CNBC The Exchange

2026-05-12 · Hosted by Kelly Evans · CNBC

Executive Summary

The Exchange opened with a pointed dot-com bubble comparison from Dan Niles (Niles Investment Management), who argued we are in “year three” of a ChatGPT-era bubble analogous to 1997–1999 — with at least one more great year ahead before the reckoning. The episode also featured Satya Nadella’s testimony in the Musk vs. Altman OpenAI trial (framing Microsoft as a neutral business partner), OpenAI’s new PE-backed deployment joint venture with a 17.5% guaranteed return, a deep dive into Sievert (SiTime), and the upfronts advertising season with streaming now overtaking linear TV. Geopolitically, the Trump-Xi Beijing summit and Iran ceasefire fragility remained central topics.

Key Stories & Changes

1. Dot-Com Analog: Dan Niles’ Framework

  • Dan Niles (Niles Investment Management): Nasdaq up 122% in the first 3 years post-ChatGPT (Nov 2022–May 2026) vs. 109% in the first 3 years of the internet (Netscape Nov 1994–Nov 1997)

  • Niles: we likely have “at least one more great year” — in 1998–1999 (years 4–5 of internet), Nasdaq added 40% then 86%

  • Key structural difference from 1999: Agentic AI (OpenClaw finalized Jan 30, 2026) requires 10–100x more compute than chat AI; token growth jumped from ~20% to over 120% in two months after OpenClaw launch

  • Anthropic revenue: from $9 billion (end-2025) to $45 billion in four months; combined Anthropic+OpenAI annualized revenue from $7B (start of 2025) to ~$70 billion now

  • What ended the dot-com bubble: Y2K liquidity flood removed, oil spiked, Fed drained markets; NASDAQ internet traffic went from doubling every 3 months to doubling every year — not bad, but not what valuations priced in

  • Niles current positioning: significant cash; would not be surprised by 10–15% semiconductor index drop in short order; prefers Alphabet (GOOGL) as “full stack” best-in-class; looking at non-tech for deployment of new capital

  • Caution: oil in mid-90s + treasury yields at year-to-date high + stock market at all-time high = “one of those things is wrong”

2. Satya Nadella Takes the Stand — Musk vs. Altman Trial

  • Microsoft CEO Satya Nadella testified in the Elon Musk lawsuit against OpenAI and its co-founders (Sam Altman, Greg Brockman)

  • Nadella framed Microsoft as a “neutral business partner” — invested ~$1 billion in OpenAI to compete with Google; computing power was the identified challenge

  • Nonprofit approved creation of for-profit entity so the mission could continue — Nadella’s stated understanding

  • Nadella: “surprised” when Altman was fired by the board; needed to ensure Microsoft customers were not affected (they weren’t)

  • Musk never reached out to Nadella to express concerns — Nadella confirmed Musk has his phone number

  • Key dynamic: Microsoft has since partnered with Anthropic; OpenAI with Amazon — the original partnership has diversified

  • Bloomberg Law note: “no smoking gun” email so far; mostly about “egos and personalities”

3. OpenAI’s PE-Backed Deployment Joint Venture

  • OpenAI acquiring consulting company Tomorrow to seed a new joint venture focused on enterprise AI deployment

  • JV starts with ~150 employees; goal: drive enterprise adoption of OpenAI software across portfolio companies

  • $4 billion investment from PE firms: TPG, Bain, Goldman Sachs, Walsh Carson, Robert Pinkus, and others

  • ~17.5% minimum guaranteed return for PE investors — making it near-risk-free for the capital providers

  • Competitive move against Anthropic, which has its own enterprise push; OpenAI expects enterprise to reach 50% of business by year-end (per CRO Denise Dresser)

  • Strategic implication: OpenAI needs its AI embedded across the broader economy before rivals do

4. SiTime (SITS) — The “Sleeper” in AI Infrastructure

  • SiTime (ticker SITS): up 140% year-to-date; up 350% over past year; $22 billion market cap

  • Makes precision timing semiconductors (oscillators, clocks, resonators) — coordinates signals across complex AI systems (GPU, switch, optical module)

  • Guiding 80% revenue growth for 2026 (vs. prior typical 25–30%); AI/data center segment showing triple-digit growth for 8 successive quarters

  • CEO Rajesh Vashist: “This one is based on revenue, profitability, and a very large set of customers” — strongly rejects bubble comparison

  • UBS called SiTime “fastest growing company in the sector” — “less like a concentrated Apple/Nvidia supplier, more like a diversified dealer to the entire industry”

  • Customers: hyperscalers, GPU/CPU companies, switch companies, optical module makers; also military/aerospace/defense, industrial, auto, consumer, smartphones

  • “Moore’s Law is basically dead” — CEO says SiTime has high barriers to entry; tried by competitors but market remains concentrated

  • Trading at 112x PE — elevated, but justified by growth trajectory in CEO’s view

5. Trump-Xi Beijing Summit — Expectations vs. Reality

  • 17 CEOs in the delegation; Trump arrives Beijing Wednesday; two full days with President Xi

  • Agenda: trade, AI, export controls, Taiwan, Iran war

  • Tobin Marcus (Wolf Research): expects this summit to primarily “reinforce and extend the détente” from the Busan truce (Oct 2025) — not transformative new deals

