Three earnings reactions in forty-eight hours proved where AI value accrues. Hardware companies posted their best days in twenty-five years while enterprise software cratered despite beating estimates — the market has learned to read AI earnings.
Three earnings reactions in forty-eight hours told the entire story of the AI economy.
Intel reported Q1 revenue of $13.58 billion against expectations of $12.42 billion. Adjusted earnings came in at twenty-nine cents per share versus one cent expected. Data center revenue climbed twenty-two percent to $5.1 billion. The stock surged twenty percent after hours, crossing the all-time high it set in 2000 — a quarter century ago. Intel is up more than eighty percent in 2026.
Texas Instruments posted $4.83 billion in revenue, beating estimates by three hundred million dollars. Data center revenue surged ninety percent year over year. The stock rose sixteen percent in a single session — its best day since October 2000. Bank of America upgraded the stock to Buy and raised its price target from $235 to $320.
ServiceNow reported $3.77 billion in revenue. Beat estimates. The stock fell as much as eighteen percent. Down thirty-three percent on the year.
Read that sequence again. Two semiconductor companies posted genuine AI revenue acceleration, and the market gave them their best days in twenty-five years. An enterprise software company beat its number and got destroyed — because Middle East deal delays, margin pressure from a $7.75 billion acquisition, and a GAAP earnings miss told the market that AI-powered workflows are not the same thing as AI revenue.
IBM confirmed the pattern. Revenue of $15.92 billion beat expectations. Earnings of $1.91 per share beat by ten cents. The stock fell seven percent because management maintained its full-year guidance rather than raising it. Down fifteen percent in 2026. When you beat and the market sells, the market is telling you the beat does not count.
The Thesis
AI value accrues to the physical layer. The market has debated this for two years, and this week it answered.
Companies selling chips that AI systems physically require are being rewarded at rates not seen since 2000. Companies bolting AI features onto existing software at premium valuations are being punished — even when they deliver the revenue number. The market has learned to distinguish AI substance from AI narrative, and the distinction is binary: either you have chip revenue from data centers, or you do not.
The distinction matters because it is structural, not cyclical. Data center buildout has years of runway — the top five hyperscalers have committed roughly $660 billion in capital expenditure for 2026, with approximately seventy-five percent flowing to AI infrastructure. That spending physically requires analog semiconductors, power management chips, networking silicon, and electricity. It does not require another workflow automation platform.
The Positions
If you believe this thesis, here is what it implies.
Long the physical layer. Texas Instruments makes the analog chips every AI system depends on — power management, signal processing, sensor interfaces. Not the GPUs that get headlines, but the unglamorous silicon that makes GPUs functional. Data center revenue grew ninety percent year over year. Unlike GPU design, analog chip IP requires decades of proprietary process knowledge. There is no DeepSeek equivalent for analog semiconductors. Lower concentration risk than NVIDIA, higher margins than AMD, and genuine moat.
Intel is the turnaround the market was right to doubt for five years and now has six consecutive quarters of data to believe. Data center revenue is accelerating into AI demand. The stock trades at a fraction of historical multiples.
Power infrastructure — Vistra, Constellation Energy, and the independent generators — supply the electricity the buildout requires. Alphabet issued a hundred-year bond to finance data center construction. That is the demand timeline.
Avoid the premium SaaS names. ServiceNow down thirty-three percent in 2026 is not cheap — it is being repriced for a world where AI agents replace the workflow automation ServiceNow sells. IBM's fifteen percent decline reflects the same dynamic: consulting revenue is real but insufficient when AI agents can do the advising.
The broader enterprise SaaS cohort trading at ten-plus times revenue faces a structural squeeze. AI features add integration and compute cost without proportional revenue. The narrative premium that held these stocks up through 2025 is collapsing into earnings reality.
Convergence Week: April 29
Alphabet, Microsoft, Meta, and Amazon all report earnings on April 29 — the same day the Federal Reserve announces its rate decision.
The question for each: are you showing AI revenue, or AI spending?
Meta just committed $115 to $135 billion in 2026 AI capital expenditure while cutting eight thousand employees. If the market rewards the combination of infrastructure commitment and headcount reduction, the thesis strengthens. If Meta's AI revenue line disappoints despite the spending, even the infrastructure winners face a demand question.
Microsoft's Azure AI revenue growth rate is the single most important data point in technology. If it decelerates, the $660 billion capex cycle faces a demand problem. If it accelerates, hardware wins bigger.
What Breaks This
The thesis fails if enterprise software companies show convincing AI revenue in the second half of 2026. If ServiceNow's Pro Plus or Salesforce's Agentforce land genuine revenue acceleration, the software layer captures value too — it was just late.
It also fails if the capex cycle decelerates before hardware revenue peaks. If hyperscalers cut 2027 guidance, the infrastructure trade reverses hard.
The ninety-day test: through July, do hardware semiconductors — TXN, Intel, Broadcom, Marvell — outperform enterprise SaaS — ServiceNow, IBM, Salesforce, Workday — on a total return basis? If yes, the rotation is structural. If the software names recover first, this week was noise.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)