Signals

See the world early.

Filter

Signal categories

Latest Signal publications and updates

Technology

Updated

Taiwan Has Approved ASE's Billion-Dollar Packaging Expansion. Equipment Suppliers Could Be Paid First.

Taiwan approved ASE’s NT$123.7 billion packaging investment plan; equipment suppliers could receive orders before new production generates revenue.

Sources
Outcome updated

Outcome updated ·

Unresolved · Newly disclosed evidence

Ainos filed the signed purchase agreement with the SEC on 17 November. It records the same US$9 million order for 5,000 units and a US$4.5 million deposit due within 60 days of signing. The remaining amount can be invoiced after all…

Read the update and evidence

First published

Read full signal

AI powered research

Technology

Updated

Before renewing an AI supplier, ask what it keeps

Before renewing an AI supplier, ask what information it keeps after the job is done, who owns the retention decision and what evidence the supplier can share.

Full explanation

UpGuard announced a dedicated research arm, not a new rule or a guarantee that a particular product is unsafe. A buyer can start with its contract and retention schedule. The announcement does not establish that customers have paid for new reviews. A team can stop using an AI service while its supplier still holds material under the contract. The practical issue is not whether a privacy policy exists. It is whether someone can explain what categories are retained, why, for how long, which exceptions apply and who will answer when the relationship ends.

Sources
Outcome updated

Outcome updated ·

Unresolved · Newly disclosed evidence

UpGuard’s launch announcement did report a specific data exposure from its own research. It did not establish a problem at the AI supplier a buyer is deciding whether to renew.

Read the update and evidence

First published

Read full signal

AI powered research

Technology

When the satellite list goes quiet, check where you are looking.

If your farm adviser sends you a map update every week, ask where the list of new satellite images comes from.

Full explanation

Copernicus is changing how it delivers that list. A team using the old direct storage bucket needs to change its setup by 20 January; a team using the supported catalogue API does not. Before the 20 January switch, compare product IDs for the same Sentinel platform and inventory dates. After it, use the supported catalogue API. A map reader can mistake a list that stopped updating for a week with no new satellite coverage. A change to how the list is delivered can explain the silence instead.

First published

Read full signal

AI powered research

Technology

The press counts impressions. How many shirts will your buyer accept by Friday?

An apparel seller promising 500 printed shirts needs to know how many will pass inspection by Friday.

Full explanation

Kornit says longer jobs supplied 84% of its added print impressions through September. That does not tell the seller how many shirts arrived on time. Ask two printers to quote the same artwork, fabric, quantity and deadline. Compare the cost per accepted shirt, including reprints and delivery, before promising a turnaround. A shop can print a shirt and still miss the buyer’s deadline if the colour or size is wrong. A press count measures work started at a machine, not the goods your customer accepts.

First published

Read full signal

AI powered research

Technology

Who handles the first refund after your app builder leaves?

Before paying someone to launch an AI-built app, ask who will handle the first refund if the builder leaves.

Full explanation

Sweden’s CloudBackend announced €6.6 million in funding and plans to start admitting selected creators on 24 November. Neither fact shows what happens when a customer needs help. Ask the builder to walk through one fictional order with you, name the support owner and show what records you can take away. A shop can pay for a booking app, yet still depend on its builder to find an order when a customer cancels. A demo shows the booking screen. It does not say who is allowed to answer a customer, approve a refund or hand the shop its records if the builder stops working for it.

First published

Read full signal

AI powered research

Technology

Expel is testing a view of why alerts were closed. Were the decisions right?

On 18 November, Expel announced changes to how its Workbench security service reviews alerts.

Full explanation

It says several tools are live, but the Cases view that lets customers inspect the reasoning is still in early access. A buyer should compare decisions on approved past records and count mistakes, reviewer time and costs. A clearer explanation can help people check a decision. It does not show that the decision was right or cheaper. A faster or better-explained review may still miss an important alert. A buyer needs to know whether the result was correct and whether time savings remain after human review and correction.

First published

Read full signal

AI powered research

Technology

Before a data centre buys power to bridge a grid delay, price the wait

A data centre waiting for grid power now has another proposal to compare: a package pairing gas generators with batteries.

