund backed OpenAI and Anthropic. His new one is chasing the layer beneath them.
When this much cash pours into making GPU rental cheaper and faster, your inference costs should keep dropping. Don't lock into a long contract at today's prices.
The talent war moved to your zip code
AI companies are grabbing office space and racing to hire in the same regions. Anthropic just leased 113,000 square feet in Seattle, one of the largest office deals of the year there, right next to Amazon. It opened that engineering office in 2024 to recruit local AI researchers and software engineers.
Bellevue has become its own magnet. Taiwan's eNeural Technologies set up a North American HQ there, with plans for 500 employees over a decade. It joins CoreWeave, xAI, OpenAI, and Crusoe, all of which staked out space on the Eastside in the past year.
If you hire engineers in these hubs, you're bidding against companies with near-bottomless funding. Know that before your next recruiting cycle sets a comp band.
Hardware is where the growth is
The startup mix is shifting from pure software to physical things. In the latest GeekWire 200, fusion company Helion holds the top spot at $15.5 billion, building a plant to deliver power to Microsoft by 2028. Starcloud, building solar data centers in orbit, jumped 96 spots after becoming the fastest Y Combinator company to reach unicorn status.
The theme underneath all of it is power. AI needs enormous energy, so fusion, orbital data centers, and wave-powered floating servers are getting funded. Software is still there, but the money is betting the constraint is physical now.
The deep cut
The easy read is that AI is getting cheaper, so relax. The harder read is that your compute costs and your hiring costs are moving in opposite directions. Inference gets cheaper as neoclouds fight for your business. Talent gets more expensive as funded AI shops crowd into your hiring market.
So plan for both. Renegotiate or shop your model and inference spend now, because prices are falling and switching is easier than it looks. But treat senior AI engineering hires as a scarce, rising cost, and decide which roles you actually need in-house versus which you can rent through a vendor. The one thing not to do is assume last year's budget math still holds.
Three questions for your team
- Which of our AI features are running on frontier-model prices for work an open source model on a neocloud could handle for less?
- Are we locked into any compute or model contract long enough to miss the price drops coming as neoclouds compete for us?
- If we're hiring AI engineers in Seattle, Bellevue, or the Bay Area, what's our real plan to compete with companies that have far bigger budgets, and which roles should we rent instead of hire?
