The Day Anthropic Signed for Generators── Securing power, narrowing losses, AI completion rates, a bill against layoffs
Download the video (MP4, 12 MB)
AI news is usually told as a race over whose model is smartest. But when I lined up today's events, I saw a different picture. AI is turning into heavy infrastructure. It runs on huge amounts of electricity and borrowed money. Yet the cases where it finishes assigned work are still limited. We handle drug information. What should we look at before we rely on an AI?
Can we judge it by the size of the company?Today we move from power and money to work quality, national interests, reading the literature, patient behavior and worker anxiety. Size alone is not enough. That is my answer from today.
01Anthropic Secures Power
Source Reuters ／ Brave New Coin ／ The Morning Call
Companies that build AI are no longer just buying computing parts. They have started arranging their own power equipment. That means parts alone are no longer enough. Securing electricity is becoming a management issue on par with technology.
$3.4B for generators
Anthropic signed a $3.4 billion deal for generators for AI data centers, Reuters reported, citing the FT.
No higher power bills
OpenAI's CFO defended data centers after the midterm backlash and promised not to raise electricity bills.
Jobs for community college grads
A Pennsylvania opinion piece argued that data centers can give state residents jobs that support a family.
The moves on screen point the same way. The ceiling on AI growth now seems set by how much electricity exists. So builders lock in power first. Meanwhile, builders promise to protect household budgets. And local voices hope the facilities will create work.
Money on this scale flows to power because power is no longer an add-on. Before speed, the question is whether electricity arrives at all.
These facilities draw power day and night and give off heat nonstop. If the power stops, every AI running on them stops too. So where they are built depends on where power can be drawn.
This is not a distant issue for pharma professionals. Your in-house AI runs only as reliably as some distant plant and grid. If local backlash grows, it may affect the price of the service, or whether it continues.
So where does the money to buy that power come from? Next, we follow the money.
02Smaller Losses, Debt Warnings
Source shattered.io ／ 24/7 Wall St. ／ FourWeekMBA
Buying generators and building facilities takes enormous sums. The nature of that money shapes how safe users can feel relying on AI.
| Report | Figure | Implication |
|---|---|---|
| Amended S-1 (shattered.io) | Operating loss narrowed from $5.2B to $1.3B | Losses shrank ahead of the listing |
| 24/7 Wall St. | Warns $895B AI investment could become a Wall Street debt problem | The scale of investment is seen as a debt risk |
| FourWeekMBA | Infrastructure took 42% of $100M+ rounds in one week | Much of the big money went to facilities |
The table holds good news and bad news side by side. The deficit from the main business shrank sharply before the listing. But with investment this large, some ask whether borrowed money can really be repaid.
A smaller deficit may signal an exit from losing more with every sale. But suppose the debt for facilities assumes sales keep growing. If that assumption breaks, only the duty to repay remains. And debt weighs heaviest when the economy cools.
Even among large fundraisings, a sizable share went to the facilities beneath AI. Investor money is also flowing to the foundation.
For firms building AI into operations, a provider's finances matter as much as its quality. If a provider stumbles, price rises, spec changes, or even shutdown can follow. So plan from procurement not to depend on one vendor. Reading a provider's financial results before signing is part of checking quality, in my view.
So is the AI that providers spent so much on actually finishing the work it is given?
03Do Agents Finish the Job?
Source BigGo Finance ／ MLQ.ai ／ Yahoo News UK ／ BigGo Finance
Is AI doing work worth all this money? The question is not how much it writes, but whether the job gets done. If output grows but unfinished work piles up, the burden on reviewers actually grows.
| Report | Fact | Implication |
|---|---|---|
| GitHub (Parikh) | Agents created 40% of pull requests | Much of the coding work is already delegated |
| Alibaba's Co-Design Bench | Even the best coding agent completed only 26.4% of real requests | More requests go unfinished than finished |
| Chicago law professor | AI deleted a syllabus and a year of lecture notes | Delegated actions produced irreversible results |
| Modal's CTO | Warns that cheap DIY agent sandboxes will fail | How the run environment is built decides reliability and safety |
The table shows a gap between volume and finish. AI that acts for people now handles a large share of the output. Yet in one evaluation, only a fraction of genuine requests were carried through. One person stepped out for lunch and came back to find a year of teaching records erased.
