Google's talent exodus sparks heated debate! 'Cash-burning giants' under pressure, 'shovel sellers' hit new highs—where is AI capital flowing?

Current market dynamics are increasingly showing thatthe market is no longer buying into 'AI storytelling,' but instead flocking to 'pick-and-shovel sellers'—companies that generate stable profits and hold pricing power over essential needs.

On June 22, Alphabet, Google’s parent company, dropped nearly 5%,marking its steepest single-day decline since May 7, 2025, and wiping out approximately $225 billion in market value—the largest such loss in its history.Its performance that day significantly underperformed the broader Nasdaq index. $Alphabet-C (GOOG.US)$$Alphabet-A (GOOGL.US)$

The immediate trigger—Within less than a week, two key figures in AI have departed one after another, both joining Google's most direct and cutting-edge competitors.

Jumper had worked at Google for nearly nine years, while Shazeer was someone Google spent approximately $2.7 billion to effectively 'buy back' through the Character.AI deal—only to leave again less than two years after rejoining.

But what truly prompted the market to 'vote with its feet'was not the departures themselves, but rather the concerns these events triggered about Google’s position in the AI competitive landscape:

(1) Organizational disadvantages in talent competition: Gil Luria, Managing Director at D.A. Davidson, pointed out thattop-tier AI research talent is extremely scarce, and startups like OpenAI and Anthropic are systematically attracting elite talent who previously favored large tech companies, thanks to more flexible incentive structures, flatter organizational hierarchies, and clearer, more focused missions.In the AI race, talent competition is paramount; for Google, this is not merely a talent retention issue, but a direct reflection of its organizational effectiveness and competitiveness in innovation culture.

(2) Imbalance between surging capital expenditures and visibility into monetization: Google has raised its fiscal year 2026 capital expenditure guidance to a range of $180–$190 billion and explicitly stated it will 'significantly increase' spending further in 2027. Yet, first-quarter free cash flow declined by approximately 47% year-over-year to $10.12 billion.When capital expenditure continues to exceed market expectations in both scale and growth rate, yet differentiated advantages in commercialized products remain unestablished, market skepticism about return on investment begins to mount.

(3) Potential pricing pressure: Public remarks by a founder of Google Ventures (GV)—recommending an 80% price cut for Gemini API tokens to undercut competitors—have instead heightened market concerns about pressure on Google's profitability.If a price war on the inference side becomes a reality, it will further compress already tight profit margins constrained by high capital expenditures.

"Cash-burning behemoths" under pressure: concerns over AI ROI emerge

Google's case is not an isolated incident but reflects the widespread financial pressure large tech companies face under the 'AI arms race' model. According to reports, $Amazon (AMZN.US)$ 、 $Alphabet-C (GOOG.US)$ 、 $Microsoft (MSFT.US)$ 、 $Meta Platforms (META.US)$ the combined 2026 capital expenditure forecasts for the four tech giants have already reached approximately $695 billion to $725 billion, representing a year-over-year increase of over 60%. This scale exceeds the annual fiscal spending of most mid-sized developed economies.

Goldman Sachs analysts noted that AI-related capital expenditures are rising faster than actual data center construction, meaning future bottlenecks will shift from model demand to financing capacity, power supply, and project execution. While the four tech giants race to dominate the AI battlefield, they each face distinct situations.

$Alphabet-C (GOOG.US)$is currently one of the few giants capable of seamlessly monetizing AI and generating tangible profits, with its cloud business serving as the most solid revenue channel: Q1 cloud revenue hit $20 billion (+63%), operating margin surged from 17.8% to 32.9%, and remaining performance obligations (RPO) doubled to $462 billion. However, the CFO explicitly warned that 2027 capex will see 'significant increases,' and Bank of America estimates that 2026 free cash flow (FCF) will plummet from $73.3 billion to roughly $44.1 billion. When capex growth far outpaces revenue growth, it directly erodes free cash flow and margins in the short term, raising concerns about 'revenue growth without profit growth.'

$Meta Platforms (META.US)$is the only one among the four that does not offer external cloud services; returns on AI investments can only be indirectly reflected through 'advertising efficiency gains'. Although Q1 revenue grew by 33% and operating margin reached an impressive 41%, capex nearly doubled compared to 2025, and the CFO admitted the primary reason for the increase was 'component price hikes' (DRAM contract prices rose 95% quarter-over-quarter in Q1), not strategic capacity expansion. The risk of uncontrolled capital spending and unclear monetization pathways makes its credibility the most fragile among the 'tech giants.'

$Microsoft (MSFT.US)$and $Amazon (AMZN.US)$The core business models of both are similar: build data centers → purchase GPUs or develop proprietary chips → sell computing power. Microsoft’s moat lies in its end-to-end full-stack flywheel—Azure computing power + Copilot subscriptions + GitHub Copilot Enterprise—which creates a 'per GPU to per dollar' reconciliation chain., RPO of nearly $627 billion is approximately 2.9x annualized cloud revenue (~$218 billion), offering the longest revenue visibility among the four. However, Nadella himself has publicly discussed the 'risk of AI commoditization,' and maintaining long-term competitiveness remains structurally challenging.Amazon AWS boasts an operating margin as high as 35%, serving as the company's profit engine. Its vertical integration strategy (in-house chips like Trainium/Graviton → reducing reliance on NVIDIA and improving long-term gross margins) gives its Capex a 'strategic asset-building' characteristic., but at the cost of all three initiatives simultaneously consuming cash (Trainium ecosystem + data center expansion + Project Kuiper satellites), resulting in relatively weak short-term cash flow flexibility.

