Week of September 14, 2026 · 10 stocks
Oracle's $664 billion backlog steals the week as AI infrastructure runs the table
A record contract backlog at Oracle set off a picks-and-shovels feeding frenzy, dragging the whole compute supply chain higher while a lone bank chart broke the other way.
The feed this week was a monoculture with a few loud exceptions: of roughly 2,000 names and 757,000-plus mentions we track, the conversation kept circling back to who sells the compute. Oracle's historic backlog print was the gravitational center, but the pull reached HPE, Dell, Pure Storage, CEVA and even a mobile-power outfit repurposed for data centers. Around the edges: Lilly held down the defensive GLP-1 corner, DraftKings caught the NFL-kickoff wave, and HDFC Bank supplied the week's only genuine breakdown.
The boring server company grew up this week. HPE's fiscal Q3 print landed non-GAAP EPS of $1.11 against a $0.95 estimate on record revenue, but the number that moved the feed was orders: up 42% year over year, outpacing revenue into a record backlog. The kicker was a gigawatt-scale networking expansion with Oracle, courtesy of the Juniper acquisition, plus AI-systems orders of $2.4B.
Bulls spent the week calling the reversal off the $56-$58 zone a "Stage 2 continuation," stacking Goldman's reiterated Buy at $75 next to a raised FY27 guide. The one bear voice worth quoting shrugged the whole thing off, filing HPE and Dell under "unremarkable hardware suppliers" late in an AI bubble.
The data leans bull: shares run roughly 83% above the 200-day, and the constraint on shipments is memory supply, not demand — the kind of problem a hardware vendor is happy to have.
This was Oracle's week, full stop. Fiscal Q1 revenue of $19.3B and EPS of $1.92 beat, but the headline was remaining performance obligations detonating to a record $664B — up 363% year over year — as cloud-infrastructure revenue jumped 121%. The stock gapped up 7.5% before giving back most of it intraday, which tells you how crowded the trade already is.
The mood was aggressively long: TD Cowen's $240 target, the OpenAI-Stargate backlog, and Jensen Huang's "AGI has arrived" all got recycled as reasons to chase calls into and out of the print. The dissent was small but pointed — Michael Burry's new short and chatter that bond desks price a one-in-six five-year default odds.
Two things support the enthusiasm: Larry Ellison shelved a plan to sell up to $7.5B of stock, and volume ran nearly 3x average. The asterisk is $28B of capex driving free cash flow to negative $5B — the backlog is real, but so is the bill.
Nothing wakes up DraftKings chatter like the NFL, and Week 1 did the job — mentions spiked 17x off baseline. The bull case has a fresh leg this year: prediction markets, now live in roughly 20 states, plus a second courtroom loss for Kalshi in Iowa that took some pressure off the competitive flank. Citizens reiterated Market Outperform at a $37 target.
Posters treated the prediction-market pivot as the real story, framing an exchange-style fee model as a structurally better business than pure sportsbook — management's own "huge opportunity" language got quoted back approvingly.
The stock has been cut nearly in half on the year, and the fundamentals explain the caution: revenue up 17% to $1.65B but operating margin hovering around zero. Worth noting for the tape-readers — an officer took roughly $21M off the table on September 2, the textbook de-risk right into the season's biggest catalyst.
Lilly is the feed's safe harbor, and the GLP-1 leader gave the faithful plenty to stack this week. Retatrutide, the "triple-G" obesity candidate, posted 70.3 pounds of weight loss in Phase 3 — enough to extend the moat well past Zepbound — while the AtaiBeckley deal closed on a 98% shareholder vote and Novo Nordisk's CEO openly conceded share loss to Lilly.
The conversation was confident bordering on smug: patents stretching into the mid-2030s, Lilly's outsized slice of 2026 biopharma buyouts. Even the skeptics flagging a weekly chart breaking below its two-month range mostly stayed long.
The engine backs the swagger — Q revenue up 56% to $19.8B at a 45% operating margin, with the last two EPS prints beating by 30% and 23%. The October 29 report is where the tirzepatide revenue line has to keep pace with the story.
Dell was the week's cleanest breakout, printing fresh all-time highs around $563 ahead of a September 21 S&P 100 inclusion that forces mechanical buying from index funds. The setup behind it: a record $47B revenue quarter, an AI-server backlog swollen to $95B, and Evercore lifting its target to $650. The meme layer — "342% since Trump said go buy a Dell" — supplied the volume of jokes.
Sentiment was near-unanimous long, with the only real friction coming from tape purists muttering that five green days in a row means the asymmetry is gone and a pullback is overdue.
