The current AI investment logic has undergone a disruptive shift, yet most ordinary investors are still stuck in the first-half GPU-compute mindset, staring at Nvidia, data centers, memory chips, and other hardware names. Top smart money has already changed lanes. Cathie Wood’s ARK Fund latest May 2026 rebalance fully exposes the new direction: large-scale reduction of traditional chip hardware positions, partial liquidation of compute names, and aggressive positioning in AI vertical applications, enterprise AI productivity, and AI commercialization.
It’s not just Cathie Wood—Musk is also adjusting at the same time, downplaying Tesla’s near-term earnings story and fully betting on robots and edge AI deployment. The consensus among top-tier capital is very clear: AI’s first half makes money by selling shovels; AI’s second half makes money by monetization and deployment.
This article breaks down the three core money-making themes of AI’s second half, divided into the certainty layer, the controversial growth layer, and the earnings-validation layer, corresponding to Broadcom, Microsoft, and Meta. It reconstructs the positioning logic of major funds and helps you rebuild your AI investment framework. If you missed Nvidia and Micron’s multi-bagger move in the first half, don’t miss the structural opportunity in AI’s second half.
Layer 1: The most certain theme in AI’s second half | AI infrastructure (Broadcom)
The first opportunity in AI’s second half is the most certain and least likely to fail AI infrastructure theme, and the core name is Broadcom—also a key area ARK has been adding to in this rotation.
Most investors are still locked into the idea that AI is only about Nvidia GPUs. But with large-scale AI model adoption and the full rollout of enterprise AI agents, the industry’s key pain point has shifted from ‘not enough training compute’ to too-high AI inference costs. Daily AI office work, AI search, AI ads, smart customer service, and frequent large-model calls create massive inference demand. Traditional general-purpose GPUs are simply too expensive for large-scale commercialization.
Against this backdrop, custom ASIC AI chips have become the best solution and the core infrastructure AI deployment cannot avoid. Broadcom is the largest hidden enabler behind the world’s cloud giants. To reduce dependence on Nvidia, Google, Amazon, and Meta are all racing to develop custom AI chips in-house, and key solutions like TPU, Trainium, and Inferentia are all deeply tied to Broadcom’s ASIC design, high-speed interconnect, and networking-chip technology.
Earnings data fully validates the growth thesis: Broadcom’s AI-related revenue in fiscal 2026 Q1 reached $8.4 billion, up 106% year over year; Q2 AI revenue guidance exceeded $10.7 billion. Quarterly AI revenue now exceeds the full-year scale of most semiconductor companies, marking Broadcom’s complete transformation into a core AI second-half infrastructure leader and its escape from the traditional chip-company label.
Industry data shows that after Google TPU was optimized based on Broadcom’s solution, inference costs in some scenarios can drop by as much as 70%, a cost advantage that crushes traditional GPUs. Future AI competition will no longer be about model parameters; it will be about the cost-efficiency ratio of AI compute, and Broadcom is perfectly positioned.
Management clearly projects that in the next few years, the company’s AI revenue could exceed $61–90 billion. Current AI order backlog is healthy, and the order cycle has already stretched years into the future. Using a 35–45x forward P/E for a top AI infrastructure leader, Broadcom’s future market cap has more than 50% upside.
Key risk warning: high customer concentration, long-term coexistence of GPUs and ASICs, and relatively large short-term sentiment-driven swings. Overall, Broadcom is the preferred core defensive allocation for AI’s second half.
Layer 2: The high-beta theme in AI’s second half | Enterprise AI productivity (Microsoft)
The second opportunity in AI’s second half is the most controversial, most underappreciated, and highest-beta enterprise AI productivity theme, with Microsoft as the core name. It is also the super-platform most likely to be underestimated in this AI cycle.
Many people wonder: Microsoft was an early AI leader, so why has its stock been weak over the past year? The core reason is not a lack of AI technology, but Wall Street’s long-running debate: is Microsoft’s massive AI capex a cash-burning black hole, or a brand-new profit engine?
Microsoft has fully integrated AI into its entire product suite: Windows Copilot, Office Copilot, GitHub Copilot, Azure AI, and enterprise intelligent-office ecosystems are all in place. The biggest market disagreements center on three things: Copilot monetization, Azure AI growth durability, and the margin pressure from high capex. Exactly this controversy creates the huge expectation gap.
