Lenders Set Terms on AI Chips── Terms on AI loans, claims of AI results, virtual tests in plants

Download the video (MP4, 11 MB)
Why would lenders want computer hardware kept on the ground?Today: how to judge AI money and AI claims. A company known for rockets raised a large sum to buy processors for AI. Reports say the people who lent the money asked that those processors stay on the ground. People who put up cash seem to care less about a grand vision than about getting paid back. That same day, shares tied to AI were sold, and forecasts split. For those of us using AI in pharma work, this matters. What evidence makes us trust it, and how much should we pay?
Share prices do not answer that.
01Conditions Attached to AI Hardware
Source The Motley Fool / The Spectator
Computing for AI needs equipment that costs enormous sums. What do the people lending that money ask for in return?
A company known for rockets has borrowed heavily to buy processors for AI. The headlines say its lenders want those processors kept on the ground. A machine within reach can be seized if repayments stop. That seems the natural reading of the lenders' thinking. But this reason is my own inference, not something reported.
Processors alone do not compute. They need electricity. New generating machines take a long time to arrive after an order. So older engine types are reportedly being put to work feeding data centres.
This is not remote from a pharma company. The AI tools we use every day run on machines in some building, fed by that electricity. If electricity or processors run short, the amount we can use, or the price, may change. If you build AI into a work process, plan first how people will keep the work going on the day it stops.
Even with processors and electricity in short supply, will money keep flowing into AI? That same day, the markets were sharply divided.
02Bulls and Bears on the Same Day
Source BigGo Finance / BigGo Finance / Marcus on AI | Substack / Fortune
On the day shares linked to AI were sold, a forecast also said companies would keep spending more on it. Why are views so far apart?
| Signal | What was reported | What it means |
|---|---|---|
| UBS forecast | AI spending of $1.5 trillion in 2028; Meta and Amazon as winners | View that spending keeps rising |
| Rating on Microsoft | BNP Paribas keeps $632 target; praises full-stack strategy | A firm owning the stack from base to product kept its rating |
| Gary Marcus commentary | Claims AI stocks fell over 50% in 5 months | Skeptic's commentary; figures need verification |
A bullish forecast and bearish share-price falls appeared in the same day's news
The optimists expect corporate outlays on AI to keep growing. A firm that owns everything from base components to finished products also kept a high rating. The pessimists say these share prices have already collapsed. But that view comes from a known doubter, and the figures should be checked elsewhere. One looks at how much money will be spent. The other asks whether real value sits behind the price paid.
The sheer size of the forecast shows how strong expectations are. And the more money pours in, the harder the fall if results do not come.
For a pharma company setting an AI budget, this split is useful. Spending more and spending well are different things. Before adopting AI, you need a way to check for yourselves whether the effect justifies the cost. If budgets rise and fall with market mood, you cannot later explain what the decision was based on.
On checking effects, another announcement that day caught my eye. It was about results that AI is said to have produced.
03AI Results and How to Verify Them
Source WSJ / New Scientist / Forbes
When we hear that AI has achieved something big, how far should we believe it?
| Result | What was reported | What it means |
|---|---|---|
| Proof of the Riemann hypothesis | Anthropic mathematician claims a proof with AI assistance | Peer-reviewed verification unconfirmed; mathematicians must check |
| GPU design improvement | OpenAI models improved AMD's flagship GPU design by up to 14% | Reported by Forbes as OpenAI's account |
The two announcements share one feature. Both came from the side that produced the result. One says a long-standing puzzle in mathematics has been cracked. The other says the layout of a computing part was made better. If true, both are big steps. But until outside experts have checked them, it is fair to treat them as assertions. The hardware story, too, reaches us as the developer's own account.
A proof is accepted only after other researchers read it line by line and look for errors. Being widely reported and being accepted as correct can be far apart in time.
This order is very like the world of medicines. A company may say a drug works, but it is not approved until an independent review passes it. Think of using AI analysis in internal documents or external explanations. If you cannot show who checked the content, the result cannot serve as evidence. Keeping claims and confirmation apart is still required when the work involves AI.
Checking claims takes effort and money. That burden has started to show up in the price of AI.
