Is AI infrastructure still a buy after the market sell-off?
After a steep correction that knocked the S&P 500 down 4.5% and the Nasdaq 7%, the source material still points to large AI infrastructure spending. The more supportable question is whether that spending is showing up as revenue visibility, cash generation or merely large financing needs.
Why AI capex is accelerating and what it means for investors
One source projects data-center spending to approach $700 billion in 2026 and potentially $1 trillion annually by 2028. The source set also cites Meta’s $600 billion U.S. data-center expansion with free technician training, Oracle’s $55.7 billion fiscal 2026 capex, Microsoft’s $190 billion fiscal 2026 capex outlook, and SpaceX’s planned $75 billion IPO raise, potentially more than $86 billion with overallotment. Deloitte’s cited data show 93% of enterprise AI spending going to technology and 7% to human capital; PwC says 20% of companies capture nearly 75% of AI’s economic value.
Chip winners versus order-flow players
1. NVIDIA reported 85% YoY revenue growth to $81.6 billion and a 74.2% gross margin, while shares slid more than 6% in a semiconductor pullback. The source also noted valuation and margin headwinds, including rising memory costs and hyperscaler in-house hardware efforts.
2. Broadcom reported $22.2 billion Q2 fiscal 2026 revenue, with AI revenue of $10.8 billion, up 143%, and a target of more than $100 billion in AI revenue in fiscal 2027.
3. AMD rose after Bank of America raised its 2030 server-CPU TAM estimate from $125 billion to $170 billion, citing AI-driven demand for server CPUs.
4. Micron shares rose after Wolfe Research projected DRAM prices roughly 200% above end-2025 levels by end-2026 and NAND prices up 216%, driven by AI-centered demand and limited supply.
5. Super Micro Computer announced about $39 billion of new AI server orders from more than 20 customers and a roughly $7 billion equity/equity-linked financing plan. The orders are not firm commitments, and the source highlights dilution, working-capital needs and low/volatile margins.
AI capex giants: Oracle, Microsoft, Meta and SpaceX
Oracle. Oracle disclosed $75 billion of prepaid GPU contracts, an initial deployment of up to 800,000 GPUs and about 1 GW of capacity this quarter, while fiscal 2026 capex rose to $55.7 billion and RPO reached $638 billion. Management projected $70 billion of net cash outlay in fiscal 2027 and plans to raise about $40 billion through debt and equity. Its AI spending is directed to GPU and server suppliers including Nvidia, AMD and Dell.
Microsoft. Microsoft projects fiscal 2026 capex of $190 billion. Its AI segment now generates more than $37 billion annually, with Azure and Microsoft 365 Copilot driving growth, and Copilot has reached 20 million paid seats.
Meta. Meta’s $600 billion U.S. data-center expansion and free technician training program indicate a large AI infrastructure commitment. The source material does not establish the funding mix or quantify execution risk.
SpaceX. SpaceX plans a June 12 IPO at $135 per share, targeting about a $1.77 trillion valuation and a roughly $75 billion raise, potentially $86 billion with overallotment. The company reported 2025 revenue of $18.7 billion and a $4.9 billion net loss, implying a valuation near 95 times 2025 revenue. Its S-1 frames a large addressable market tied to enterprise AI, but the source also notes IPO volatility risk and limited near-term catalysts.
Power, grid and construction beneficiaries beyond chips
Data-center buildouts require power and connectivity. Chaikin Analytics identifies Quanta Services for U.S. electrical-grid expansion, MasTec for electrical, pipeline and communications infrastructure, and Argan for power-plant development as companies linked to the AI infrastructure cycle. QXO agreed to acquire TopBuild for $17 billion; TopBuild reported more than 17% year-over-year quarterly sales growth, which the source tied to insulation work for AI data centers.
Viasat won a $219 million U.S. Space Force contract under the Protected Tactical SATCOM-Global program, with potential follow-on work under a $4 billion program ceiling. The source presents this as a defense communications win and validation of geosynchronous satellite infrastructure demand.
Oklo’s Preliminary Documented Safety Analysis for the Aurora reactor was approved by the DOE, reducing one regulatory hurdle. The source still describes the stock as speculative because further approvals and capital-intensive execution remain.