  • Busan truce: tariff easing (IEPA tariffs from 20% to 10%), mutual pause on export controls (US paused “50% rule” on Chinese entities; China paused rare earth restrictions)

  • Bigger win scenario (Chinese investment pledges into US factories) always been politically fraught — Chinese EV factories in Michigan/Wisconsin unlikely to be politically popular

  • Marcus: Iran will dominate discussions; Trump expected to pressure Xi to reduce China’s support for Tehran

  • China: Tehran’s largest oil buyer; sold military technology used against US forces in region; US sanctioned 3 Chinese satellite companies for providing imagery to Iran

  • Kevin Warsh Fed Chair confirmation: “glide path” — no Republican opposition; will be confirmed this week; first FOMC press conference in June; faces divided committee on inflation vs. growth

6. Advertising Upfronts — Streaming Overtakes Linear TV

  • For the first time, streaming TV upfront ad spending projected to exceed primetime linear TV upfront spending (per eMarketer)

  • Amazon and YouTube: each at ~12% streaming ad market share; followed by Roku, Hulu, Peacock; Netflix younger/smaller but growing

  • eMarketer warning: a sustained energy shock could reduce connected TV ad spending growth from 12% to under 10% in 2027

  • Linear TV projected to decline 14% in ad spending next year; NBC Universal, Fox showcasing pilots this week

  • Major shift: live sports (especially NFL) is the dominant anchor; NFL package announcement expected Thursday

  • Retail media networks (Kroger, Albertsons, DoorDash, Walmart via Vizio acquisition) growing as a new ad category

  • Monday.com: biggest revenue beat in 15 quarters; initial jump of 25%+ pared to ~4%; company pivoting from seat-based pricing; keeping headcount flat for the rest of year

1. Agentic AI as the True Inflection Point

Dan Niles’ token generation data is the most concrete evidence to date that the AI compute demand curve stepped up sharply in early 2026. The 120%+ token growth rate following OpenClaw finalization in January is qualitatively different from the steady ~20% prior growth. If Niles is right, the structural demand for compute (CPU/GPU/memory) should remain robust through at least early 2027 — but the lap against hard 2026 comparables begins then.

2. OpenAI’s Enterprise Land Grab

The PE-backed deployment JV with a 17.5% guaranteed return is a sophisticated corporate strategy: use other people’s capital to embed your technology across thousands of portfolio companies. The 17.5% guaranteed return is a remarkable sweetener that signals OpenAI’s confidence in its ability to monetize enterprise deployments — but also suggests it is willing to make aggressive commitments to win the race against Anthropic.

3. “Sleeper” Stocks Reveal AI’s Depth

SiTime’s trajectory — up 350% in a year while most retail investors have never heard of it — illustrates how deep the AI infrastructure trade runs. Precision timing chips are not glamorous, but they are essential: every AI system requires coordinated signal timing, and SiTime has pioneered and dominates this niche. Its 112x PE and 80% revenue growth guidance suggest the market is beginning to recognize these deeper infrastructure plays, not just the headline GPU/memory names.

4. Advertising Hits a Structural Inflection

The streaming-overtakes-linear milestone is more than a media industry factoid — it represents a fundamental reallocation of brand advertising dollars that has been decades in the making. The emergence of retail media networks (Walmart/Vizio, DoorDash, Kroger) as a new category creates a fragmented but growing third pool of ad dollars that didn’t exist five years ago. —-

Sentiment Analysis

Overall Market Sentiment: Cautiously Bullish with Structural Concerns

Niles is the episode’s intellectual anchor: bullish for 2026 but strategically holding meaningful cash, watching oil and bond yields as the key tells. The overall tone is not fearful but is more thoughtful than the pure momentum shows.

Risk Factors Highlighted

Dot-com echo risk: Nasdaq +122% in three years post-ChatGPT mirrors internet bubble trajectory precisely; 1999 analog suggests one more great year before potential 80%+ correction

Oil + yields + stocks trilemma: Dan Niles’ key concern — oil mid-90s, treasury yields at YTD highs, and stocks at ATH cannot all be right simultaneously

Compute cost pressure on software margins: Monday.com and others seeing gross margin compression as token and memory costs rise; could halt software sector re-rating

OpenAI guaranteed return obligation: 17.5% minimum return to PE investors creates a financial obligation that must be funded; revenue projections must be met

Iran dominating China trip agenda: Primary Trump-Xi focus expected to be Iran, not trade — limits upside for business deals that markets are pricing in

Kevin Warsh’s divided FOMC: New Fed Chair inherits above-target inflation + energy shock + competing FOMC factions; June meeting policy direction uncertain

Token demand lap: Niles explicitly flagged that beginning of 2027, the easy token growth comparisons end — potential for demand deceleration hitting compute-exposed stocks

Geopolitical risk to SiTime and similar: Military/aerospace/defense is 20% of SiTime revenue; China/Taiwan tensions could create export control risk for precision timing chips used in defense systems

This episode was covered in today’s The Market Signal — 2026-05-12, a cross-source synthesis of multiple podcast reports.

Keep Reading