Apple India's 66% profit jump, MeitY's Meta advisory, and AI on COBOL
Apple's India business got far more profitable, the government wants Meta to rethink its feeds, and OpenAI takes aim at the old code Indian IT has billed on for years. Plus a YouTube reset and a scam-check prompt worth saving.
- Long Press · Daily Five21 Oct 202601 / 05
commentaryIndiaApple India's profit jumped 66%, and it isn't discounting
Apple India made Rs 5,311 crore in net profit in FY26, up 66% from Rs 3,196 crore, on operating revenue of about Rs 1.04 lakh crore, Inc42 reported on October 20. Profit grew roughly twice as fast as sales, so Apple isn't buying its way into India with discounts; it's keeping more of every rupee here. That's the year India quietly stopped being Apple's interesting side market and became a properly profitable one. The line I'd watch is the licence fee Apple India pays within the group, which went from Rs 4,258 crore to Rs 10,374 crore, because money that starts as Indian revenue doesn't all stay as Indian profit.
The questionHow much profit did Apple India make in FY26?
In one lineApple India reported a net profit of Rs 5,311 crore for FY26, the year ended March 31, 2026, up 66% from Rs 3,196 crore, according to figures Inc42 reported on October 20, 2026. Operating revenue crossed Rs 1.04 lakh crore, up about 31%.
The problem
For years India was the market Apple praised on earnings calls but barely made money in: a huge population, a small share, and buyers who mostly picked Android on price.
What changed
FY26 figures reported by Inc42 show net profit up 66% to Rs 5,311 crore and operating revenue up 31.4% to about Rs 1.04 lakh crore, from Rs 79,060 crore. Total expenses rose 28%, slower than revenue, and iPhone exports from India crossed Rs 2 lakh crore.
The catch
The licence fee paid within the Apple group rose about 2.4 times to Rs 10,374 crore, so a growing slice of what India earns goes back to the parent before it shows up as local profit. And nothing here hints at lower prices in the shop.
What it means for you
Don't hold out for a desperate Apple price cut in India; on these numbers it simply isn't under that kind of pressure. Festive bank offers and last year's model are still the real ways to pay less for an iPhone.
Questions people ask
Why did Apple India's profit grow faster than its revenue in FY26?
Operating revenue rose about 31% to roughly Rs 1.04 lakh crore while total expenses grew 28%, so Apple India kept more of each rupee it sold. Its biggest cost, buying the goods it resells, rose only 21.6%. The one cost that jumped was the licence fee paid within the Apple group, up about 2.4 times.
Does Apple India's profit jump mean iPhones will get cheaper in India?
Nothing in the FY26 numbers points that way. They show Apple India earning more per rupee of sales, not cutting prices to grow. Local assembly helps Apple's costs and exports, which crossed Rs 2 lakh crore in FY26 according to Inc42, but the filings say nothing about lowering retail prices for Indian buyers.
How much does AppleCare earn in India?
AppleCare and support services brought in Rs 4,292.8 crore in FY26, up 7.6% year on year, according to Inc42's report on Apple India's numbers. That's small next to the Rs 99,068 crore Apple India made from selling goods, so the India business is still overwhelmingly about hardware rather than services.
- Long Press · Daily Five21 Oct 202602 / 05
commentaryIndiaMeitY wants Meta to change what your feed shows
MeitY has told Meta to review the recommendation systems behind Facebook and Instagram so users aren't pushed a single or limited set of narratives, in an advisory dated October 10 and reported by The Pioneer. Context labels on recycled or doctored event footage are a good idea, and frankly overdue. The rest worries me. Limited narratives has no definition, Meta reports back to the ministry rather than to the people using the app, and nobody outside that conversation will be able to tell whether a feed was fixed or just tuned to suit whoever asked. If feeds are going to change on government advice, both sides should publish what changed.
The questionWhat did MeitY ask Meta to change on Facebook and Instagram?
In one lineIn an advisory dated October 10, 2026, MeitY asked Meta to review Facebook and Instagram's recommendation systems so users aren't shown a single or limited set of narratives, to add context to unclear event visuals, and to escalate content that could cause disorder. Meta hasn't publicly responded.
The problem
Recommendation systems decide what crores of Indians see first, and old or edited protest footage travels fastest when it's posted as if it happened today.
What changed
The MeitY advisory, dated October 10 and reported by The Pioneer on October 18, asks Meta to avoid disproportionate exposure to a limited set of narratives, label visuals whose date or place is unclear, quickly escalate content that could lead to violence or disorder, and report the action taken. It cites the IT Rules, 2021.
The catch
It arrived during protests in Delhi over electoral roll revisions, and the claim that Meta's algorithms favoured certain content comes from unnamed ministry sources rather than any published finding. Meta hasn't responded publicly, The Hans India notes.
What it means for you
If a protest or emergency clip lands in your feed this week, check when and where it was shot before you forward it. That part of the advisory is good advice for all of us, whatever Meta ends up doing with the rest.
Questions people ask
Is MeitY's advisory to Meta legally binding?