Such AI picks up tools and keeps going through many tasks without asking. Not checking each step is the source of its convenience. But the same trait allows errors that cannot be undone, where no one is watching.
Pharma work assumes that records are kept. If trial records or submission documents vanish, a redo is not enough. If you bring in such AI, limit what it may touch. Require human approval before any deletion. Keeping its workspace cut off from other systems serves the same goal.
And a fight has broken out over which countries may receive the computing power that runs such AI.
04CEOs Split on Chips for China
Source Wccftech
The computing power behind this work comes from semiconductors. Industry leaders have now clashed head-on over who may buy them.
- Position
- Opposes a total ban on AI chips to China
- Statement
- “I completely disagree” / “America is about competition”
- Position
- Seeks a total ban on AI chips to China
- Statement
- No new statement from him in today's sources
Huang's rebuttal as reported by Wccftech
The two on screen work in the same field but give opposite answers. One wants sales stopped altogether. The other wants to keep selling. But today's reporting carries no new words from one of them. So treat this conflict as a picture drawn by the press. It is safest to assume neither view is separate from its own business.
Whether exports stop is decided by governments, not companies. Which view wins depends on lawmakers and officials. The companies are lobbying toward that decision. And once a government decides, it is hard to reverse for business reasons.
Drug companies also use AI across borders. If research sites or contractors are abroad, the tools and computing you can use may change suddenly with one government decision. When choosing a tool, check which country's rules it depends on. Even one backup option helps prepare for sudden change.
Now let us return to our own daily work. With that computing power, how will AI change the way we read the literature?
05AI That Reads Papers as Images
Source DR Pixel: Page-Native Deep Research with Vision-Language Models
Drug decisions rest on numbers published in papers. Many of those numbers sit in tables and charts, not the body text. So an AI that reads only text can miss the most important parts.
These methods used to first turn a page into a plain string of characters. In that step, table rows and columns fall apart. Chart meaning is lost too. The new approach searches the page as it looks, and reasons on it as it looks. Without extra tuning, it reportedly did better than converting to text.
| Evaluation | Result | Implication |
|---|---|---|
| Authors' own benchmark | Retrieval recall up ~15 points; accuracy up 10+ points | Own yardstick; needs outside confirmation |
| External benchmark not used in training | Within 1.8 points of an ideal baseline given the correct pages | Close to ideal even on unseen documents |
| OCR methods compared, and cost | Not stated in the abstract | Not yet enough to decide on adoption |
But the improvement after further tuning was measured with a test the authors made. A small gap from the ideal on an outside test is encouraging. Still, you cannot commit without knowing the comparison and the cost. The better a number looks, the sooner I check who measured it, and how.
This matters for literature work. An AI summary can misread a number and still read smoothly. So for now, efficacy and side-effect numbers still need checking against the original page. The new approach is a candidate to cut that effort.
Checking numbers against the source is not only for experts. The same thing is happening on the patient side.
06In the Pharma Field
Source Healthcare IT Today ／ The Guardian
Change is also happening among people who receive medicines. US reports say many people already consult AI when picking health coverage. Medical information seems to reach people more often without passing through clinicians.
48% for insurance choices
48% of Americans used AI to review their health insurance options, per a Healthcare IT Today weekly roundup headline.
68% verify drug information
In the same headline, 68% said they would independently verify medication details generated by AI.
A flood of complaint letters
UK MPs' offices, councils, schools and others are flooded with chatbot-written complaints and claims (The Guardian).
The moves on screen show two sides of people. They lean on AI, yet grow careful when medicines come up. They say they confirm with other sources. AI may be the first place they ask, but not the last check. Meanwhile, in Britain, easy AI-written grievances are swamping public offices. They are hard to read and hard to handle.
Careful people look for somewhere to check. One such place could be information from drug makers. But as AI-written grievances grow, truly urgent requests may get buried. If a side-effect report is buried, that becomes a safety issue.