Full explanation

Trina Storage formally unveiled it in Las Vegas after a September preview. Before paying for a review, ask for the likely grid connection date and the cost of waiting. A named customer, an operating permit and measured fuel savings are three different pieces of evidence. The launch supplies none of them. If a utility cannot supply power when promised, a developer faces two bills: the cost of waiting and the cost of bringing power to the site temporarily. If the grid arrives earlier, the developer has fewer months over which to recover the temporary system’s costs.

First published

Read full signal

AI powered research

Technology

A cheaper AI answer needs a stopwatch, not a brochure

ADATA’s TRUSTA has announced a toolkit upgrade that moves some of an AI service’s working memory off its graphics chip.

Full explanation

For companies that answer similar questions repeatedly, reusing earlier work may be cheaper than doing it again. Test that on a workload you have permission to use before buying storage. The vendor’s speed figures describe one part of a particular test, not a customer’s bill or an order. A help desk might send the same long instruction sheet with hundreds of different questions. Reusing work on the shared beginning could shorten the wait for the first word. If the questions share almost nothing, loading old work from storage can take longer than starting again.

First published

Read full signal

AI powered research

Technology

Updated

Forty Satellites Were Launched. None Reached Orbit. The Next Product Is Proof of Reliability.

Launch capacity is one bottleneck in space broadband.

Full explanation

A failure can move demand into independent testing, revised hardware inspection, launch-pad scheduling, replacement production, customer-service delay and risk insurance, but none yields a new paid order automatically. Network speed depends on operational satellites and ground services, not units manufactured.

Outcome updated

Outcome updated ·

Unresolved · Milestone

In an 18 November 11:00 EST update, ULA said its initial review of integrated flight data indicated that inadequate fuel supply from the vehicle to the BE-4 engines, rather than engine performance, accounted for the low booster thrust.

Read the update and evidence

First published

Read full signal

AI powered research

Technology

Updated

Europe approved chip projects. Who will prove they work?

More than 20 cumulative state-aid decisions are associated with over €33 billion in planned public and private investment, the European Commission said on 17 November.

Full explanation

That figure is not aid paid out, completed factories or testing revenue. The question for European technicians, materials testers and component suppliers is which buyers will outsource qualification work.

Outcome updated

Outcome updated ·

Unresolved · Newly disclosed evidence

We subsequently located the European Commission's 16 November daily-news announcement. It confirms direct-grant approvals of €43.5 million for Lidrotec's laser wafer-dicing equipment facility and €14.1 million for Dockweiler's…

Read the update and evidence

First published

Read full signal

AI powered research

Technology

Updated

Airia says it can label AI tasks. Does that make a finished job cheaper?

Airia says its gateway can classify AI requests by task, giving a team a way to ask what work its bill represents.

Full explanation

A label is not proof of useful work or lower costs. Compare the full cost of a finished, accepted task, including corrections, human review and the gateway fee. The announced classifier labels individual prompts in gateway traffic. Whole-conversation classification and automatic routing to cheaper models were described as future enhancements.

Outcome updated

Outcome updated ·

Unresolved · Newly disclosed evidence

Airia's 27 October release notes already described early-access rules that can route gateway requests by task category. Its 17 November announcement described task classification for gateway customers and future integration for routing to…

Read the update and evidence

First published

Read full signal

AI powered research

Technology

Updated

Nvidia's Investigation Could Create Work for Independent Security Specialists.

Nvidia's board announced an independent investigation while saying it had no reason to believe secrets were compromised.

Sources
Outcome updated

Outcome updated ·

Unresolved · Milestone

On 17 November NVIDIA announced a standing board committee for national-security and export-control oversight and said the committee intends to retain independent external experts in law, national security and trade compliance.

Read the update and evidence

First published

Read full signal

AI powered research

Technology

Updated

Email Coming Back Does Not Mean Every Customer Request Has Been Recovered

A working inbox is not the same as a recovered business day.

Full explanation

Check requests taken elsewhere against delayed messages before work is repeated or missed. A backup route can fail if its login depends on the same inbox, and temporary requests may need reconciliation.

Sources
Outcome updated

Outcome updated ·

Unresolved · Newly disclosed evidence

In a notice dated 14 November, DAFF said affected clients should again be able to send and receive email and that no action was required. Its earlier notice said IMS biosecurity directions and SAC auto reports would queue for resending; it…

Read the update and evidence

First published

Read full signal

AI powered research

Technology

An AI assistant says the invoice was sent. What would actually prove it?

