AI Software Premium Repriced — ServiceNow Marks the Break

Weekly Executive Summary

• The market's central rerating this week was not about AI spending levels — it was about whether AI spending translates into earnings, and that test failed for software while hardware passed emphatically.

• Semiconductors extended a historic run — the SOXX rising 35% over the trailing month, ARM gaining 40% on the week, AMD up 24%, and Intel posting its best single day since 1987 — while the IGV software ETF fell 6%, making the hardware-software divergence the defining aggregate pattern of the week.

• Tone opened cautious as the NASDAQ snapped a 13-day streak on geopolitical nerves, rebuilt conviction through Alphabet's TPU reveal and Tesla's earnings beat, then landed in a fractured place where record index highs coexisted with ServiceNow's largest-ever single-day drop and IBM's post-earnings selloff.

Where the AI Cycle Repriced Itself

Semiconductors vs. Software: The market drew a hard line between hardware that builds AI and software that monetizes it

The week made concrete what had been a theoretical worry: investors are no longer willing to pay for AI potential in software; they demand demonstrated AI revenue. ServiceNow's 17.6% single-day collapse — despite 22% subscription growth and a raised AI contract target of $1.5 billion — was the sharpest expression of this repricing. IBM's near-7% post-earnings decline on the same logic reinforced it. The rotation was not subtle: capital moved out of the IGV ETF (down 6% on the week) and into the SOXX (up 35% over the trailing month), with hardware names like TXN, AMD, and ARM serving as the destination. The governing market logic is that physical infrastructure — chips, power, cooling — is a mandatory prerequisite whose revenue is visible now, while software AI monetization remains a future promise still awaiting proof.

AI Infrastructure Spending: Capex at scale is restructuring balance sheets and labor forces simultaneously

The $650 billion in projected industry-wide AI capital expenditure was not an abstract number this week — it showed up in headcount decisions at Meta (10% workforce reduction, 8,000 jobs) and Microsoft (voluntary buyouts targeting 7% of U.S. staff), and in capex commitments that are directly negative for near-term free cash flow. Tesla's decision to triple capex to $25 billion while projecting negative free cash flow during the investment cycle is the clearest single example of how the AI buildout is subordinating conventional financial metrics to a long-term infrastructure logic. Google's $40 billion commitment to Anthropic, structured as a phased investment starting at $10 billion, confirmed that the largest players are treating foundational model access as a balance-sheet priority, not a venture bet.

Vertical Silicon Integration: The competitive moat is shifting from who owns the model to who controls the chip

Across the week, the most durable competitive signals came from companies moving to own their own silicon. Google's launch of the TPU 8T (training) and TPU 8i (inference) — with cloud revenue growing nearly 50% and surpassing Azure's 40% — positioned it as the lowest-cost token provider through supply-chain ownership. Tesla's $3 billion Terrafab chip facility, built on Intel's 14A process, represents the same logic applied to robotics and autonomous vehicles. Rivian's in-house 800-TOPS silicon and Level 3 autonomous roadmap, supported by the Volkswagen deal, shows the pattern extending into EVs. The common thread: vertical silicon integration is being repriced from a cost-efficiency tactic into a primary source of competitive moat.

The CPU Renaissance: General-purpose chips re-entered the AI conversation as first-tier assets

Intel's record-high session and Texas Instruments' best day since 2000 were not coincidental — both reflect the market's recognition that the "agentic era" of AI, where AI systems take multi-step autonomous actions, shifts compute workloads back toward high-performance CPUs and analog components, not just GPUs. The SOX index reaching 42% above its 200-day moving average, a level not seen since 2000, confirmed that this broadening is happening at scale. Intel's foundry momentum — 18A production scaling 6x-7x in Q2, Data Center/AI revenue at $5.05 billion, and the Terrafab design win — transformed the company from a legacy concern into an active beneficiary of AI infrastructure nationalism.

The IPO Pipeline as a Liquidity Event: $200 billion in announced capital demand is already reshaping positioning

The week's framing of Anthropic and OpenAI as the defining IPO events of 2026 — with secondary market valuations for Anthropic approaching $350 billion post-Google's commitment — is not a forward story. It is already influencing portfolio construction. The concern surfaced explicitly: a $200 billion pipeline concentrated in a handful of frontier AI names risks redirecting institutional capital away from mid-cap and equal-weight tech. Cerebras Systems' IPO refiling at a $2 billion target, with gross margins declining from 42% to 39% despite doubled revenue, illustrated the pressure that highly concentrated customer rosters and competitive pricing create for smaller AI infrastructure companies trying to access the same capital pool.

The Companies That Defined the Week's Repricing

GOOGL (Alphabet): Vertical integration validated across silicon, cloud, and capital deployment

Alphabet appeared in every daily issue as the convergence point for the week's dominant themes. The launch of the TPU 8T and TPU 8i established a two-track silicon strategy — training and inference — that no peer has matched in public product form, while Google Cloud's 50% growth surpassing Azure's 40% gave the infrastructure argument a revenue number behind it. The $40 billion Anthropic commitment, structured to start at $10 billion against a $350 billion valuation, is a capital allocation decision with specific logic: it secures preferential compute placement for Anthropic's workloads on Google's own TPU infrastructure, creating a closed loop between investment and infrastructure utilization. The Marvell partnership on memory processing units adds a custom silicon layer beneath the TPU stack. The week's consistent message from Alphabet is that it is building a hardware-to-model supply chain that competitors will have to rent rather than own.

