Week of August 17, 2026 · 11 stocks
The AI build-out goes physical: turbines, storage IP and robot silicon crowd the feed while Cerebras stumbles
The week's loudest names weren't chatbots but the picks-and-shovels powering them — even as a Cerebras hardware miss and a run of biotech surprises split the crowd's attention.
Nearly 1.36 million stock mentions crossed X this week across a universe of 1,864 names, and the through-line wasn't the models — it was the machinery underneath them. Flash-storage IP, steam turbines, inference chips and robot silicon dominated the loudest threads, with a cluster of biotech readouts and small-cap defense plays fighting for the back half of the feed. We stripped out the cashtag imposters first — the $QNT chatter was about the Quant/Overledger crypto token and $TRX was all Tron, not the listed companies — and kept only the names where the conversation is genuinely about the equity.
Pure Storage — now rebranded Everpure — landed the week's cleanest hardware story: a design win with a top-four hyperscaler, with Meta the name most often floated, for its DirectFlash technology. The stock jumped nearly 24% and drew a Piper Sandler upgrade to Overweight. The twist is the model — this is licensed IP, not boxes, with Pure earning fees on flash deployed inside a hyperscaler's own storage stack. That single customer already chipped in roughly $30M last quarter at 90%-plus gross margins.
The volume of chatter was thin next to the megacaps — just 19 posts on the week — but the ones talking framed it as the quiet AI-infrastructure name finally getting paid for a decade of flash engineering, with sell-side notes penciling $500M-plus in hyperscaler revenue by next year.
The fundamentals back the enthusiasm without stretching: fiscal Q4 revenue grew 20% to $1.06B at a 70% gross margin, and the shares sit up 62% over the trailing year, digesting the run near their 50-day line. The line the sell-side keeps drawing is 'priced for perfection' — impressive execution with little room to slip.
Liquidia's YUTREPIA ramp is doing the thing biotech bulls dream about. Q2 net product sales hit $170.4M, up 31% sequentially, with $74.7M of net income and $96.3M of adjusted EBITDA — a fourth straight quarter of rising profitability — and the inhaled-treprostinil launch now pushes near 30% of the prostacyclin market by revenue, with management reiterating a $1B-plus target for 2027.
The feed is close to one-sided. Posters treat every earnings-day dip as a chance to add, cite a launch outpacing United Therapeutics' Tyvaso, and point to lifted price targets from Needham and BTIG. Their near-term fixation is the pending '327 patent decision — described as landing 'any day now' with a 'buckle up' tone and barely a bear voice in the window.
The numbers support the mood, with the stock up 156% year-to-date on genuine commercial traction rather than hope. The asterisk the crowd is under-weighting: director Stephen Bloch sold about $24M across four days in late July, and the patent ruling is a true binary — a favorable outcome confirms the moat, an adverse one reopens the door to competition.
Reddit was the week's biggest attention magnet — 2,166 posts, chatter up 88% day-over-day — after S&P Dow Jones confirmed it joins the S&P 500 before the open on August 18, replacing AvalonBay Communities as Equity Residential absorbs the apartment REIT. It's an off-cycle add, and it makes Reddit only the second pure-play social platform in the index after Meta. The stock jumped double digits on the news.
The bull case in the threads has migrated from ads to AI royalties. Posters keep flagging Reddit as the single most-cited domain in ChatGPT, Perplexity and Gemini answers, arguing its $130M-plus data-licensing deals with Google and OpenAI are a revenue line ad-only peers can't copy — and framing the index inclusion as a mechanical, non-discretionary bid from passive funds.
The Q2 numbers give them something concrete: net income roughly tripled to $252M year-over-year on minimal capex, with 61% revenue growth and margins social platforms rarely post. The offset even the bulls concede — CEO Steve Huffman sold $2.88M of stock in early August, just ahead of the inclusion headline, and logged-in user growth is the metric bears keep circling into the October print.