An advisory isn't a new law. It leans on the IT Rules, 2021, which set the cooperation and takedown timelines platforms must meet, and it asks Meta to act with immediate effect and report back on what it did. The advisory doesn't define how anyone will measure whether Meta has complied.
Why did MeitY send the advisory to Meta in October 2026?
The advisory doesn't name an incident. The Pioneer reports it came amid protests in Delhi over electoral roll revisions, and that ministry sources claimed agency reports showed Meta's algorithms favouring some content over others. That claim comes from unnamed sources, not a published audit, and Meta hasn't commented publicly.
Will my Instagram or Facebook feed change because of MeitY's advisory?
Possibly, but you probably won't be told. Meta reports its action to the ministry, not to users, and neither side has said what will change. The visible part, if Meta follows the advisory, would be more context labels on old, edited or misdated videos of public events.
- Long Press · Daily Five21 Oct 202603 / 05
tipYour YouTube Home is trainable. Here's the reset
You can reset YouTube's Home feed with tools already in the app: the three dots on any video give you Not interested and Don't recommend channel, and deleting plus pausing watch history wipes the slate, as YouTube Help lays out. Most of us just scroll past the stuff we hate, and that teaches the app nothing. Saying no is quicker. Do the full wipe if a kid or a guest took over your account for a week, then switch history back on so it learns from you again.
The questionHow do I reset YouTube recommendations on my Home feed?
In one lineUse the three-dot menu on any video and pick Not interested or Don't recommend channel to remove specific stuff. For a full reset, delete your YouTube watch history and turn it off; YouTube's help pages say that's how you stop video recommendations on Home until you switch history back on.
The problem
Watch three cooking videos at a relative's place, or hand your phone to a kid for an evening, and YouTube's Home spends the next month convinced that's who you are.
What changed
On any video, the three-dot menu has Not interested and Don't recommend channel, which take that video or that channel out of Home and Watch Next, per YouTube Help. For a clean slate, delete your watch history and turn it off; YouTube's page on removing recommended content from Home says that stops video recommendations there.
The catch
With history off, YouTube has nothing to go on, so Home goes quiet and search gets more generic. It's a reset, not a way to live.
What it means for you
Use Don't recommend channel on the three worst offenders in your feed today. If things are beyond saving, delete and pause history for a day, then turn it back on and watch what you actually want.
Questions people ask
Does Don't recommend channel block a YouTube channel completely?
No. It stops that channel's videos, Shorts or posts appearing in your Home and Watch Next recommendations, according to YouTube's help page, but the channel itself isn't blocked. That makes it the right tool for channels you watched once and never want suggested again, without deleting anything or unsubscribing from anyone.
What happens to YouTube recommendations if I turn off watch history?
YouTube's help says deleting and turning off watch history is the way to stop video recommendations on Home, and that past watches stop shaping suggestions and search results while it's off. The app feels much emptier, so treat it as a reset button and turn history back on once the junk is gone.
- Long Press · Daily Five21 Oct 202604 / 05
aiIndiaA 20-second AI check before you pay any bill-due link
Before you tap any 'your power will be cut tonight' link, paste the message into ChatGPT or Gemini with this prompt and let it list the red flags: "Check this message for scam signs. Who does it claim to be from, and does that match how that organisation really contacts people? List every red flag (urgency, links, APK files, personal numbers, payment requests). Then tell me the official way to verify it without using anything in the message. Message: [paste here]" AI isn't smarter than you here. It's calm when you're not, and these scams run entirely on panic. On October 20, The420.in reported an Indore trader who paid Rs 1.45 lakh through exactly this kind of fake electricity-bill link and got all of it back, because the complaint went in fast enough for police to stop the money.
The questionHow can I use ChatGPT to check if a bill message is a scam?
In one linePaste the message, with your personal numbers removed, into ChatGPT or Gemini and ask it to list scam red flags and the official way to verify the bill. It's a quick second opinion, not proof. Confirm in your electricity board's own app, and if you've already paid, call 1930.
The problem
Fake electricity, gas and FASTag 'bill due' messages work because they land at 9 pm with a deadline, and nobody thinks clearly with a power cut threatened in an hour.
What changed
In the Indore case reported by The420.in on October 20, a textile trader paid Rs 1,45,000 by credit card through a WhatsApp link from someone posing as an electricity official. After a complaint on the national cybercrime portal, police traced it through the merchant channel and refunded the full amount within 24 hours. The prompt above would have flagged that message in seconds.
The catch
A chatbot only reads the text you give it. It can't open the link safely or see your real bill, and a well-written fake can sail through, so a clean answer only means it isn't obviously a scam.
What it means for you
Save the prompt in your notes app now, check any bill in your discom's own app or website, and if money has already gone, call 1930 within the hour.
Questions people ask
What should I do if I paid through a fake electricity bill link in India?
Call 1930 or report at cybercrime.gov.in straight away. In an Indore case reported on October 20, 2026, a trader paid Rs 1.45 lakh by credit card through a fake electricity official's WhatsApp link, complained on the national portal, and police reversed the full amount within 24 hours, before it reached the fraudsters' accounts.
