You saw the fireworks: Oracle ripped 27% this week, its strongest performance since June 1999, after expanding a Bloom Energy deal for up to 2.8 GW of fuel cell capacity—including 1.2 GW already deploying—and receiving a $400 million warrant to buy Bloom shares at $113.28. AMD added 13-14% on the week, part of a 42% surge over a 13-day winning streak to a fresh all-time high, its longest run in over 20 years. Microsoft bounced nearly 14%, marking its best week since April 2015, right after its worst quarter since 2008. The iShares Expanded Tech-Software Sector ETF (IGV) surged 14%, on track for its best weekly gain since October 2001. Consensus is convinced: the YTD pain from AI disruption fears is done, power deals prove demand is real, and this rotation into beaten-down software and semis confirms a durable recovery.
Reality is the punchline. This is textbook short-covering relief in oversold names that got hammered earlier in 2026, not evidence that AI infrastructure spending is converting into sustainable revenue acceleration or margin stability. Oracle entered the week down roughly 29% YTD amid AI cloud competition worries—the exact pressures this headline gigawatt deal is supposed to address. The $400 million Bloom warrant delivered paper gains as Bloom shares spiked, but no quantified revenue lift from the power contract appeared in recent filings, and one optics-heavy announcement doesn't close the gap between announced capacity and hyperscaler capex actually hitting order books or utilization metrics.
Zoom out to valuation patterns. Software and semis like these have snapped back hard after 20-30%+ drawdowns before, only to revert when customer spend data failed to materialize. The IGV's 14% week echoes post-2001 rallies that faded without earnings validation. Microsoft added over $350-370 billion in market cap this week alone after shedding nearly 25% in Q1—the steepest drop since the financial crisis—yet its cloud/AI growth still faces the same capex efficiency questions. Hyperscalers are guiding toward $660-690 billion in combined 2026 AI infrastructure spend, with Microsoft tracking $120 billion+, but conversion lags persist: Microsoft has cited an $80 billion unfulfilled Azure backlog tied to power constraints, not demand softness. A bounce from depressed levels doesn't rewrite that math.
Take AMD specifically. Its 42% run over 13 days—the longest streak since 2005—came without fresh MI450 shipment beats or margin expansion data. The move rode broader risk-on sentiment after geopolitical easing around the Strait of Hormuz more than operational inflection. Broader April semis action (Intel +55%, Marvell +41%, strong moves in Broadcom and Micron) follows the same oversold playbook after heavy YTD pressure on valuation and demand concerns. Meanwhile, no major hyperscaler raised Q2 capex guidance directly tied to these laggards in recent reports. The rally prices in broad recovery; the data still shows a sharp split between pure-play AI infrastructure winners and everyone else scrambling for relevance.
Here's the deadpan fact bomb: Oracle secured that $400 million Bloom warrant alongside the power expansion, yet its monster week arrived only after a ~29% YTD decline driven by the very AI cloud pressures it now claims to solve with fuel cells. Power deals generate great headlines and warrant upside—Bloom shares jumped 15-22% on the news—but they haven't translated into disclosed revenue acceleration or margin sustainability that would re-rate multiples in the next 10-Q. Historical reversion in similar software snaps shows these moves often stall when concrete spend figures don't follow the optics.
Connect it to capital allocation and macro sensitivity. Hyperscalers are pouring record capex—projected $690-750 billion across the top players in 2026, up sharply YoY—but efficiency scrutiny is rising. Oracle's own capex plans sit around $50 billion, a big jump, yet software multiples remain vulnerable if utilization metrics don't accelerate. The market crowded the rebound narrative after easing tensions and sentiment rotation. Fundamentals in these names still reflect pressured positioning versus the narrow AI leaders that ground higher through the YTD volatility. This snapback buys breathing room, not absolution for execution gaps.
The rebound stalls within 1-3 months. When Q2 results and next earnings cycles show AI-related revenue growth missing 30% YoY thresholds, margin contraction versus guidance, or no fresh capex raises validating the hype, valuation compression will hit hard. The gap between headline gigawatts and actual conversion reasserts itself every time.