Old-line Babcock & Wilcox — boilers and environmental controls since the 19th century — just re-rated on AI. Q2 revenue surged 130% to $319.7M, net income swung to a $14.3M profit from a year-ago loss, and backlog exploded 533% to $2.6B. Management raised full-year EBITDA guidance to $80-105M and authorized a $50M buyback, then on August 11 signed Siemens Energy to build 20 steam turbines — a full gigawatt — for its FastPower data-center program.
Sentiment in the threads is close to euphoric. The framing is 'sleep three years': a legacy industrial suddenly holding a 4-to-6 GW pipeline of AI-data-center power projects, with the stock popping 41% after hours on the print. The one measured note came from a poster flagging EBITDA-margin compression as the Base Electron project ramps.
The fundamentals underwrite the excitement rather than contradict it — a genuine profit turn, a $14B-plus opportunity pipeline, and a balance-sheet cleanup that included redeeming the remaining ~$61M of preferred stock. The risk the tape is watching is execution: signed-contract slippage, or thin margins on that incremental data-center revenue.
SIFCO is the week's information-arbitrage puzzle. The tiny aerospace-forgings maker — engine and structural parts for Boeing, Airbus and defense OEMs — posted sales up 39% to $26.4M and adjusted EPS of $0.51 against two cents a year ago, but a GAAP headline of -$0.01 and a clumsy press release triggered a 35% pullback. Book-to-bill ran 1.3x, the kind of number that says orders are still outrunning shipments.
The bull thread reads like a mispricing hunt: posters argue the sell-off was a mechanical reaction to a badly-worded release, not the business, and that the adjusted print and rising backlog tell the real story. Bears counter with the gross-profit line, but among the small pool of voices here the arbitrage framing dominates.
Under the hood, the operational inflection looks real — 20% gross margins, a defense-and-aerospace cycle tailwind, and a stock up 393% year-to-date — trading near 10x EV/EBITDA with a 4.5% free-cash-flow yield. The figure to watch is book-to-bill: hold above 1.0 and the coil stays intact; slip below and the pullback gets a second leg.
indie Semiconductor turned its Q2 call into a robotics pitch. Revenue grew 24% to $64M with the operating loss narrowing to $8.9M from $14.5M, and Q3 is guided up 30% — but the crowd fixated on the iND880 vision chip. CEO Donald McClymont pitched its DRAMless design as a fix for customers who 'are unable to source memory at all,' at roughly a $10 average selling price.
The bulls have latched onto the humanoid angle. indie flagged design wins with Unitree and Agibot — each of which shipped over 5,000 robots last year — and McClymont's line that you can think of 'a humanoid robot as a car with legs' became the thread's rallying cry. Some are penciling $100-plus of silicon content per robot; the technical crowd is marking fib levels for the next leg.
The valuation gives the story room — about 3.9x sales on a roughly $854M cap, with consensus revenue climbing toward $360M by 2027 — while the Volvo Tier 1 radar win anchors the base business. The catch the data flags: that humanoid revenue is still a narrative, and a production slip at Figure or Unitree pushes it out to 2028.
HIVE Digital dropped its fiscal Q1 2027 results after the close on August 14, with the earnings call falling on the 17th — and the run-in chatter was all about the pivot from Bitcoin mining to AI-GPU hosting. The bull pitch: roughly $200M in ARR from AI data centers layered on 27 EH/s of mining capacity, a 320MW AI hub planned for the Greater Toronto Area, and all of it on renewable power.
The tone heading into the print was unabashedly promotional — 'AI plus green power equals the perfect setup,' with one poster calling it 'one of the biggest wealth-creation opportunities of our generation.' The lone caveat surfaced was housekeeping: a tiny crypto 'Hive' token shares the cashtag, so some of the $HIVE noise isn't about the miner at all.
The data keeps the story honest. Prior-quarter revenue grew 130% to $71.8M, but gross margin sat deeply negative as the company spends hard to stand up GPU capacity — a burn that needs AI revenue to land fast, or more equity to fund it. CFO Darcy Daubaras' $319K sale in mid-July is the small pre-earnings tell the bulls have to explain.