For drug makers, people who come to check are important. Is official information easy to find, accurate, and easy to compare with AI answers? Inquiry desks may also get more long, AI-drafted questions. Each needs a correct answer, with a record kept.
And the jobs of the people who staff those desks are also starting to shake as AI spreads.
07Anti-Layoff Bill and Job Cuts
Source Philstar.com ／ Tech Observer Magazine ／ ad-hoc-news.de ／ Vietnam.vn
Finally, the people who work. The more AI takes over tasks, the more real the choice to cut staff becomes. Efficiency quickly becomes a question of livelihoods. How is society responding?
| Actor | Reported move | Implication |
|---|---|---|
| Philippine Senate (Sen. Tulfo) | Filed labor law amendment to protect workers from AI layoffs | A bill to prevent layoffs reached a national legislature |
| Indian CEOs (KPMG survey) | Plan AI-driven workforce cuts while hiring more | Hiring and cutting run side by side |
| ServiceNow | Job cuts in UK, US and India; pivot to AI agents | Shifting spending from people to AI |
| Lao Gan Ma (China) | Boycott over replacement rumors; online sales briefly fell up to 90% | Rumor alone made consumers cut sales |
The table shows moves that differ by position. In one country, a bill would use law to prevent dismissals. On the business side, some leaders reportedly plan to trim staff with AI while still recruiting. And one company faced a boycott over a mere rumor, and its web sales dropped sharply for a while. Protection by law and cuts arrive at the same time. People seem sensitive enough to act on hearsay.
This is also a matter of reputation for firms adopting AI. Replacing people without visible care can drive away customers and staff. For drug makers, it touches the trust of patients and clinicians. Decide what to give AI and what to keep for people, in a form you can explain. The public watches how you replace, more than whether. Without explanation, talk of cuts spreads as rumor, faster than the company's own words.
What we put into that explanation answers the question that runs through all of today's events.
Watch the terms of Anthropic's generator deal and whether the pledge not to raise electricity bills holds.
Open the full transcript
Intro
AI news is usually told as a race over whose model is smartest. But when I lined up today's events, I saw a different picture. AI is turning into heavy infrastructure. It runs on huge amounts of electricity and borrowed money. Yet the cases where it finishes assigned work are still limited. We handle drug information. What should we look at before we rely on an AI?
Can we judge it by the size of the company?Today we move from power and money to work quality, national interests, reading the literature, patient behavior and worker anxiety. Size alone is not enough. That is my answer from today.
CH 01 Anthropic Secures Power
Companies that build AI are no longer just buying computing parts. They have started arranging their own power equipment. That means parts alone are no longer enough. Securing electricity is becoming a management issue on par with technology.The moves on screen point the same way. The ceiling on AI growth now seems set by how much electricity exists. So builders lock in power first. Meanwhile, builders promise to protect household budgets. And local voices hope the facilities will create work.Money on this scale flows to power because power is no longer an add-on. Before speed, the question is whether electricity arrives at all.These facilities draw power day and night and give off heat nonstop. If the power stops, every AI running on them stops too. So where they are built depends on where power can be drawn.This is not a distant issue for pharma professionals. Your in-house AI runs only as reliably as some distant plant and grid. If local backlash grows, it may affect the price of the service, or whether it continues. So where does the money to buy that power come from?
Next, we follow the money.
CH 02 Smaller Losses, Debt Warnings
Buying generators and building facilities takes enormous sums. The nature of that money shapes how safe users can feel relying on AI.The table holds good news and bad news side by side. The deficit from the main business shrank sharply before the listing. But with investment this large, some ask whether borrowed money can really be repaid.A smaller deficit may signal an exit from losing more with every sale. But suppose the debt for facilities assumes sales keep growing. If that assumption breaks, only the duty to repay remains. And debt weighs heaviest when the economy cools.Even among large fundraisings, a sizable share went to the facilities beneath AI. Investor money is also flowing to the foundation.For firms building AI into operations, a provider's finances matter as much as its quality. If a provider stumbles, price rises, spec changes, or even shutdown can follow. So plan from procurement not to depend on one vendor. Reading a provider's financial results before signing is part of checking quality, in my view. So is the AI that providers spent so much on actually finishing the work it is given?