The fundamentals earn the run rather than just riding hype: Q2 revenue of $47B with EPS beating 43%, and 342% on the year still leaving the stock at roughly 30x earnings. Silver Lake trimmed about $15M on September 9 — routine sponsor deleveraging, not a tell that the story cracked.
Here's the week's quiet outlier: a Southeastern road-and-infrastructure builder that jumped 20% on a print almost nobody was watching. Fiscal Q3 revenue hit $999M, up 28%, EPS beat at $1.08, the backlog set a record at $3.36B, and management raised the FY26 guide above consensus. The hook that got fintwit's attention — data-center construction now pegged at 70-75% of new US building inside the company's existing footprint.
What little chatter there is runs strongly bullish, with a soft spot for the governance flex: zero board turnover in 23 years.
And yet the stock still sits at the 11th percentile of its 52-week range even after the pop — the market is openly skeptical that 28% growth holds. Fourth straight beat versus a tape that refuses to believe it is exactly the kind of disconnect that resolves violently, one way or the other, at the November 19 close.
CEVA rode the "physical AI" tag this week, bucketed alongside INDI, OUST and VPG as a way to own the sensor-and-signal layer of the compute boom. The substance underneath: Q2 revenue of $29M up 13%, licensing revenue climbing 21% to a three-year high, and an LG Electronics partnership to push its ultra-wideband IP into cars.
The bull framing leaned on operating leverage finally showing up on a licensing inflection. The pushback was equally specific — one trader flagged the valuation as rich at roughly 6x sales, and one bruised holder confessed to "taking the biggest bath ever" after the print.
The fundamentals say patience: trailing operating margin is still negative on an R&D step-up, and the whole thesis rides consensus EPS climbing from $0.55 this year toward $1.09 by 2028. The tape is cooling into it — shares sit 14% under the 50-day — so the November 9 report carries the burden of proof.
Pure Storage rode the same AI-infrastructure updraft on quieter chatter — the flash-storage vendor is the less-obvious way to play hyperscaler capex, and it keeps posting clean acceleration. Revenue grew 20% to $1.06B at a 70% gross margin, and sell-side models build toward roughly $4.4B in FY27 revenue on the back of price increases and hyperscale adoption.
There was no real crowd voice to quote this week — the name trends on velocity, not debate — but the sell-side split is the familiar one for anything this far into an AI run: "priced for perfection" set against "will keep riding the hype anyway."
The tape reflects the standoff. Shares sit near both the 50- and 200-day lines, digesting a year that's up 26% and a trailing-twelve up 62%. It's an uptrend catching its breath, waiting on the next hyperscale-customer datapoint to break the tie.
Solaris is the AI trade's least glamorous corner — mobile power generation — repointed straight at data centers, and it popped this week on a genuine catalyst. The company acquired heavy-civil and data-center EPC specialist Omega for around $101M and raised its Q3 adjusted-EBITDA guide to $110-130M from $90-105M. (Fair warning: the $SEI cashtag doubles as a Solana-adjacent crypto network, so scroll past the on-chain noise to find the equity.)
The stock-side conversation was bullish, anchored on the power-for-AI pivot and an insider open-market purchase into strength — the tell that tends to matter more than the average grant or sale.
The growth is real: Q1 revenue up 55% to $196M at a 26% operating margin. The catch sits on the balance sheet, which already runs near 2x debt-to-equity — so how Solaris funds the Omega deal is the specific thing the November 2 print has to clarify.
For a week this bullish, HDFC Bank was the conspicuous red mark. HSBC pulled its rating from Buy to Hold and cut the target to $26.10 from $30.80, and the chart chatter turned outright grim — "on the verge of a major breakdown," with $22 flagged as the "last hope" support line as CEO-succession worries linger after the Jagdishan exit.
The bear read dominated, though a contrarian minority argued the ADR near $23 is simply too cheap for a franchise this size. The tape sided with the bears: the stock sits at 10% of its 52-week range and 17% below the 200-day.
The awkward part for the bears is the underlying business — strip out the merger-effect noise and Q3 grew 14% year over year at a 21% operating margin. The franchise keeps compounding while the market discounts integration friction; the October 16 print is where those two stories get reconciled.
Attention is a leading indicator, not a crystal ball — see the live mention velocity, sentiment and per-name analysis on TickerTalks.
Want this live, not weekly?
See what's trending right now →TickerTalks is a research tool, not financial advice. This recap describes what accounts on X are discussing — it is not a recommendation to buy or sell any security. Social attention is not predictive of future price moves; investing involves risk of loss.