Cerebras handed the AI-chip crowd a genuinely split print. Q2 revenue of $180M missed the ~$194M Street number and hardware sales actually fell 23% to $54.1M — the line a wafer-scale story can't afford to shrink — sending the stock down almost 14%. Yet the company raised full-year guidance to $880-890M, beat on the adjusted loss, and pointed to a $20B multi-year OpenAI deal for 750MW of inference compute.
The feed stayed bullish even as the tape sold off. Posters leaned on 281% cloud-revenue growth and a $25.4B backlog, the OpenAI 'Ultrafast' mode clocking 750 tokens per second, and the AMD partnership — treating the drop as a shakeout. The bearish contribution was mostly a single 'timberrrrr' post on the red candle.
The data is where the caution lives. Against an 87x price-to-sales multiple, a 23% hardware decline is a real crack, and the load-bearing red flag is insider: the COO trimmed roughly 90% of their direct stake before the print — hard to wave away as tax planning. Cloud strength is real; the hardware ramp is what November has to fix.
Merlin, an aircraft-autonomy micro-cap, is the week's contrarian argument. The August 13 Q2 print was a clear miss — revenue of $2.19M against a $3.78M estimate and a 409% EPS surprise to the downside — and the stock dropped 15% on the day, now off 64% over the year. On the numbers alone it looks broken.
The remaining bulls are arguing catalysts over the quarter. They point to a cleared SOI 3 with New Zealand's civil-aviation authority, coordinated with the FAA, a closed FAA issue paper on AI/ML natural-language processing, and an autonomous-landing demo at Oshkosh in July — 'real progress while nobody was watching' — with visible frustration that the crowd rotated to ONDS and KTOS instead.
The data urges patience. Analysts model revenue building from ~$17M in FY26 toward $65M by FY28, with EPS negative for years — a long runway on a $429M market cap where regulatory progress may already be priced. What resolves it either way is a specific defense contract or FAA milestone; absent one, the base case is more volatility.
dLocal, the cross-border payments rail for emerging markets, drew a rising-tier crowd after a revenue-strong Q2. Sales grew 56% to $399.7M and total payment volume jumped 92% year-over-year, prompting management to lift its 2026 TPV guide to up 60-70% and buy back 6.9M shares at an average $12.47. EPS of $0.18 missed by a single penny — small enough to read as rounding against the volume story.
The threads frame it as the cleanest way to own emerging-market payment growth — Latin America first, then Africa and Asia — collecting local-currency payments for the likes of Amazon, Netflix and Microsoft. Bulls acknowledge the take-rate and margin pressure but treat it as an accepted cost of chasing that volume; the tone leans bullish despite an initial post-earnings dip.
The economics hold up: a mid-30s gross margin, a 34% return on invested capital, and a double-digit free-cash-flow yield, all at roughly 2.7x sales — a discount to slower-growing payment peers. The yellow flag: founder-CEO Sebastián Kanovich sold about $1.13M on August 5, largely tied to an option conversion, a reminder that risk/reward at $14-15 is more balanced than it was at the $8 lows.
AbCellera spent years being called a 'platform' with no drug of its own — and then on August 10 the platform produced one. Top-line Phase 2 data for ABCL635, a monthly non-hormonal injection for menopausal hot flashes, showed an 83% reduction in flash frequency versus 33% on placebo and a 58% cut in severity versus 12%, all from a single dose across 92 women. The stock leapt roughly 50% and cleared $10 for the first time in years.
The feed read as an inflection-point party. Posters seized on the follow-up $200M raise as management signaling it will take the drug to market itself rather than license it away, and on Truist's $1.1-2.4B peak-sales model. The strongest tell they cited was insider conviction, with almost no counter-narrative beyond a few 'wait, is this actually bullish?' jokes.
The data backs the excitement about as far as Phase 2 data can. Two directors bought in the open market this week — chairman Michael Hayden's $481K and CEO Andrew Booth's $384K, both near the raise price — the strongest insider signal you get right after a positive readout. What could stall it: a stricter FDA bar for the Phase 3 endpoint, or a competitor posting better efficacy at the same stage.
The live mention counts, crowd sentiment and per-name verdicts are all on TickerTalks — the feed refreshes daily, and next week's leaders are already forming.
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.
