AI chip stocks have been volatile this year as investors weigh the enormous amount of spending going toward AI infrastructure and how sustainable it is. However, with cloud computing companies seeing strong returns on their AI chip and networking investments with quick payback periods and locked-in contracts, it appears that robust spending will continue.

Three semiconductor companies are now trading 15% or more below their recent highs, presenting what some analysts view as attractive entry points for long-term investors.

Broadcom: Custom Silicon Demand Keeps Building

Down 23% from its highs set this spring, Broadcom (AVGO) looks like one of the most compelling chip stocks to buy on the dip. The company is a leader in data center networking and custom AI ASICs, or application-specific integrated circuits, and has a significant growth opportunity ahead.

Broadcom helped Alphabet (GOOGL) develop its powerful Tensor Processing Units, or TPUs, which are set to be a major growth driver. Alphabet is spending aggressively on capital expenditures this year and has indicated it plans to spend significantly more next year, which should feed directly into Broadcom’s TPU and networking business. Alphabet has also allowed Anthropic to place TPU orders directly with Broadcom, adding another growth driver.

Given the success of TPUs, other hyperscalers have turned to Broadcom to help them create their own custom AI chips. Broadcom has projected it will see more than $100 billion in ASIC revenue in fiscal 2027, while Citigroup estimates that figure will rise to $180 billion in fiscal 2028.

With the stock trading at below 20 times forward analyst earnings estimates, the pullback appears to offer a reasonable valuation for a company positioned at the center of the custom silicon trend.

AMD: Betting Big on Inference and Agentic AI

Advanced Micro Devices (AMD) has been a hot stock in 2026, but the recent chip pullback has left it about 18% off its highs. That makes it a potentially attractive time to consider a company riding two of the biggest trends in AI: inference and agentic AI.

After losing to Nvidia (NVDA) in AI model training, AMD is making sure it will grab a piece of the larger and faster-growing inference segment. It formed two large partnerships with OpenAI and Meta Platforms (META) centered around inference, which helped give AMD a strong foothold in this market. At the same time, through its chiplet design, which can package more memory, and deals to acquire memory optimization company MEXT and inference chip company Taalas, AMD is aggressively positioning itself as a leader in this space. It also teamed up with Cerebras for a disaggregated system where its GPU-powered Helios solution will handle the pre-fill phase more cheaply, with Cerebras’ more expensive technology reducing latency.

As a leader in server central processing units, AMD is also set to ride the wave in agentic AI. AI agents are creating a huge need for advanced CPUs, and AMD sees this becoming a $220 billion market in the next few years.

SK Hynix: The Memory Bottleneck Play

Down around 20% from its high following its initial public offering this year, SK Hynix is a top memory stock to grab on the pullback. The Korean company is one of the big three DRAM makers and the market share leader in high bandwidth memory, or HBM.

HBM is currently the driving force in the memory market, as GPUs and other AI chips require this specialized form of DRAM to reduce latency and optimize performance. However, a combination of factors, including HBM requiring upwards of three times the wafer capacity of ordinary DRAM, is keeping capacity tight while demand continues to grow.

With long-term deals in place and as the main supplier of HBM to Nvidia, SK Hynix looks like the best-positioned memory maker over the long term. The company sees the market being imbalanced until at least 2030, although there is a good chance this supercycle lasts much longer. With a forward price-to-earnings ratio around 6 times, the stock looks like a buy on this dip.

The Broader AI Investment Picture

Global AI investment is expected to reach $900 billion in 2026, with the lion’s share, $730 billion, coming from the four largest US hyperscalers: Amazon (AMZN), Alphabet, Microsoft (MSFT), and Meta Platforms. The risk of losing market share due to a lack of computing power is larger than that of short-term margin compression, which is why the spending continues despite concerns about sustainability.

Global AI investment is forecast to rise further by 33% to $1.2 trillion in 2027, year over year. Massive capital flows to vendors that supply the specialized infrastructure equipment that powers critical AI workloads.

Nvidia’s Rubin generation, which entered full production in June 2026, stacks HBM4 memory delivering roughly 2.75 times the bandwidth of Blackwell’s HBM3e. That makes Taiwan Semiconductor Manufacturing Company (TSM), or TSMC, and SK Hynix indispensable partners rather than mere suppliers.

ETF Alternatives for Broader Exposure

For investors who prefer not to pick individual winners, exchange-traded funds offer a diversified approach. The VanEck Semiconductor ETF (SMH) tracks the largest global semiconductor companies and benefits from the current global AI chip crunch. Should AI growth broaden to include enterprise software, the Global X Artificial Intelligence & Technology ETF (AIQ), with broader exposure to the software and hardware sectors, offers another option.

Investors can also consider a core-satellite portfolio approach, anchoring the core position in diversified passive ETFs while actively managing the satellite portion exposed to key players that benefit from global AI investments.

CompanyTickerDecline from HighKey AI CatalystBroadcomAVGO23%Custom AI ASICs, TPUsAdvanced Micro DevicesAMD18%Inference, agentic AISK HynixKRX: 00066020%HBM memory leadershipVanEck Semiconductor ETFSMHN/ADiversified chip exposureGlobal X AI & Technology ETFAIQN/AHardware and software exposure

Note: Decline figures reflect approximate drawdowns from recent highs as of late August 2026.

What Investors Should Watch

Rather than trying to time short-term market movements, the key is to position in companies that generate long-term value. A high price-to-earnings multiple is not always bad; investors should ensure the company’s earnings growth is comparably high. Earnings are only good if most of them are converted into actual free cash flow, the lifeblood of a sustainable business.

The AI supercycle does not always move in a straight line. AI model winners like Alphabet and OpenAI drove the initial phases of the growth cycle from 2023. Today, players that anchor the most critical supply chain bottlenecks, such as high-bandwidth memory and the advanced expertise required to manufacture it, have taken over the momentum. As capacity constraints are resolved and enterprise monetization scales meaningfully, overall growth should broaden sustainably and include successful players across all phases.