AI infrastructure became a Wall Street product, software creation attracted another mega-round, and the era of permanently cheap frontier models started to look rather optimistic. Nvidia recruited six of the world's biggest financial institutions to help create more than $500 billion of AI compute financing. Databricks raised $5 billion at a $190 billion valuation while growing revenue by more than 80%. Lovable doubled its valuation to $13.3 billion and is hiring heavily across machine learning, product, infrastructure and security. DeepSeek priced its new flagship up to 14 times above its faster alternative, and Apple is building a separate AI stack for China. Capital is concentrating around companies that own a genuine bottleneck: compute, enterprise data, software creation, model economics, security or market access.
Key Takeaways
AI infrastructure is now part technology programme, part project finance: hire AI infrastructure and data-centre programme leaders, capacity planning and GPU fleet engineers, FinOps and AI unit-economics specialists, power procurement and grid specialists, and infrastructure finance and risk teams
The enterprise data layer keeps winning: distributed systems and database engineers, data platform and streaming engineers, AI gateway and agent-platform engineers, governance, privacy and lineage specialists, and forward-deployed enterprise architects
AI coding did not remove engineers, it moved demand to people who make software generation dependable: ML and agent-product engineers, platform and SRE, application and cloud security, product engineers who own outcomes end to end, and developer experience engineers
Better models can carry a very large premium, so measure quality-adjusted cost: model evaluation and benchmarking engineers, AI FinOps and inference optimisation, model-routing and fallback platform engineers, and agent reliability and observability engineers
Global AI products are becoming portfolios of regional systems: multilingual model evaluation, AI policy and privacy engineering, regional data and safety teams, partner engineering and product leaders experienced in regulated markets
A 70,000-applicant field experiment found applicants interviewed by AI voice agents were 12% more likely to receive offers, with higher starts and retention, when humans still made every hiring decision
This week: add AI economics to one senior technical interview, audit every internally created AI or vibe-coded application, run a structured-interview pilot with mandatory human approval, and build a five-company sourcing watchlist
Show Notes
Nvidia signs memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create compute-financing platforms targeting more than $500B of third-party capital, with Nvidia potentially backstopping up to $125B (25%) as Big Tech AI spending is expected to exceed $730B this year
Nvidia is reported to have cut its proposed initial guarantee for OpenAI's 10GW Ohio data-centre project from around $250B to less than $120B after investors raised concerns about its exposure
Databricks raises $5B at a $190B valuation, up from roughly $134B six months earlier, passing a $7B annualised run-rate with 80%+ year-on-year Q2 revenue growth, Lakebase above $100M run-rate and Lakehouse warehousing past $1.5B
Lovable raises $400M Series C at $13.3B, doubling its December valuation, with ARR nearly tripling from $200M toward $600M by end of August, 60M+ projects created, 900M+ monthly visits to Lovable-built apps, and plans to grow to roughly 450 people hiring most heavily in ML, product, infrastructure and security
DeepSeek releases V4 Pro at $1.32 per million input tokens and $3.96 per million output tokens, about nine times the input and 14 times the output price of V4 Flash, scoring 53 on the Artificial Analysis Intelligence Index versus 40 for Flash, while planning to at least double staffing
Apple has reportedly trained a China-specific large language model with Alibaba's support, with Apple Intelligence expected in China after regulatory clearance and Qwen incorporated into the local product
Funding watch: River AI $1.1B, Neros Technologies $250M Series C at $2.5B, Silicon Data $30.5M Series A, Mindgard $30M Series A, Cytix $7M Series A
Quick bytes: a draft US letter reportedly tells 35 partner countries that a US-led AI coalition is incompatible with Beijing's framework; Z.ai says GLM-5.3 scored 84.5% on CyberGym versus 83.8% for Mythos 5, unverified; Aurora and Kodiak get California permits for heavy self-driving truck tests; SafePal discloses unauthorised access affecting about 39,798 customers
AI Tool Spotlight: Screenloop — an ATS plus interview intelligence, with an AI notetaker that transcribes interviews, summarises evidence against predefined attributes and drafts scorecards for human review
Frequently Asked Questions
What does Nvidia's $500B compute-financing push mean for hiring?
AI infrastructure decisions now blend engineering with project finance. Senior infra interviews should ask how a candidate compares buying, leasing, reserving or cloud capacity, and what evidence would make them reduce a long-term commitment.
Why does Databricks' $5B raise matter for engineering hiring?
The market is rewarding the layer that makes enterprise AI usable, governed and connected to live company data. Look for evidence a candidate moved messy, permissioned, business-critical data into a reliable production workflow with measurable adoption.
What does Lovable's hiring plan tell us about AI-era software roles?
When millions of users can generate applications, the scarce work becomes architecture, platform reliability, secure defaults, payments, governance, observability and turning prototypes into products that survive real traffic.
How should teams respond to DeepSeek's premium pricing?
Measure quality-adjusted cost, not token price. Give candidates two models with different quality, latency and price, and ask them to design the evaluation and routing policy that optimises cost per successful outcome.
How should a team pilot Screenloop this week?
Pick one role with at least 10 interviews, agree five anchored competencies up front, draft scorecards with Screenloop, require interviewers to approve or edit every AI-generated assessment, and calibrate on anonymised evidence. Track time to feedback, correction rate, evidence coverage and candidate experience.