CH 03 Do Agents Finish the Job?
Is AI doing work worth all this money?The question is not how much it writes, but whether the job gets done. If output grows but unfinished work piles up, the burden on reviewers actually grows.The table shows a gap between volume and finish. AI that acts for people now handles a large share of the output. Yet in one evaluation, only a fraction of genuine requests were carried through. One person stepped out for lunch and came back to find a year of teaching records erased.Such AI picks up tools and keeps going through many tasks without asking. Not checking each step is the source of its convenience. But the same trait allows errors that cannot be undone, where no one is watching.Pharma work assumes that records are kept. If trial records or submission documents vanish, a redo is not enough. If you bring in such AI, limit what it may touch. Require human approval before any deletion. Keeping its workspace cut off from other systems serves the same goal. And a fight has broken out over which countries may receive the computing power that runs such AI.
CH 04 CEOs Split on Chips for China
The computing power behind this work comes from semiconductors. Industry leaders have now clashed head-on over who may buy them.The two on screen work in the same field but give opposite answers. One wants sales stopped altogether. The other wants to keep selling. But today's reporting carries no new words from one of them. So treat this conflict as a picture drawn by the press. It is safest to assume neither view is separate from its own business.Whether exports stop is decided by governments, not companies. Which view wins depends on lawmakers and officials. The companies are lobbying toward that decision. And once a government decides, it is hard to reverse for business reasons.Drug companies also use AI across borders. If research sites or contractors are abroad, the tools and computing you can use may change suddenly with one government decision. When choosing a tool, check which country's rules it depends on. Even one backup option helps prepare for sudden change. Now let us return to our own daily work. With that computing power, how will AI change the way we read the literature?
CH 05 AI That Reads Papers as Images
Drug decisions rest on numbers published in papers. Many of those numbers sit in tables and charts, not the body text. So an AI that reads only text can miss the most important parts.These methods used to first turn a page into a plain string of characters. In that step, table rows and columns fall apart. Chart meaning is lost too. The new approach searches the page as it looks, and reasons on it as it looks. Without extra tuning, it reportedly did better than converting to text.But the improvement after further tuning was measured with a test the authors made. A small gap from the ideal on an outside test is encouraging. Still, you cannot commit without knowing the comparison and the cost. The better a number looks, the sooner I check who measured it, and how.This matters for literature work. An AI summary can misread a number and still read smoothly. So for now, efficacy and side-effect numbers still need checking against the original page. The new approach is a candidate to cut that effort. Checking numbers against the source is not only for experts. The same thing is happening on the patient side.
CH 06 In the Pharma Field
Change is also happening among people who receive medicines. US reports say many people already consult AI when picking health coverage. Medical information seems to reach people more often without passing through clinicians.The moves on screen show two sides of people. They lean on AI, yet grow careful when medicines come up. They say they confirm with other sources. AI may be the first place they ask, but not the last check. Meanwhile, in Britain, easy AI-written grievances are swamping public offices. They are hard to read and hard to handle.Careful people look for somewhere to check. One such place could be information from drug makers. But as AI-written grievances grow, truly urgent requests may get buried. If a side-effect report is buried, that becomes a safety issue.For drug makers, people who come to check are important. Is official information easy to find, accurate, and easy to compare with AI answers?
Inquiry desks may also get more long, AI-drafted questions. Each needs a correct answer, with a record kept. And the jobs of the people who staff those desks are also starting to shake as AI spreads.
CH 07 Anti-Layoff Bill and Job Cuts
Finally, the people who work. The more AI takes over tasks, the more real the choice to cut staff becomes. Efficiency quickly becomes a question of livelihoods. How is society responding?