The latest earnings report has already challenged the skeptics: Microsoft’s fiscal 2026 Q1 total revenue was $82.9 billion, up 18% year over year; Azure cloud revenue grew 40%; annualized AI revenue surpassed $37 billion, with growth above 120%. Meanwhile, Microsoft 365 Copilot paid seats surpassed 20 million, proving that enterprise willingness to pay for AI is real, not just hype.
Microsoft’s core moat is that it doesn’t need to build a user ecosystem from zero; it can directly empower a global billion-user enterprise workflow. Must-have work tools such as Word, Excel, Teams, GitHub, cloud development, and enterprise security have all been rebuilt with AI, completely rewriting white-collar work, programming workflows, and enterprise operations, and materially raising productivity across the whole enterprise stack.
At the same time, Microsoft’s internal AI transformation continues to improve operating efficiency, validating the commercial loop of AI for cost reduction and productivity gains.
Valuation upside estimate: Microsoft’s market cap is currently $3.1 trillion, while its forward P/E is only around 22x, putting it near historical lows. If Copilot monetization keeps exploding and Azure growth remains strong, the company could re-rate to above 30x P/E, implying a share price of $540–700 and potentially doubling market cap.
Key risks: overly high AI capex, slower-than-expected enterprise adoption, and competition from Google and others. Best suited for investors seeking high beta and betting on an expectation gap.
Layer 3: The earnings-validation theme in AI’s second half | Commercial efficiency delivery (Meta)
The third opportunity in AI’s second half is the most solidly supported by real earnings theme in commercial efficiency, with Meta as the core name. It perfectly captures AI’s leap from ‘telling stories’ to ‘making real money.’
The market still thinks of Meta as social media and the metaverse, but today Meta’s core value is rebuilding the global ad business with AI, making it the tech giant with the most complete AI commercialization story.
Meta’s core profit engine is advertising, and AI is reshaping ad delivery end to end: AI-generated creatives, precise audience matching, dynamic budget optimization, and real-time ROI improvement. Its Advantage+ smart ad system lets small and midsize businesses run sophisticated campaigns without specialized operators, greatly improving conversion efficiency across the industry.
The earnings data strongly validates the thesis: Meta’s fiscal 2026 Q1 total revenue was $56.3 billion, up 33% year over year; ad revenue was $55 billion, also up 33%; average ad price rose 12% year over year. The logic behind rising ad prices is that AI improves delivery performance, and advertisers are willing to pay more for higher ROI—this is the signature sign of AI entering the income statement.
Beyond ads, Meta AI search, AI business assistants, and the AI agent ecosystem continue to roll out, allowing Meta to fully escape a pure social-platform identity and transform into a global top-tier AI commercial services platform.
The supplementary name Uber also confirms the theme: through AI dynamic pricing, smart dispatch, route optimization, and automated customer service, platform efficiency has improved dramatically, with Q1 2026 operating profit up 42% year over year, further validating the broad applicability of AI cost reduction and efficiency gains.
Valuation upside estimate: Meta’s market cap is currently $1.8 trillion. As AI advertising continues to improve and commercialization broadens, using a 28–32x P/E for an AI platform suggests the stock could reach $900–1,000, with the market cap nearing a doubling.
Key risks: still-high AI capex and concerns about repeating the metaverse cash-burn story. But AI is already delivering real profits, which is fundamentally different from the pure-concept stage of the metaverse.
III. Full summary: the core investment logic of AI’s second half
The core of AI’s first half was compute scarcity—fighting over GPUs, hardware, and production capacity; the core of AI’s second half is commercialization—fighting over efficiency, profits, and enterprise productivity gains. Cathie Wood’s latest rotation precisely caught the industrial turning point, with major capital moving decisively from hardware shovels to application monetization.
These three themes correspond to three types of opportunities, fitting different investing styles:
1. Certainty layer | Broadcom: essential AI ASIC infrastructure, defensive and stable, with 50%+ upside, suitable for conservative investors;
2. Growth layer | Microsoft: the enterprise AI productivity leader, rich in expectation gap and offering double-digit upside, suitable for event-driven investors;
3. Validation layer | Meta: real AI earnings delivery, mature business model, suitable for most ordinary investors.
The AI rally is not over—it has simply changed leadership entirely. Leaving hardware overcrowding behind and embracing AI applications and productivity transformation is the true wealth theme for AI’s second half in 2026.
