Three Signals VPs Were Circling This Week (May 19–25)

Three independent threads surfaced in VP-level public commentary this week — compute economics, org restructuring, and immigration risk — and they converged on the same underlying question: who controls the inputs to the next phase of AI?

Google I/O flipped the conversation on infrastructure cost

Sundar Pichai, CEO of Google and Alphabet, opened Google I/O on May 19 with a data point that reframes how senior leaders should think about model strategy: Gemini APIs now process roughly 19 billion tokens per minute, with monthly throughput growing 7x year-over-year to over 3.2 quadrillion tokens 1. Pichai also disclosed that Google's 2026 capital expenditure is expected to run $180–$190 billion, roughly 6x the 2022 level.

The reveal that matters most for early-career readers watching VP behavior: the new Gemini 3.5 Flash model is priced at less than half the cost of comparable frontier models and outputs tokens 4x faster 1. Pichai's team positioned this as a business-level shift: moving 80% of workloads to Gemini 3.5 Flash could save a large enterprise over $1B annually. That framing — using model choice as a P&L lever — is how cost conversations at the VP level now begin.

Jeff Dean, Chief Scientist at Google DeepMind and the technical face of Gemini, underscored the same point directly on X on May 19: "Highly capable models that are fast are super important. Our new Gemini 3.5 Flash model is a great mix of fast and capable." 2

Jeff Dean @JeffDean·7d

Highly capable models that are fast are super important. Our new Gemini 3.5 Flash model is a great mix of fast and capable.

View on X

正在加载内容卡片…

Liz Reid, Google's VP of Search, spelled out what that agentic infrastructure actually changes for the product layer: Search will no longer be defined by ranked links. It now dispatches information agents that work in the background 24/7, builds generative UI on the fly for individual queries, and lets users assemble stateful "mini apps" using natural-language commands 3. In a pre-I/O briefing, Reid described agents that "map out a monitoring plan for you, including the tools and the data it needs to access — like our real-time finance data." The shift is structural: humans increasingly receive synthesized conclusions from agents rather than retrieve information themselves.

Sundar Pichai opens Google I/O 2026 — Pichai revealed that Google's 2026 capex is expected to reach $180–$190B, 6x the 2022 level 1

Meta's VP called the org shift "unprecedented" — in writing

While Google was onstage, Meta was issuing mandatory transfer notices to thousands of engineers. Peter Hoose, Meta's VP of Production Engineering, wrote in an internal post viewed by The Guardian 4:

"Our work, infrastructure and our products are fundamentally changing as a result of the continued acceleration of AI. The pace of what we are building is unprecedented, and these are exactly the kind of challenges that define what we do best."

Hoose's post covered two new teams: one building AI cloud infrastructure, another building an internal AI agent codenamed Hatch. Transfers to both were described as mandatory — employees earlier told they were "selected" had already seen a precedent: in April, at least 1,000 engineers were moved to a data-labeling team called Applied AI (AAI) on what started as an opt-in but quickly became explicit: "Transfers aren't optional."

Meta also began rolling out Model Capability Initiative (MCI), a monitoring tool that captures mouse movements, keystrokes, laptop open/close events, and copy-paste data to train AI models. Over 500 employees have signed a petition against it. A group in the UK is organizing with United Tech and Allied Workers (UTAW).

For someone building a career in big tech: a VP-level memo that describes org transformation as "unprecedented" and simultaneously makes transfers non-negotiable is a signal about where executive priority actually sits, independent of what the earnings call says about employee investment.

Satya Nadella, Chairman and CEO of Microsoft, put the same shift in business-strategy language on May 13: "Every firm will need to reconceptualize work as they build agentic systems. As AI and agents take on more of the execution, the opportunity is to expand human agency and redesign how work gets done." 5 He linked to an in-depth analysis from his team. What Nadella frames as opportunity, the 500 Meta petition signers are experiencing as friction — both descriptions are accurate.

Immigration policy as an AI infrastructure risk: three independent voices

The most consistent signal this week — and the one most likely to be undersold by mainstream coverage — was VP-level concern about US immigration policy as a constraint on AI capabilities.

Andrew Ng, co-founder of Coursera (now chairman of the combined Coursera–Udemy company 6) and former head of Google Brain, posted on May 22 with unusual directness 7:

Andrew Ng @AndrewYNg·4d

The new White House policy requiring green card applicants to apply from outside the US is a capricious attack on legal immigration. It will hurt families, leave us with fewer doctors, teachers and scientists, and hurt American competitiveness in AI.

View on X

正在加载内容卡片…

The post drew 12,000 likes and 1,600 retweets — atypically high engagement for Ng, who typically posts course launches and technical frameworks. Jeff Dean retweeted related commentary from a colleague calling the same policy "a huge blow to America's tech sector." Yann LeCun — Executive Chairman at AMI Labs and former Chief AI Scientist at Meta — retweeted multiple independent threads on the same topic across May 23–24, including Erik Brynjolfsson's note that "historically, the US has been the greatest magnet for talent on earth. Whether that continues to be the case is not inevitable." 8

Three researchers from different institutions, none coordinating publicly, all flagging the same structural risk in the same 72-hour window — that's a meaningful convergence at the VP/chief-scientist altitude.

Ng's education take: what VPs think about grading and talent pipelines

In a separate, longer post on May 22, Andrew Ng pushed back on Harvard's decision to cap A grades at roughly 20% of undergraduates — a move framed as a counter to grade inflation 9. His argument was direct:

"Having hired a large number of people from Harvard and other institutions, I can say confidently that GPA is not an important signal. We have screening and interviewing processes that give far more accurate ways to figure out if someone is truly skilled."

Ng rejects the "elite through exclusion" model. His alternative: "Set a high bar and teach elite, cutting-edge skills, but strive relentlessly to help everyone succeed." The post, originally from his The Batch newsletter, received 2,100 likes. For early-career readers, this matters: the person who built DeepLearningAI and helped found Coursera is on record saying GPA is not how VP-level hiring works at the firms you're trying to join.

What to watch next week

Gemini 3.5 Pro is due next month (currently in internal testing). If benchmarks hold, it intensifies competitive pressure on Claude Opus 4.7 and GPT-5.5 on the latency-per-dollar dimension — the metric Pichai's team is explicitly calling out.
Meta's Muse Spark (the first model out of Meta Superintelligence Labs) is live. Watch whether Zuckerberg starts positioning it as a talent retention tool or purely as a product offering — that framing will signal which internal constituency he's managing.
The Coursera–Udemy merger is officially closed, with Ng as Chairman. The combined platform will be the largest employer-facing learning provider at scale. VP learning-and-development teams will be making platform decisions based on this.