The table shows moves that differ by position. In one country, a bill would use law to prevent dismissals. On the business side, some leaders reportedly plan to trim staff with AI while still recruiting. And one company faced a boycott over a mere rumor, and its web sales dropped sharply for a while. Protection by law and cuts arrive at the same time. People seem sensitive enough to act on hearsay.This is also a matter of reputation for firms adopting AI. Replacing people without visible care can drive away customers and staff. For drug makers, it touches the trust of patients and clinicians. Decide what to give AI and what to keep for people, in a form you can explain. The public watches how you replace, more than whether. Without explanation, talk of cuts spreads as rumor, faster than the company's own words. What we put into that explanation answers the question that runs through all of today's events.
Wrap-up
Today's events show that the worth of AI is not measured by size. Who carries the heavy load of electricity and money?Who checks that assigned work is really done?We, as users, should watch those two things. Keep both at hand when choosing an AI. Look at how well the aftermath is handled, not how big the numbers are. Tomorrow, we follow where the builders shift that burden.
- CH 01Reuters「Anthropic secures $3.4 billion generator deal for AI data centers, FT reports」 reuters.com
- CH 01Brave New Coin「OpenAI CFO Defends AI Data Centers After Midterm Backlash, Promises No Higher Electricity Bills」 bravenewcoin.com
- CH 01The Morning Call「Opinion: For Pa. community college graduates, data centers are an opportunity」 mcall.com
- CH 02shattered.io「Anthropic S-1: Loss Narrows to $1.3B From $5.2B [2026]」 shattered.io
- CH 0224/7 Wall St.「Anthropic’s $895 Billion AI Gamble Could Become Wall Street’s Next Debt Nightmare」 247wallst.com
- CH 02FourWeekMBA「AI Rounds of $100M+ Raised $3.32B in a Week; Infra Took 42%」 fourweekmba.com
- CH 03BigGo Finance「Microsoft's Jay Parikh on GitHub at 225 Million Developers: 40% of Pull Requests Are Now Written by Agents」 finance.biggo.com
- CH 03MLQ.ai「Alibaba's Co-Design Bench: the best coding agent fully completes only 26.4% of real user requests」 mlq.ai
- CH 03Yahoo News UK「Chicago law professor returns from lunch to find AI deleted his syllabus and a year's lecture notes」 uk.news.yahoo.com
- CH 03BigGo Finance「Modal CTO Akshat Bubna: Building Agent Sandboxes the Cheap Way Will Fail on Reliability, Security and Scale」 finance.biggo.com
- CH 04Wccftech「NVIDIA’s Jensen Huang Rejects Anthropic’s Total China Chip Ban Push, Says “I Completely Disagree” And Believes “America Is About Competition”」 wccftech.com
- CH 05DR Pixel: Page-Native Deep Research with Vision-Language Models（「論文の情報の多くは表・図・数式にあるが、文献を調べる AI の多くは、論文を OCR で文字に起こしてから読む。」）
- CH 06Healthcare IT Today「Bonus Features – November 8, 2026 – 48% of Americans have used AI to review their health insurance options, 68% would independently verify AI-generated medication details, plus 34 more stories」 healthcareittoday.com
- CH 06The Guardian「‘AI clog’: UK public bodies battle onslaught of chatbot-generated complaints」 theguardian.com
- CH 07Philstar.com「System to shield workers from AI layoffs pushed」 philstar.com
- CH 07Tech Observer Magazine「Indian CEOs plan AI-driven workforce cuts even as hiring rises: KPMG」 techobserver.in
- CH 07ad-hoc-news.de「ServiceNow Trims Headcount While Betting Its Future on AI Agents and a ChatGPT Foothold」 ad-hoc-news.de
- CH 07Vietnam.vn「Lao Gan Ma is in turmoil due to rumors of using AI to replace workers.」 vietnam.vn
Articles used
- AI Daily News 2026-11-09
- AI Daily News headline translations 2026-11-09 morning
- AI Daily News headline translations 2026-11-08 evening
- AI that researches papers as page images, without converting them to text
Today's related reports
- AI Daily News November 9, 2026
- AI & Economy News November 9, 2026
- AI & Finance News November 9, 2026
The narration is synthetic speech. The content draws only on this site's articles from the same day. Each edition passes seven plain-language checks before publication. That means known obstacles to comprehension are held below threshold — it is not a guarantee of comprehension.