Dan Niles: Why Cash Makes Sense, Even as the AI Boom Continues

The Fed under Warsh is likely to raise rates at the September 16, 2026 meeting, because hiking on October 28 would be too close to the November 3 midterms.

Warsh said 65 months of elevated inflation is the central bank's responsibility. Niles says listen to what officials say. Hiking in late October would anger the White House days before the midterms, so September is the window. One soft payrolls or CPI print won't change it after 65 months of bad data.

Rate hikes from several central banks at once (US, Japan, UK) put heavy pressure on stocks, and investors should not fight a tightening Fed.

Many central banks face high inflation and rising inflation expectations, so tightening is global. His rule is 'don't fight the Fed': if the Fed is raising, don't bet against it.

The Fed should hike, because a quarter-point increase won't damage a near-full-employment economy, and inflation hurts most the 40% of households with no stocks or home.

Unemployment is near 4% versus a 'normal' closer to 6%. The 40% of US households with no stocks and no home get no offset from inflated asset prices while paying more for heat, food and gas.

A quarter-point hike won't kill the AI data center buildout, and a pause in data center expansion is healthy, not a sign the boom is over.

If a small hike could kill it, the bubble callers would be right, and he doesn't think they are. Cisco had a couple of 30% selloffs while rising about 4,000% from end-1994 to March 2000. A pause also makes companies question whether they are overbuilding just to match rivals (Larry Page: 'I'd rather go bankrupt than lose this race').

Markets that go straight up are dangerous, because greed and leverage turn the eventual reversal into a collapse, so periodic pauses protect investors.

Markets swing between fear and greed. Leveraged buyers get wiped out when the move reverses, like Korean investors in leveraged Samsung/Hynix ETFs in the late June to July selloff. After Nasdaq rose 86% in 1999 and 24% more in early 2000, it fell 78%. Cisco's bookings went from +70% to -30% year over year within months (May 2001 release).

The Strait of Hormuz disruption is likely to last until at least the November 3, 2026 midterms, keeping oil and gas prices high.

High gas prices hurt the sitting party in midterms, and Republican polling is bad, so Iran has leverage. Iran held US hostages for 444 days and released them hours after Reagan was sworn in, once Carter had been crushed in the election. Trump seems to have ruled out military action other than retaliation.

More than one Fed hike wouldn't surprise him, because high oil and the resource demands of data centers keep pushing inflation up.

Oil staying high seeps into inflation. The global data center buildout soaks up resources and adds upward price pressure. Inflation has been above the 2% target for 65 months. He will adjust as data arrives rather than predict a set path.

Rising long-term bond yields are driven mainly by soaring government debt worldwide, and deficits won't be fixed until a crisis forces it.

The US runs a 6% deficit, the highest outside a major war, with about $40T of debt against about $33T of GDP. Japan and the UK have multi-decade-high yields. Stimulus from the financial crisis and COVID was never pulled back. Politicians chasing reelection keep promising 'free stuff', so he takes the cynical view that only a crisis brings austerity.

The US political shift toward bigger deficits and democratic socialists is bad for big business and the stock market.

Democratic socialists are gaining seats, and their platforms call big business bad. With the political winds blowing this way, he expects deficits to only go up.

In midterm election years the S&P's peak-to-trough drawdown from end of July to November 9 is about 10%, double the roughly 5% in other years.

Based on data from 1990 through 2025 that he posted. He first called it the average, then the median, and first stated the numbers backwards before correcting himself. There are outliers both ways, and the sample includes the 1990 Gulf War, which he sees as similar to now. He declined to give an S&P range.

AI economics are healthy: token prices fell over 50% since end of May, but token volume grew more than 2.5x, and cloud growth and profits improved.

AWS, Azure and Google Cloud growth sped up from 35% in the March quarter to 43% in the June quarter, and operating margins rose 2 points. Anthropic says it reached adjusted profitability in the June quarter. Falling costs and rising volume with growing profits is how technologies normally mature.

When a technology is seen as world-changing, companies overbuild to win the race, so a bubble break with bankruptcies is built in, but money can still be made before the peak.

If companies believe the winner makes huge money, they would 'rather go bankrupt than lose this race'. So once growth plateaus, the bubble breaks and some companies go bankrupt, as with canals, railroads and the internet. If that peak is still a couple of years out, investors can still profit.

China is clearly winning on energy for AI, while US data centers face political pushback ahead of the midterms.

China's central government decides where a nuclear reactor goes, with no community or political resistance. US politicians are moving from supporting data centers to shelving or opposing them to avoid losing their seats.

US AI companies won't fall far behind China on low-cost models, even though China built strong cheap models out of necessity.

Cut off from top Nvidia chips, Chinese labs focused on better algorithms and lower costs. Niles expects Google's upcoming Gemini Flash to be very competitive on low-cost token production, while Anthropic and OpenAI lead on frontier models.

AI demand will split into tiers: about 90% of tasks will use cheap low-end models and only the top 5–10% will need expensive frontier models.

Using frontier models for tasks like summarizing emails is too expensive. You don't need a Ferrari to go to the corner store for milk; a Ford works fine. Frontier models stay valuable for the hardest use cases.

China will take significant share in memory chips, including high bandwidth memory, because chip self-sufficiency is a national security priority.

After being cut off from US chips, China treats chip supply like a battleship or aircraft carrier. It doesn't need the leading edge; Japan in the 1970s–80s and Korea in the 1990s didn't start there either. CXMT is already public and YMTC will list soon, raising capital. China buys roughly 20% of the world's PCs and smartphones, so it can reach 20% share selling at home.

Markets have an earnings bubble more than a valuation-multiple bubble, and memory makers' roughly 80% operating margins will fall below average once China ramps up.

Companies are over-earning compared with normalized earnings. Valuations are high relative to GDP and normalized earnings. Memory makers' 80% margins are not normal, and he 'firmly believe[s]' China's capacity ambitions will push them back below average.

Apple's choice to license Google's Gemini instead of spending hundreds of billions on AI capex has worked well, though he's not sure it was planned.

Rumors say Apple pays a couple of billion dollars to license Gemini while Google spends about $200B, so Apple isn't burning cash flow. Apple's stock gains came mostly from multiple expansion, not products; it is late on AI phones and foldables that Samsung has had since about 2019.

Apple should see a huge iPhone upgrade cycle sometime in 2027, once its AI Siri is out of beta and localized worldwide and better products ship.

The AI Siri launching now is US-only, English-only and in beta. The ecosystem's stickiness keeps users waiting rather than switching to Android. A foldable iPhone and wider language support should drive upgrades.

Solid yields on cash and bonds worldwide now compete with stocks, which is part of why the stock market is struggling.

Japanese investors can get a reasonable, safe yield at home for the first time in maybe three decades, and investors everywhere are running the same comparison.

Shelter in cash, held in money market funds, until the early-November 2026 midterms instead of stocks or bonds.

He sees a good chance of a drawdown more than double normal. Valuations are high, earnings are inflated, and hikes, oil and yields are all risks. Bonds are selling off, so they aren't a safe alternative. Money market funds pay solid yields in every country and stay liquid, as in 2022 when both stocks and bonds fell.

AI has not hit a fundamental peak: agentic AI is only about 7 months old, so he expects at least another year of solid growth.

He dates agentic AI to OpenClaw on January 30, 2026, only seven months ago. He admits many smart people think the bubble is breaking now, pointing to falling token prices and chips that have nowhere to go without data centers.

Broadcom's latest results came in far below expectations, and he expected the stock to fall the next day.

Broadcom reported after the close with numbers 'nowhere near' what people expected, unless the conference call changed the picture.