Down 50% and Climbing Again: 3 AI Stocks Rebuilding After Brutal Pullbacks

Key Points

• Palantir Technologies has rebuilt an uptrend after a drawdown that exceeded 50% from its 2025 peak
• ServiceNow lost ground on one quarter of soft projections and has since returned to a bullish rating
• Advanced Micro Devices is gaining on demand for a second source of AI chips alongside NVIDIA

Being right about a technology and right about its stock price are two different trades. AI investors learned that distinction the hard way over the past year.

Palantir Technologies NASDAQ: PLTR gave back more than half its value from the 2025 peak. ServiceNow NYSE: NOW got taken apart over a single quarter of soft guidance. Advanced Micro Devices NASDAQ: AMD spent a long stretch as the also-ran in a market that only wanted one chip name.

All three have since reclaimed their uptrends, which is the part the drawdown headlines skipped.

So the question changes shape. Not whether AI is real, but whether the companies whose valuations already got reset have more room than the ones printing new highs every week.

The Obvious AI Winners Are Already Crowded Trades

Marc Chaikin, founder of Chaikin Analytics, has been tracking this cycle against the 1990s tech boom, and roughly three years in, the AI trade has split into tiers.

The top tier is easy to spot, which is the problem. Dell Technologies NYSE: DELL, SanDisk NASDAQ: SNDK and NVIDIA NASDAQ: NVDA are printing new highs, and obvious beneficiaries make for crowded entries, per Chaikin.

His screen is narrower. Companies where AI is central to the business, where the market got wildly enthusiastic, pushed prices to an extreme, then had second thoughts. The ones worth revisiting are those whose AI revenue kept compounding through the correction while competitors fell away. Earnings quality does the tiebreaking, because plenty of software names carry bullish technical ratings with questionable financials underneath.

CapEx Fear Is Creating Better Entry Points

The fear sitting on AI stocks has a name: capital expenditure (CapEx).

Alphabet NASDAQ: GOOGL, Amazon.com NASDAQ: AMZN, Meta Platforms NASDAQ: META, and Oracle NYSE: ORCL are spending enormous sums on data centers and borrowing to do it. Companies that were once nearly debt-free no longer are, and investors want to know whether that spending converts into profit.

Chaikin's answer comes from the internet buildout. Fiber got laid faster than traffic could fill it, and the industry spent years talking about dark fiber before usage caught up. Compute is the same shape of problem. Capacity could eventually outrun demand and the market could react badly, but that point looks a couple of years out in his read, and the spending in the meantime is driving an earnings wave well outside tech, into the firms that build, electrify and cool the sites.

Politics adds a second layer. Data center construction has become a live midterm issue over electricity costs and local impact, and that pressure could ease once the votes are counted.

His cycle work supplies the timing. Midterm years tend to produce a low, and across roughly a century of data, the average run from that low into the pre-election-year peak is about 20% over the following nine months, pointing toward mid-2027. That bottom is forming now, he argues, visible in advance-decline statistics rather than in the headline S&P 500.

Data Platforms and Enterprise Software Reclaim Bullish Ratings

Palantir sits at the center of this. Government agencies use it to organize and reach data they already hold, and the company has strong political backing right now.

That exposure cut both ways. After a series of 2025 peaks, headlines tying data software to strikes during the Iran conflict painted the stock with a broad brush, and the drawdown ran past 50%.

Chaikin's rating went neutral through it and has since turned bullish again, with the stock up better than 60% over three months and an uptrend rebuilt while the price stays well below its old high.

The moat argument is less exotic than it sounds. Once software is embedded in a corporate or government workflow, ripping it out is expensive, the way Salesforce NYSE: CRM and Adobe NASDAQ: ADBE demonstrated when both sold off hard and then didn't go out of business.

ServiceNow ran a quieter version of the same script. Its suite handles support systems and integration for corporate America, and it is now using AI inside its own products to make them cheaper for subscribers.

One quarter of downbeat projections was enough to knock the stock down, because the market prices six months forward.

The rating is bullish again, and the uptrend is back in place.

What matters next is whether attached AI revenue keeps growing fast enough to carry the multiple, and Palantir's high beta means a 20% market move translates into considerably more in either direction.

A Second Source of AI Chips Becomes a Government Priority

Advanced Micro Devices broke the pattern by going straight to new highs.

Buyers of AI compute have decided they don't want a single chip supplier, and NVIDIA has been that supplier. AMD, long cast as the tortoise while it fought Intel NASDAQ: INTC, crossed a trillion-dollar valuation on the strength of being the credible second source.

There's a less-covered piece as well. AMD is central to a federally funded program, three years in development and now live, linking supercomputers across national labs from Oak Ridge in Tennessee to Lawrence Livermore in California, aimed at energy and medical research.

Chaikin's team flagged the name roughly six weeks ago after a 12% pullback, with money flow strong, the rating bullish and the stock short-term oversold. It has run about 30% since then, which creates an obvious problem for anyone arriving now. The approach isn't to pay new-high prices but to wait for the 5%-8% pullbacks a volatile name like this delivers regularly.

Rebuilt Uptrends Matter More Than New Highs Right Now

What connects these three isn't the AI label. It's the sequence of a real revenue stream, a violent repricing, then a rebuilt uptrend most investors haven't registered.

The setup leans on compute demand staying ahead of compute capacity and on earnings continuing to arrive on schedule. It doesn't depend on any of them quickly reclaiming old highs, and it doesn't require the hyperscaler CapEx debate to be resolved this year. It does require the midterm low to hold.

The risk is overcapacity arriving early, or political resistance to data centers outlasting the election. The upside is a second buying wave in names that have already absorbed a correction the market's AI leaders have not.

Stay focused on whether compute demand keeps outrunning capacity, because that's what decides how much of this earnings wave is left.

Readers can see Marc Chaikin's latest Power Gauge warning and stock alerts to follow how his ratings shift as the underlying data changes.