AI software, ARR and enterprise adoption
Adobe. Adobe reported fiscal Q2 2026 revenue of $6.62 billion, up 13%, total ARR of $27.1 billion, and AI-first ARR above $500 million, up threefold year over year. The source also cited operating-margin targets above 45% and $2.17 billion of operating cash flow.
Abridge. Abridge raised $830 million at a $5.3 billion valuation, serves 300+ health systems and 250 million patients, and has strategic partnerships with Eli Lilly and NVIDIA. The cited risks are regulatory complexity around data security and AI-generated medical records.
Enterprise adoption. The source material suggests enterprise AI value is uneven: Deloitte reports 93% of AI spending going to technology versus 7% to human capital, ServiceNow says about 90% of enterprise AI use cases focus on internal productivity and cost management, and PwC says 20% of companies capture nearly 75% of AI’s economic value.
Valuation risk and market timing
BofA strategist Savita Subramanian flagged seven of ten bear-market signposts, including the “Rule of 20,” extreme rich/poor valuation gaps, sentiment indicators and credit stress signals. The semiconductor sector fell 10.3% on June 5, with $1 trillion of market value lost.
May inflation was 4.2% and energy-driven, which one source says may temporarily delay Fed rate cuts. High-multiple or leverage-sensitive examples in the source material include SpaceX, CrowdStrike and Strategy.
How to think about the examples
Core examples to evaluate: Adobe, NVIDIA and Microsoft all have source-supported AI revenue, ARR, capex or gross-margin data, but each also carries valuation, capex or margin questions that investors should weigh.
Watchlist examples: Oracle, Broadcom and AMD have visible AI-related demand signals, including Oracle’s prepaid GPU contracts and RPO, Broadcom’s AI revenue growth, and AMD’s raised server-CPU TAM estimate.
Speculative examples: SpaceX, Super Micro, Oklo and Abridge each have large opportunity sets or partnership/order signals, but the sources also flag valuation, dilution, regulatory, profitability or data-security risks.
What would make me change my mind?
Key variables to monitor include whether Oracle’s prepaid GPU contracts and RPO convert into cash generation after heavy capex and financing; whether Microsoft’s AI usage and agent pricing can offset its $190 billion capex outlook; whether SpaceX’s AI capex and contracts support revenue enough to justify its valuation; whether memory price forecasts for Micron materialize; and whether Oklo advances from safety approval toward commercial deployment.
FAQ
Should I tilt my portfolio toward AI chips or the broader infrastructure ecosystem? The sources show strong AI chip revenue growth and separate sources for grid, construction and satellite/defense infrastructure. They do not prove that one group has a better risk-adjusted return.
Is the SpaceX IPO a must-have for an AI-focused portfolio? The source material frames SpaceX as loss-making with a valuation near 95 times 2025 revenue, so it is better described as a speculative IPO candidate than a must-have.
How do prepaid GPU contracts affect Oracle’s risk profile? They increase visible demand and RPO, but Oracle also disclosed very large capex and plans to raise about $40 billion through debt and equity.
Do AI software ARR metrics outweigh hardware spending in determining winners? The sources do not prove that ARR outweighs hardware spending. They show that ARR can provide visibility, while hardware spending shows the scale of capex demand.
What macro signals could trigger a renewed sell-off in AI stocks? The source material points to persistent inflation, delayed Fed rate cuts, credit stress signals and semiconductor volatility as risks for high-multiple names.
Conclusion – A practical checklist for investors
- Separate AI exposure from cash conversion and revenue visibility.
- Treat prepaid contracts, RPO and disclosed backlogs as evidence of demand, but weigh them against capex and financing plans.
- Treat non-binding order announcements as less certain than revenue, backlog or contracted demand.
- For infrastructure beneficiaries, focus on disclosed contracts, revenue growth or earnings beats rather than broad theme exposure alone.
- For speculative names, focus on valuation, profitability, dilution, regulatory status and data-security risk.
The source material supports a large AI infrastructure spending cycle, but it also shows uneven enterprise adoption, valuation risk and financing needs. Investors should match each AI exposure to its evidence of revenue visibility, cash conversion and risk tolerance.
This article is for informational purposes only and does not constitute individualized investment advice.