Microsoft says it is working with more than 40 organisations to make AI actions easier to check. For a buyer, the useful test is whether a second person can establish what happened without trusting the assistant’s answer.

First published

Read full signal

AI powered research

Technology

More people are building automations. Is anyone saving time?

IFTTT says more people are building small automations since its October launch.

Full explanation

Before a business pays for one, try a reversible job and count errors, corrections and credits as well as minutes saved. The platform’s growth figures do not tell you whether a client will pay.

First published

Read full signal

AI powered research

Technology

Fortra’s AI can help explain a file transfer. Who approves the change?

Fortra announced selected-customer testing of AI-assisted workflow and log analysis for GoAnywhere MFT and Globalscape EFT.

Full explanation

A readable explanation is not authority to alter a payroll or shipment route. The company expects wider availability in early 2027; performance, savings and paid outside work have not been established.

First published

Read full signal

AI powered research

Technology

Updated

Germany’s E-Prescription Tender Makes the Supplier Handover Worth Testing.

A healthcare system must keep working when suppliers change.

Full explanation

Germany's procurement makes a handover rehearsal worth investigating, but no separate paid testing job has been confirmed. An experienced testing firm can ask an eligible bidder for a defined task. Existing tools may already cover it. Use synthetic records, agree an acceptance test and secure a funded scope before spending. The opportunity is conditional, not a direct contract for beginners.

Outcome updated

Outcome updated ·

Unresolved · Newly disclosed evidence

Rechecking the buyer's existing procedure clarifies a distinction omitted from the original summary: applicants may submit participation requests for one or both lots, declaring a preference if applying for both. They may be invited to…

Read the update and evidence

First published

Read full signal

AI powered research

Technology

Updated

New Rental Records Test the Business of Checking Who Uses AI Chips

New rental-access reporting sharpens an existing OTR compliance thesis. Equipment, customer and access records may need to be connected under applicable rules; no blanket rental ban is established.

Outcome updated

Outcome updated ·

Unresolved · Milestone

On 15 November 2026, Megaspeed told CNBC that it had retained external legal counsel to investigate media reports concerning its business, supply chain and global operations. This later company statement describes a broader internal…

Read the update and evidence

First published

Read full signal

AI powered research

Technology

AI Cloud Customers Are Paying Early. The Hidden Sale May Be Getting Their Data Ready

Emerging data:

