Franklin is optimistic that the AI infrastructure bull market will continue, and the recent decline in semiconductor stocks is a healthy correction.

Franklin Templeton investment strategist Katrina Dudley said the multi-sided case for AI infrastructure spending is overpowering the bearish case. Dudley noted that this round of investment theme could last at least until 2027, and even look ahead to 2028.
(Background: Viewpoint: Semiconductors take over from the “Magnificent Seven” to lead gains; the S&P 500 is preparing to break through 8,000 points.)
(Background supplement: Unfazed by Hynix’s stock price falling 30%; SK ADR’s U.S. share subscription is oversubscribed by 7 times; official trading begins 7/13)

How long will the AI wave continue? Franklin Templeton investment strategist Katrina Dudley said Thursday in an interview on Bloomberg TV that the bearish argument (meaning the side betting against this wave of investment) mainly focuses on two things: supply-chain efficiency not living up to expectations, and excessive capital expenditure.

Capital expenditure, simply put, is the money companies spend building data centers and buying chips. If they spend too much and it’s not put to use properly, it becomes the bubble warning bears love to point to most, and also the most tense topic in the market ahead of every earnings season. But Dudley observed that companies are actually quite rational when it comes to infrastructure investment—not blindly throwing money at the hot trend—and that the return on investment is enough to support continued spending.

Return on investment is the ultimate standard for judging whether this investment is worth it, and it is also the key dividing line at the core of her bull-versus-bear assessment. She therefore believes that what will truly win is the bullish argument: the sustainability of the AI theme can at least carry through the second half of 2026, 2027, and even into 2028. However, she also reminded that anyone who claims, with chest-thumping confidence, that they can see the trend after 2028 is probably being incomplete. Investors can listen, but don’t treat it as gospel—and there is no need to stake their entire net worth on a single optimistic forecast. After all, forecasts have limits; the farther out the forecast, the more it deserves to be discounted.

A semiconductor pullback: a warning or price discovery?

The rally in AI concept stocks this year has indeed been astonishing. The Philadelphia Semiconductor Index has surged 78% since the beginning of the year, and in the second quarter of this year it posted the strongest quarterly performance since records began in 1994. The magnitude of the gains has been so fierce that even seasoned hands have called it rare.

But chip stocks have recently swung violently despite still-strong fundamentals. Within a week, the Philadelphia Semiconductor Index (SOX) fell sharply by 8%, and Intel was hit by selling pressure as well, causing some investors to start questioning whether the rally has run out of steam. Doubt has been significant, and bearish discussions on social media have clearly increased.

Dudley interprets this pullback as price discovery. In simple terms, the market is helping stocks find a reasonable level, which is a healthy, normal process—not a structural warning. The two are completely different in nature, and investors should not treat every correction as an end-of-the-world signal. She specifically pointed to the production capacity data released by SK Hynix and Samsung Electronics as being quite rational. Samsung’s stock price once plunged more than 10% in Seoul. Even though its quarterly operating profit surged 19 times, the strong financial results still led to a drop in the stock price—there’s a stark contrast.

On the other hand, SK Hynix is about to list in the U.S., with an offering subscription ratio of about 7 times, exceeding the 2019 listing record of Saudi Aramco’s $25.6 billion. The funds raised will be used for capacity expansion and EUV equipment procurement—positioning SK Hynix ahead of the next round of the capacity race. It also allows the capital market to reassess the pricing logic of the memory sector, because signals of a memory shortage are more direct than any earnings call.

Positive return on investment is the real signal

Dudley said that the signal she is truly focused on is actually very simple: whether the capital deployment companies announce is an investment with positive return on investment. As long as this threshold holds, the AI spending cycle will continue creating long-term value, and she believes the data seen so far still supports that this threshold has not been broken.

Market estimates show that this year the capital expenditure commitments of hyperscale cloud providers have already reached about $750 billion, and high-bandwidth memory production capacity has already been sold for much of 2027. This means visibility into orders is already arranged for the next two or three years. This is not optimism that’s just shouted out with words—it’s optimism backed by actual orders. The level of confidence is completely different, and it is not a paper plan that can be called off at any time.

When return on investment is still positive, the bulls have the confidence to continue adding positions, treating every pullback as an opportunity to get on board. But once one day return on investment turns negative, no matter how high the rally is, in the end it is only a numbers game built on inventory. The higher it rises, the harder it will fall later.