Oracle booked $65B in AI cloud deals—including Meta—in 30 days, but the stock slipped after-hours despite raised 2030 targets. Investors want proof on 30–40% margins, not bookings. Demand's real. Execution's unproven. The scoreboard moves there.
Anthropic wired Claude into Microsoft 365, chasing institutional memory while Microsoft hedges with model choice inside Copilot. The fight isn't about chat features—it's about who mediates your company's knowledge day after day.
Good Morning from San Francisco, Apple promoted its Siri overhaul chief. Three weeks later, he joined Meta. That makes twelve
Apple's newly promoted Siri AI chief lasted three weeks before jumping to Meta—the twelfth senior researcher to leave this year. With a March overhaul deadline looming and more exits expected, Apple's talent strategy collides with reality.
The defection hits Apple’s AKI unit as a March 2026 Siri overhaul looms, deepening a brain drain to Meta.
Ke Yang left Apple for Meta less than a month after being promoted to run the Answers, Knowledge and Information (AKI) team—the group charged with making Siri pull fresh facts from the web and speak in ChatGPT-style paragraphs. The move trades a date-certain ship target in March 2026 for a research-first post inside Meta’s Superintelligence Labs, as first reported by Mark Gurman of Bloomberg. It also extends a year of senior AI departures from Apple, several to the same Meta unit, including former Foundation Models lead Ruoming Pang. That’s the signal, not the noise.
Apple didn’t just lose an engineer. It lost the executive accountable for Siri’s most meaningful upgrade in years—weeks after elevating him. The promotion reads like a countdown he chose not to finish.Mission over milestone. That’s the tell.
The tension isn’t base pay. Both firms pay at the top of the band. It’s cadence and structure. Apple’s AI groups sit in product organizations with annual release trains, heavy review gates, and confidentiality that chills publishing. Meta’s lab reports directly to Mark Zuckerberg, optimizes for capability gains over calendar quarters, and encourages papers. Presented with “ship Siri’s web answers by March” versus “push the frontier of language models,” many PhDs pick the latter. They always have.
Benoit Dupin now oversees AKI. He runs machine-learning infrastructure, not Siri product strategy. On paper, that’s continuity. In practice, it looks like triage—folding a high-stakes feature team under an executive already tasked with keeping the pipes flowing. That matters.
The Breakdown
• Ke Yang departed for Meta's Superintelligence Labs three weeks after being promoted to lead Apple's Siri web-search overhaul team.
• Apple's Foundation Models team lost roughly twelve senior researchers this year, most to Meta's new lab offering publication freedom.
• The March 2026 Siri deadline now falls to infrastructure chief Benoit Dupin, who inherited the role alongside his existing responsibilities.
• Remaining team members expect more departures before year-end as Apple struggles to retain researchers prioritizing breakthroughs over shipping dates.
From Apple’s side, this isn’t normal attrition. Roughly a dozen senior exits from the Foundation Models and adjacent teams in a single year is startup-grade churn during a critical window. The Siri rework was already delayed once. The calendar didn’t become friendlier.
From Meta’s point of view, it’s targeted acquisition. Yang steps into a group that already houses Pang and other ex-Apple researchers—people who know Apple’s approach, constraints, and integration philosophy. They now work without a quarterly launch drumbeat, with heavy compute budgets, and with publication as career currency. That’s a recruiting advantage.
For the researchers themselves, the choice is clarity. Apple recruits for research but manages to product milestones. Meta built a lab to avoid that mismatch. The scoreboard reflects it.
Apple says Siri’s overhaul is scheduled for March 2026. The work spans three interlocking challenges: stream fresh web data, synthesize answers credibly, and make it all comport with Apple’s on-device privacy architecture. Each is hard. Together, they’re brittle.
Integration is the boss. Not the calendar.
Yang’s role was to align those pieces and unblock daily decisions. With AKI moved under infrastructure, the single-owner model blurs. First-time product leaders rarely hit aggressive timelines when they’re also running their previous remit. That’s not cynicism; it’s operations.