Full explanation

Baidu released its third-quarter results on 16 November 2026. Its AI-cloud infrastructure business reported RMB8.2 billion in revenue, up 94% from a year earlier. Revenue from its GPU cloud category rose 191% year on year, while the company said more customers were making prepayments and choosing longer contracts. The amounts of those advance payments and the exact contract lengths were not separately disclosed. This is stronger than a press release promising future demand. It is still the company's own unaudited operational reporting, not an independently sampled customer-invoice book. At first, it looks like a story about computing chips in China. It may actually be a story about the work a business must finish before a prepaid server earns its keep. A hospital, factory, retailer or insurer does not get value simply by renting AI computing power. It needs approved data, protected access, software connections, reliable testing and a process for handling mistakes. These are narrower paid jobs than building a model from scratch. The contradiction makes the opportunity interesting. While AI cloud grew fast, Baidu's online marketing revenue fell 22% to RMB11.9 billion and its overall quarterly revenue slipped to about RMB30.8 billion. Its AI-powered businesses grew but did not erase the older business's weakness. This means cloud growth is real as reported, but not a blanket instruction to buy an AI company. OTR's earlier lesson says AI demand is not the same as profitable implementation. Here, an extra clue is that Baidu says customers are accepting longer terms and prepaying. That reduces some collection risk but may bring delivery, capacity and service obligations into the future. Our proposed connection: Data preparation and access-control packages could sell to customers before cloud deployment; workflow integration and acceptance testing could earn a fee when a customer signs off a working process; and monthly monitoring, audit and repair services could earn repeat income after deployment. The best small-firm entry is not to speculate on GPUs. It is to show one firm how to move one expensive business task from a broken manual process into a measured, safe workflow. That requires lawful customer permission and strong security. The public investment exposure is Baidu (NASDAQ:BIDU, HKEX:9888). As the report was researched, a verifiable US close was US$78.14 at 4:00 pm New York time on 13 November, with 16 November premarket around US$77.7 at a quoted early-morning snapshot. This is not a verified 16 November closing price. The issuer reported about RMB276.6bn of cash and investments, a 5% GAAP net margin, 15% adjusted EBITDA margin, RMB2.8bn operating cash flow, but RMB4.6bn negative free cash flow for the latest quarter based on its calculations (data label/Q3 scope must be checked in tables before use in a trade). Strong revenue growth can coexist with heavy capital spending. What the six questions reveal. (1) The fresh change is reported prepaid demand and longer customer commitments, not the existence of AI. (2) GPU capacity, customer data, security, integration and old ad weakness connect. (3) Potential money flows to authorised cloud integrators and data/security contractors; no named partner order is verified here. (4) If a customer has prepaid cloud, waste starts while the system remains unused; test the first job now. (5) The company's revenue is evidence of paid cloud use, but no evidence proves the implementation service is underpriced in the market. (6) The idea fails if customers already operate their systems well, do not fund services or their cloud commitments are canceled. Second to fifth consequences. Second: more prepaid GPU capacity could push customers to speed up internal integrations. Third: more active integrations could require steady security, testing and monitoring rather than one-time projects. Fourth: a set of working workflows could raise repeat monthly demand for the cloud provider. Fifth: large providers may bundle those services free, squeezing independent consultants; if AI does not save staff time or grow sales, customers may reduce future renewal budgets. Twenty intelligence checks: what the evidence really establishes 01. Money movement. Baidu Q3 cloud infrastructure revenue RMB8.2bn, 94% growth; issuer reports more prepayments. Prepayment quantity, project-level price and cash collection split not separately disclosed. 02. Early buying. Longer contracts and prepaid use are issuer-reported buyer behaviour, stronger than downloads or searches; exact signed customer counts NOT VERIFIED. 03. Bottlenecks. GPU availability, power, data rights, security reviews and integration queues. GPU cloud grew 43% sequentially according to issuer; customer utilisation per paid hour unknown. 04. Hidden beneficiaries. Qualified data migration specialists, audit-log vendors, security trainers, system integrators and managed-service teams rather than only chip suppliers. 05. Financial exposure. Baidu BIDU last independently confirmed close US$78.14 (13 Nov) and Nov16 early premarket roughly US$77.7. Q3 consolidated margin 5% GAAP, adjusted EBITDA 15%, cash/investments RMB276.6bn; AI-cloud segment's stand-alone operating margin, large-customer concentration and future cost of delivery are not all public. 06. Unit economics. ILLUSTRATIVE: an authorised integrator receives RMB40,000 for one signed, accepted deployment; spends RMB18,000 labour, RMB4,000 testing, RMB3,000 tools and RMB5,000 support/warranty reserve. Leaves RMB10,000 before overhead and tax. No buyer paid this quoted price in our evidence. 07. Ownership/institutions. Baidu reported US$373m of share repurchases since Q1 2026. That does not establish director buying or investor faith in the narrow prepayment mechanism. 08. Hiring. No dated named vacancy tied to the 16 Nov disclosure established. Specialist AI deployment jobs are a future confirmation check. 