Is it safe to paste a suspicious message into an AI chatbot?
Paste the message, not your life. Remove your name, consumer number, account and phone numbers first, and never paste an OTP or UPI PIN. The chatbot only reads text, so it can't safely check whether a link is live, and you shouldn't tap the link to test it yourself either.
Can ChatGPT or Gemini catch every scam message?
No. They're good at the usual patterns, like a deadline, a personal mobile number posing as an official, or an APK attachment, but a carefully written fake can pass. Treat a clean result as not obviously a scam, then verify with the sender through a phone number or app you found yourself.
SourcesThe420.in ↗
- Long Press · Daily Five21 Oct 202605 / 05
hot takeOpenAI just put a price tag on Indian IT's slowest work
OpenAI says GPT-6.1 Sol now migrates COBOL with 86.6% accuracy at low reasoning effort, where the previous model managed 5.4%, and 97.9% at high effort, according to its October 19 post. This is aimed at the most comfortable revenue in Indian IT. Mainframe modernisation has been the multi-year, many-hands contract that nobody wanted to touch and everybody was happy to bill. These are OpenAI's own tests and it still says experts must review the output, so the work doesn't disappear. But a bank CIO who's read this won't sign a three-year migration at last year's rates, and the firms that reprice first will keep the deals.
The questionCan OpenAI's GPT-6 models really migrate COBOL code to Java or Python?
In one lineOpenAI said on October 19, 2026 that GPT-6.1 Sol scored 86.6% on its COBOL migration test at low reasoning effort, against 5.4% for GPT-5.6 Sol, and 97.9% at high effort. These are OpenAI's own benchmarks, and it says production code still needs expert review and testing.
The problem
Banks and insurers sit on decades of COBOL that few young engineers can read, and rewriting it has been slow and risky enough to keep large teams busy for years.
What changed
In an October 19 post, OpenAI says GPT-6.1 Sol reaches 86.6% on COBOL migration at low effort, against 5.4% for GPT-5.6 Sol, and 97.9% at high effort, at 39% lower API cost. GPT-6 Astra averaged 87.4% across 20 languages on its MultiLangBench test, with COBOL the hardest at 72.1%.
The catch
These are OpenAI's numbers on OpenAI's tests, no customer is named, and the company says production code still needs expert review. A benchmark is not a bank's 40-year-old core system full of undocumented quirks.
What it means for you
If you work in IT services, the skill that keeps its value is the checking: knowing what the old system really does and proving the new one matches it. If you buy these projects, ask your vendor how the quote changes now.
Questions people ask
What does OpenAI's COBOL result mean for Indian IT services companies?
Legacy modernisation, rewriting old bank and insurer COBOL into Java or other modern code, is long, people-heavy, billable work for Indian IT services firms. If OpenAI's numbers hold up on real projects, clients will expect those migrations faster and cheaper. OpenAI's post itself doesn't mention India or any IT services company.
Why is COBOL still important in 2026?
Because banking still runs on it. Citing IBM, OpenAI says COBOL supports more than 40% of online banking systems, 80% of in-person credit card transactions and 95% of ATM transactions. Few new developers learn it, which is a big reason migrating away from it has stayed slow and expensive.
Which OpenAI models are best at COBOL and Fortran?
OpenAI named GPT-6 Astra, which averaged 87.4% across 20 languages on its MultiLangBench test, including 72.1% in COBOL and 83.5% in Fortran, and GPT-6.1 Sol for COBOL migration. GPT-6 Luna scored lower at 40.3% but costs about 78% less per sample than GPT-5.6 Sol.
SourcesOpenAI ↗Engadget ↗
Taps
Quick hits. One line each, for when a long press is too much.
- IndiaInc42 ↗
Flipkart is reportedly launching food delivery by the end of October
Called Eat-in by Minutes, it's pitching restaurants about 10% commission against the usual 20-25%, per Inc42. Flipkart hasn't confirmed it.
- IndiaInc42 ↗
Centre weighs a hardware security layer for Starlink and other satcom operators
All satellite traffic would pass through a government-overseen device; specs and timing aren't final, Inc42 reports. No satcom service has launched in India yet.
- IndiaInc42 ↗
CCPA gets 3,940 explosive-chemical listings pulled from IndiaMART, Justdial and others
IndiaMART alone had 3,092. Dial4Trade and ExportersIndia were fined Rs 30 lakh between them, per Inc42.
- GlobalThe Register ↗
A power failure in one Google Cloud London zone knocked UK banks offline
Lloyds, HSBC and Revolut were among those hit from 06:08 UTC on October 20, and service mostly returned eight hours later, per The Register.
- GlobalThe Deep View ↗
Inception's Mercury 3 diffusion model claims around 1,000 tokens a second
Launch price is $0.10 per million input tokens and it claims 4-8x the speed of Haiku 5.5 and GPT-6 Luna, per The Deep View. Vendor claims so far.