There’s a broader pattern, too. When research groups get pulled toward quarterly deliverables, people who optimize for papers and breakthroughs leave for places that optimize for… papers and breakthroughs. Facebook’s FAIR saw it during the pivot to monetization. Apple is seeing it now from the other direction, trying to turn research into a shippable assistant on a date certain.
Three near-term signals will show whether this is a wobble or a stall. First, does Apple ship on schedule—and with how much scope intact? Second, do more Foundation Models veterans exit before year-end? Third, does John Giannandrea remain atop Apple’s AI org, amid reports that Apple has sounded out external candidates? One “yes” can offset a “no.” Two “no’s” can’t.
If Apple hits March with narrowed features, it chose deadline over depth. If it slips, it admits the churn hurt. If it lands late with compromises, the date was more investor relations than engineering reality. None are fatal. All are revealing.
You can retain chip designers with equity cliffs and campus prestige. You can’t retain frontier AI talent without publication freedom and research autonomy. Meta built a lab to exploit that asymmetry. Apple is still asking researchers to live on a product clock.
Yang’s tenure as AKI head was measured in weeks, not quarters—long enough to see the constraints, not long enough to be trapped by them. The March deadline remains. The team tasked with meeting it is thinner, more cross-loaded, and—crucially—missing the leader hired to make the pieces click when complexity peaks. That’s the story.
Q: What exactly is Meta's Superintelligence Labs?
A: Meta created Superintelligence Labs in 2025 as a research division focused on achieving artificial general intelligence. It's led by Ruoming Pang, Apple's former Foundation Models chief, and reports directly to Mark Zuckerberg. Unlike product teams, researchers can publish openly, access heavy compute resources, and work without quarterly release deadlines—making it a magnet for academic-minded AI talent.
Q: Why does publication freedom matter so much to AI researchers?
A: AI researchers build reputations through papers published at conferences like NeurIPS and ICML, not product launches. Publications lead to speaking invitations, faculty positions, and future job offers. At Apple, legal review slows publishing to protect trade secrets. At Meta's lab, researchers can post to arxiv.org immediately. For PhD-level talent, that's the difference between advancing a career and stalling it.
Q: What's Apple's AKI team actually building for Siri?
A: AKI (Answers, Knowledge and Information) is building web-search capability into Siri so it can pull fresh data and respond in ChatGPT-style paragraphs instead of short scripted answers. The March 2026 update should let Siri synthesize information from multiple sources, handle complex multi-step requests, and access personal data—capabilities Perplexity and ChatGPT already offer.
Q: How much do senior AI researchers earn at Apple and Meta?
A: Senior AI researchers at both companies typically earn $400,000 to $800,000 in total compensation—base salary plus equity. Staff-level researchers and team leads can reach $1 million or more. Meta sometimes offers retention bonuses exceeding $2 million for particularly sought-after talent. Compensation isn't the differentiator; it's structure, autonomy, and publication rights that drive movement between firms.
Q: Who's John Giannandrea and why does his job security matter?
A: Giannandrea is Apple's Senior Vice President of AI and Machine Learning, hired from Google in 2018 to lead Apple's AI push. Bloomberg reports Apple's interviewing external replacements, which typically signals board-level concern about execution. If Giannandrea exits, it would mark the third layer of leadership turnover—after Yang and the dozen Foundation Models researchers—suggesting systemic problems, not isolated departures.
Marcus Schuler
Marcus Schuler
Maria Garcia
Oracle booked $65B in AI cloud deals—including Meta—in 30 days, but the stock slipped after-hours despite raised 2030 targets. Investors want proof on 30–40% margins, not bookings. Demand's real. Execution's unproven. The scoreboard moves there.
Anthropic wired Claude into Microsoft 365, chasing institutional memory while Microsoft hedges with model choice inside Copilot. The fight isn't about chat features—it's about who mediates your company's knowledge day after day.
Anthropic's Haiku 4.5 delivers May's frontier coding performance at one-third the cost, collapsing the capability-to-commodity timeline to five months. Multi-agent systems just crossed the economic threshold for production deployment.
Oracle bets AI's next phase runs on private enterprise data, not public models. The database giant promises secure training without data exposure, backed by $500B in infrastructure. But tight coupling with Nvidia and OpenAI creates dependency risks.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