09. Government procurement. No government budget award tied to Baidu's 94% figure verified; this is mainly private commercial revenue. 10. Foreign-language lane. Chinese issuer reporting and the original RMB disclosure are the right base; no distinct Chinese supplier purchase-order original independently verified for this client-use angle. 11. Payment versus attention. Issuer says prepayments and longer contracts; this is stronger than search interest. Individual invoices, renewals and customer savings are unobserved. 12. Market contradiction. Cloud revenue +94%, ads -22%, total revenue down; capex and cash generation may overwhelm the headline. 13. Missing expected activity. No public breakdown of prepaid balance attributable to AI cloud, no integration utilisation figures, no verified small-firm service wins. 14. Commercialisation. Paid cloud → secure data loaded → workflow accepted → logged output quality → production use → contract renewal. Revenue at step one is not proof of end-user ROI. 15. Competition. Alibaba Cloud, Huawei Cloud, Tencent Cloud and in-house teams can supply similar computing and implementation; prices may fall faster than workload rises. Evidence and timestamps. The 16 Nov trigger is the issuer's new earnings release, which reports a July-September period. Third-quarter transactions themselves are not 16 November transactions. Press release and slide deck share an issuer family; Reuters gives separate reporting, not independent audit of customer ledgers. What readers can do now: If you are a small software provider, do not start by purchasing cloud credits and hoping clients appear. Find a business already paying for AI computing or automation and ask for one task that currently fails. With its permission, map who enters data, who approves it, what the existing cost is and how an error is corrected. Propose a paid fixed-scope workflow integration with an agreed acceptance test before any build. Keep data protected and define who can stop the system. The first 30-day test should measure baseline minutes per case, new minutes per case, wrong answers needing human repair, customer sign-off and cash collected. Forecast: if prepayments are recurring and cloud customers struggle to deploy, paid integration orders should appear before every new AI model release. Stop if the customer cannot supply lawful data or a budget holder will not sign for a measurable improvement. 16. Geographical spillover. Chinese cloud build-out can help domestic services; privacy, data-location and export restrictions prevent assuming identical workflows transfer abroad. 17. Historical discipline. OTR's lessons on AI adoption and proof warn against confusing higher demand with successful customer outcomes; older reported percentages are context, not today's payment receipts. 18. Awareness and valuation. Revenue release, wire coverage and share quotes show this is public information. The precise overlooked earnings value of integration is NOT PRICED OR VERIFIED. 19. Disproof. Falling prepayment renewal, severe data breaches, margin loss from computing costs, cloud outages, cancelled deals, free bundled integration and wider advertising decline. 20. Immediate trigger. Find one existing customer with a signed cloud service agreement before proposing a pilot. Set a 30-day acceptance test with success threshold (minutes saved, error rate, uptime and data permission). Cap pilot spending to a written fee; terminate if no signed acceptance or no measurable savings. Editorial clarification · 17 November 2026 Recognised revenue does not establish individual customer cash receipts. The reported prepayment amounts remain undisclosed. The cash-flow and margin figures above describe Baidu’s consolidated quarter, not the AI-cloud segment or an integrator’s return. The possible service pathway and illustrative economics are not observed customer outcomes. Sources 01. FRESH, 16 Nov 2026, original issuer Beijing, Q3 revenue, AI cloud and customer prepayment report: https://ir.baidu.com/news-releases/news-release-details/baidu-announces-third-quarter-2026-results 02. FRESH, 16 Nov 2026, issuer Q3 presentation, GPU and longer-term contract figures; same issuer family: https://ir.baidu.com/static-files/23cafbd4-dbc5-4c25-8e70-f00dbae0d7e2 03. FRESH, 16 Nov 2026, Reuters independent report, opposing advertisement and cloud changes: https://www.reuters.com/business/media-telecom/baidu-third-quarter-revenue-slips-1-meets-market-estimates-2026-11-16/ 04. FRESH, 16 Nov 2026, original investor PDF financial statements, cash and margins, not a separate source family: https://ir.baidu.com/node/14996/pdf 05. MARKET SNAPSHOT, Nov 13 close and Nov 16 premarket ET, Nasdaq BIDU provider; regular session live price not confirmed at review cutoff: https://stockanalysis.com/stocks/bidu/ 06. FRESH, 16 Nov 2026, Baidu investor second official host, same earnings disclosure, not independent corroboration: https://baidu.gcs-web.com/news-releases/news-release-details/baidu-announces-third-quarter-2026-results

First published

Read full signal

AI powered research

Technology

Updated

A Nexperia Court Update Does Not Approve a Factory’s Next Chip Batch.

A reported court update does not establish that any particular replacement chip is legally or technically approved.

Full explanation

A buyer must check each proposed part and lot against its own requirements. An authorised specialist could investigate a defined documentary gap, but no buyer, affected batch or paid review is established.

Update published

Update published ·

Plan changed

Wingtech disclosed its intention to appeal the Enterprise Chamber's 11 November interim measures in a Sunday stock-exchange filing, according to Reuters reporting on 15 November 2026. The event date records this reported disclosure of…

Read the update and evidence

First published

Read full signal

AI powered research