CICC Global Research noted,The tech giants’ race to ramp up investments resembles a prisoner's dilemma: everyone knows over-investment is likely, yet no one dares slow down first.Balance sheets remain healthy in the short term, providing a 1–2 year window for commercialization to materialize; however, if ROI remains unclear within this window, valuation compression could shift from multiple contraction to a deeper reassessment.

‘Pick-and-shovel sellers’ keep hitting new highs: the earnings-driven hardware realization phase

In stark contrast to valuation pressures faced by downstream giants, upstream AI hardware and infrastructure suppliers are now in a golden period of earnings realization.

Memory: The tightest supply-demand balance—AI’s 'hard currency'

June 22, $Micron Technology (MU.US)$ Closed at $1,211.38, with a total market capitalization of approximately $1.37 trillion, hitting a new all-time high, and posting a single-day gain of nearly 7%. Market focus is centered on the Q3 earnings report to be released after market hours on June 24:

Management guidance: revenue of $33.5 billion (±$750 million), Non-GAAP EPS of $19.15 (±$0.40), and gross margin of approximately 81%; consensus market expectations have been revised upward to EPS of approximately $20.57 and revenue of around $35 billion, significantly exceeding the upper end of management's guidance.

The core thesis is that HBM (High Bandwidth Memory) has become the physical bottleneck for scaling AI computing power.Only three companies globally ( $Micron Technology (MU.US)$、 $SK Hynix (000660.KR)$、 $Samsung Electronics (005930.KR)$) currently possess mass production capabilities, and these three major manufacturers are actively shifting capacity toward HBM and DDR5, squeezing supply of conventional DRAM and NAND.The supply-demand gap is expected to persist through 2027. Additionally, Micron, Samsung, and SK Hynix have all secured strategic partnerships with Anthropic in dual roles as both shareholders and suppliers, highlighting the strategic scarcity of upstream capacity.

Recently, the 'Memory Seven' (Micron, SanDisk, $Seagate Technology (STX.US)$ 、 $Western Digital (WDC.US)$ 、 $Kioxia Holdings (285A.JP)$ , Samsung, and SK Hynix) have shown strong performance. Aside from Micron, SK Hynix reached a new all-time high on June 22, surpassing Samsung Electronics intraday and ending Samsung’s more than 20-year reign as South Korea’s top company by market capitalization since 2000, with its stock price rising nearly 300% cumulatively over the past 26 years.

Semiconductor Equipment: WFE Enters a Super Cycle

For decades, the semiconductor industry has exhibited clear cyclicality, but UBS Group believes this time is different, as the emergence of AI is gradually transforming the semiconductor sector from a traditionally cyclical industry into a long-term growth industry. According to UBS Group's latest reportSignificantly raised its global WFE forecast for the next three years and provided a very aggressive outlook,The semiconductor industry is in the early stages of a new supercycle.

UBS Group stated that the largest incremental capital expenditure over the next few years will actually come from the memory segment, particularly DRAM.To meet future HBM demand, multiple large-scale memory fabs will come online globally over the next few years, and these projects will continue to drive equipment procurement demand.。UBS Group has comprehensively raised its equipment investment forecasts for the next three years, revising WFE upward to $147 billion in 2026, $198 billion in 2027, and $247.5 billion in 2028.

According to publicly available data, the nine U.S.-listed semiconductor equipment leaders with market caps exceeding $10 billion have all gained more than 75% year-to-date, seven of which have doubled, and all have hit all-time highs. $Applied Materials (AMAT.US)$ 、 $KLA Corp (KLAC.US)$ 、 $Lam Research (LRCX.US)$ 、 $MKS Inc (MKSI.US)$ 、 $KLA Corp (KLAC.US)$ 、 $Teradyne (TER.US)$

Optical Communications: A Reflection of AI Cluster Interconnect Bottlenecks

As AI clusters scale from tens of thousands to hundreds of thousands or even millions of GPUs, interconnect bandwidth between GPUs has become a new performance bottleneck.Japanese fiber manufacturer Fujikura announced a 30% price increase on DCI (Data Center Interconnect) specialty optical cables due to surging demand from North American AI data centers and constrained capacity expansion, while simultaneously raising its earnings guidance.

This news has recently driven strength across the entire sector! $Corning (GLW.US)$ Up approximately 7.65% in a single day, the stock price hit a new all-time high of $212.39. The company is planning to expand production capacity to meet NVIDIA data centers' fiber optic demand; Hong Kong-listed fiber optic leader $YOFC (06869.HK)$ Surged over 31% in a single day; A-share $Yangtze Optical Fibre And Cable Joint Stock (601869.SH)$ Locked at the daily trading limit, with total market capitalization exceeding RMB 300 billion.

The ongoing transition from 800G to 1.6T optical modules is driving a 3x to 5x increase in fiber usage per cluster. Given that the expansion cycle for optical fiber preforms takes 18 to 24 months, supply tightness is expected to persist through 2026–2027, offering sustained tailwinds for the optical communications sector.

Conclusion

The long-term trend of the AI industry remains intact, but capital markets have entered a more cautious phase,Over the past two years, capital chased the 'biggest story'; now, the focus has shifted to 'who offers the strongest certainty and the most stable pricing power.'This round of structural repricing is reshaping the investment landscape of the AI era.

1.2M Views