Blog

Wall Street just agreed to finance half a trillion dollars of Nvidia’s customers — with computing power as the collateral. If you are recruiting into banking, private credit, or infrastructure investing, this is the week the AI buildout stopped being an equity story and became a debt story — and you should expect to be asked about it. On Monday, Nvidia announced preliminary agreements with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize more than $500 billion of third-party capital into dedicated lending pools for Nvidia’s customers. Jensen Huang said he approached only those six firms and none declined; the structures will reportedly use compute capacity itself as collateral, issued through special-purpose vehicles that lease that compute onward, with Nvidia backstopping up to 25% of a given deal. Goldman, the only bank in the group, is positioned as lead bookrunner. The skeptic’s read is worth holding onto: as one analyst put it to Bloomberg, this makes Nvidia’s product cheaper without cutting GPU prices, while making future demand far more sensitive to credit conditions.

Stanford’s entry-level employment gap widened again — and the mechanism is reduced hiring, not layoffs. This is the single most directly relevant labor-market finding of the week for anyone about to enter the job market, and its mechanism matters more than its headline: firms are not firing junior people, they are hiring fewer of them. On August 12, Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen released a revised version of their “Canaries in the Coal Mine” paper, using ADP payroll data through mid-2026. Employment for 22–25-year-olds in the most AI-exposed occupations now sits about 19% below where it would be had it tracked similarly aged workers in less-exposed roles — up from 15% at the July 2025 data vintage. Experienced workers show no comparable gap. The new revision adds a useful distinction: declines concentrate in occupations built on codified knowledge (formal, documented, teachable), while employment holds or rises in roles built on tacit knowledge acquired through practice and mentorship. Read the causal claim carefully — the authors are explicit that these are descriptive patterns, that they cannot rule out other drivers, and that the gap shrinks once you control for education.

Databricks raised $5 billion at $190 billion — and, unusually for this cycle, showed the revenue to justify it. Use this as your counterweight to the AI-bubble argument: enterprise spending on AI infrastructure is showing up as real, accelerating revenue at scale, not just as valuation. On Thursday, Databricks closed a $5 billion round led by Coatue with Blackstone, MGX, and T. Rowe Price, at a $190 billion valuation — its second raise of the year. Alongside it, the company disclosed that it crossed a $7 billion revenue run-rate, growing more than 80% year over year in Q2, an acceleration from prior quarters. Proceeds are earmarked for its AI-agent stack: Lakebase (its serverless Postgres database for agents, now past a $100 million run-rate), Genie, and Unity AI Gateway.

Claude’s text now carries an invisible watermark — assume anything you paste from a chatbot is detectable. This one lands directly on your own workflow, in class and at your internship: the operating assumption should now be that AI-generated text you copy out is machine-identifiable. Anthropic confirmed on August 11 that it is watermarking model output to comply with the EU AI Act’s Transparency Code, which took effect August 2. All models released after that date embed a watermark in generated text (and C2PA metadata in files); it is applied at the model level, so it travels across the API, Claude, Claude Code, and Cowork alike, persists through copy-paste, and may survive some editing. Older models will be covered too. Google, Meta, Microsoft, OpenAI, Black Forest Labs, and Synthesia have all committed to the same EU code, so expect this to become the default rather than a Claude quirk. Predictably, a chunk of users spent the following day objecting on Reddit — though, as TechCrunch noted, most of the complaints amounted to being caught.

Google hit a billion Gemini users and halved its coding-model price — while its flagship is still late. Google is winning on distribution and price while still losing on the frontier, which is a useful reminder that “best model” and “winning market” are different competitions. On August 11, Sundar Pichai announced the Gemini app passed one billion monthly users — Google’s 14th product to do so, up from 400 million in May 2025 — with the app now generating over 150 million images a day and 63% of users now talking directly to Gemini rather than typing. Two days later Google shipped Gemini 3.7 Flash, released just three weeks after 3.6 Flash, at half the previous price through year-end ($0.75 per million input tokens, $3.75 output); Google claims it beats comparable Anthropic and OpenAI models on nine benchmarks. Note what is still missing: Gemini 3.5 Pro, Google’s delayed flagship, remains unreleased — the same delay that sat behind this month’s DeepMind leadership shakeup.

OpenAI spent the week building a commercial engine: ads abroad, a new revenue chief, and speed as a paid tier. Three separate announcements in three days point the same direction — OpenAI is converting research leadership into a segmented revenue model, which is the part of the AI business most likely to show up in your strategy and marketing coursework. On August 11 it extended its ChatGPT advertising pilot into the UK, Mexico, Brazil, Japan, and South Korea (ads run only on Free and Go tiers, and OpenAI says they do not influence answers). On August 13 it appointed Dali Rajic as Chief Revenue Officer. Also on August 13 it previewed Ultrafast, a new API tier running GPT-5.6 Sol up to 14x faster at up to 750 output tokens per second on Cerebras hardware — monetizing latency as a distinct product rather than a byproduct. Jane Street, Podium, and Rogo are among the named preview customers; access is limited while capacity scales.