INTC (Intel): A foundry narrative that moved from credible to catalyzed

Intel began the week as a laggard — down 4% on Monday while the semiconductor index extended its streak — and ended it having posted its best single day since 1987, with shares touching a record session high of $85.22. The Q1 earnings beat ($13.58 billion vs. $12.36 billion estimated) was the trigger, but the substance beneath it was the Terrafab design win: Tesla selecting Intel's 14A process for its proprietary chip facility is external validation of a process node that the market had treated skeptically. The combination of 18A capacity scaling 6x-7x in Q2 and the framing of Intel as the "only viable domestic alternative" to TSMC — at a moment when geopolitical pressure on foundry sovereignty is high — gave the company a positioning argument it has not had in a decade. The week's trajectory for Intel was not a rebound; it was a reclassification.

TSLA (Tesla): The earnings beat was the least important thing that happened

Tesla reported EPS of $0.41 against a $0.34 estimate, with gross margins at 21.1% versus 17.7% expected — a clean beat that sent shares up 3% in after-hours trading Wednesday. But the week's real Tesla story was the capex commitment: $25 billion in 2026, more than double 2024's $11 billion, directed at Cybercab, Optimus, and the Terrafab chip facility. Free cash flow will turn negative during this cycle, and operating cash flow of $15 billion falls well short of the capex figure. The $3 billion Terrafab facility — built on Intel's 14A process — is the element that distinguishes Tesla's AI pivot from peers: it is not simply buying compute capacity but building a proprietary silicon supply chain for its robotics and autonomous stack. FSD subscribers reaching 1.28 million provides the demand-side anchor that justifies the capital intensity.

NOW (ServiceNow): The week's clearest signal on software's valuation problem

ServiceNow's 17.6% single-session decline — the largest in the company's history — was not a reaction to deteriorating business performance. Subscription growth ran at 22%, and the company raised its 2026 AI-specific contract target to $1.5 billion. The selloff was a verdict on the market's willingness to underwrite future AI monetization in software at current valuations, and it came despite metrics that would have been rewarded in prior quarters. The verdict was that growth without demonstrated AI pricing power is no longer sufficient. For a $15 billion-revenue enterprise growing above 20%, that represents a fundamental shift in the valuation framework applied to enterprise software, and its implications extend across every SaaS name that has embedded AI features without yet repricing them in contracts.

AAPL (Apple): A leadership transition landed in the middle of an unresolved AI identity

Apple's announcement that John Ternus will succeed Tim Cook as CEO on September 1, 2026 was the most consequential leadership event of the week, and the market's reaction — a $98 billion single-day market cap decline — reflected the timing as much as the selection. Ternus brings hardware engineering credentials at precisely the moment Apple's AI strategy remains externally dependent: the Siri overhaul relies on Google's Gemini for on-device data analysis, and AI spending is described as "magnitudes smaller" than rivals. The April 30 earnings call will be the first conducted under the shadow of the transition, and the degree to which Ternus can articulate a proprietary AI roadmap — rather than a partnership-dependent one — will likely define how the market rereads the succession premium embedded in the stock.

Sentiment, Catalysts, and What the Narrative Is Missing

The week opened with the NASDAQ snapping a 13-day winning streak on Monday — geopolitical anxiety and fatigue in mega-cap concentration providing the proximate causes — but the tone was not genuinely bearish. It was skeptical and waiting. The first real reset came Wednesday, when Alphabet's TPU reveal and Tesla's earnings beat arrived within the same session and restored directional confidence in hardware. The NASDAQ reached record highs, the SOX extended its streak, and for roughly 36 hours the week looked like an unambiguous hardware vindication story. Then Thursday reintroduced complexity: ServiceNow's collapse and IBM's guidance-less earnings beat broke the momentum, making clear that the rally's logic applied to physical infrastructure, not the software layer sitting above it. By Friday, the headline numbers — Intel at record highs, the S&P and NASDAQ at records, ARM up 40% on the week — created a surface reading of broad strength that the underlying sector dispersion did not fully support. The week ended with two markets: hardware in acceleration and software in a repricing that has not yet found a floor.

• The "$15 Trillion Wednesday" — the aggregate market cap of companies reporting in a single session around April 29 — will determine whether the current AI capex cycle receives earnings validation or triggers a correction in names that have been priced for perfection.

• Apple's April 30 earnings call, the first under the Ternus succession announcement, will be scrutinized for any articulation of a proprietary AI hardware or silicon roadmap beyond the current Gemini partnership.

• Intel's 18A production ramp — guided at 6x-7x volume increase in Q2 — is the near-term execution test for the foundry credibility thesis that drove the week's record session, with any slip in yield or customer conversion reversing the reclassification the market awarded.

The week's prevailing narrative framed the hardware-software divergence as a clean rotation: capital leaving software and entering semiconductors based on relative AI monetization visibility. That framing is accurate but incomplete, and it obscures a more specific tension. The companies most rewarded by the market this week — Intel, TXN, Vertiv, ARM — are beneficiaries of AI infrastructure spending, not AI revenue. Their earnings reflect what hyperscalers are buying, not what AI products are generating for end customers. Meanwhile, the companies punished — ServiceNow, IBM — are actually closer to the end-customer revenue layer, the place where AI monetization must ultimately be proven. The market is simultaneously demanding proof of AI monetization and selling the companies best positioned to provide early evidence of it, while rewarding suppliers whose revenue depends on the capex decisions of a handful of hyperscalers running negative free cash flow. That internal contradiction — penalizing enterprise software for not yet monetizing AI while rewarding chip suppliers for selling into a build-out that has not yet generated that monetization — is the tension the weekly narrative has not yet reconciled.

Important Disclosure

This newsletter is for informational purposes only and does not constitute investment advice. Content is generated by AI and may contain inaccuracies; always verify data independently before trading. Investing involves significant risk of loss. AlchemyJ is not a registered financial advisor. By reading this, you agree to our terms.

View our Full Legal & AI Disclaimer here