04Will Free AI Last?
Source 36Kr / Dealroom / finance.biggo.com
Will using AI at no cost stay normal from here on?
Free offerings withdraw
Chinese outlet 36Kr reported that limits on free AI usage quotas are spreading and DeepSeek-style free providers are exiting
Domestic money invests
A Hengdian-linked fund in China reportedly raised $247 million from Zhejiang firms for a stake in DeepSeek
Running it locally
DeepSeek V4.1 Flash was run on a homemade server with 12 used V100s. A former member released the agent Linger
In China, providers that gave service away are cutting how much people can use, or leaving the market. Running these models costs money every single day. Meanwhile, domestic capital is gathering around a leading developer, and the likely survivors are becoming clear. Some people now run models on their own machines. The choice is shifting to paying, or running it yourself. The plan of attracting users at no cost and charging later may be hitting a wall.
A cap on use means the tool could stop in the middle of important work.
If staff rely on no-cost tools for their work, those tools could vanish one day. With such tools, it can also be hard to confirm how the data you enter is handled. Running models in-house keeps data inside, but then upkeep and checking fall on you. The company should first decide which tools may be used for work.
So how much are the big drugmakers paying for AI? We now look at those who pay, not those who charge.
05What Pharma Majors Pay
Source Oncodaily / Foreign Policy Journal / TradingView
We have talked a lot about money. But where did drug companies themselves put theirs?
| Deal | Reported amount | What it means |
|---|---|---|
| Renewal of a pharma major's oncology deal with Tempus AI | $200M; partnership extended to 2031 | Headline does not show if this is the total or the renewal amount |
| Another pharma major's joint venture framework with nference | No amount given | Targets obesity, diabetes and heart disease |
| Share sales by Tempus AI CEO and others | Over $645K | Personal sales, not corporate investment |
The only amount reported as big pharma's own spending was the $200M Tempus AI renewal
Only one deal came with a clear figure. It extended an existing cancer partnership. It stayed with a partner it already had. Another deal sets up shared business with a partner in lifestyle-related disease, but no sum was given. The third is a share sale by executives as individuals, and has nothing to do with company spending. In practice, that one extension is the only corporate outlay with a known size.
This kind of shared business differs from a contract where you simply pay. Both gains and losses are split. Nobody picks it without planning a long relationship.
Within these reports, the big drugmakers are not throwing money at AI at random. They seem to prefer long ties with partners they already know. With share prices swinging hard, that is a cautious choice. When choosing an AI partner, starting small and expanding after seeing results is a useful order. Working deeply with one partner, rather than many, also makes results easier to check.
Try small, check results, then expand. This idea is taking shape inside drug factories, too.
06On the Pharma Plant Floor
When new machinery arrives, what happens if you test it live and it fails? Every hour it is down delays medicines reaching patients.
At one site, AI reads the instructions for checking machinery. The machinery is then run inside a factory built in a computer, before trials on the real line. Finding problems early cuts the time real lines stand idle. For a site making medicines every day, testing without stopping the line is a big gain.
AI does not skip the real check. It sits just before it. The effort falls, but the check itself remains.
At a closely watched site, every change to machinery or methods raises the question of records. The newspaper behind the story also said moving quickly to a site without people is hard. The realistic path keeps human checks, and uses AI to speed up the preparation before them. Problems found by AI are still judged by people at the end, and that judgement is recorded. Quality and manufacturing staff need a design that keeps AI on the helper side of checking.
Keeping human checks also raises a question. When something goes wrong, who is to blame?
07What Is Being Overlooked
Source Bergen Record / Investing.com / PYMNTS.com / Anadolu Ajansı
When a forged picture made by AI hurts someone, is the company that built the tool to blame?
- Position
- Plaintiff in lawsuit over Grok fake images
- Statement
- Grok was designed to make sexual images
- Position
- Defendant in the UK lawsuit
- Statement
- Denies liability. Obviously fake images do not violate privacy
In the US, too, the Muslim Girl founder sued xAI over sexualized Grok image edits
The people suing say the tool was built from the start to produce such pictures. The company being sued replies that a picture anyone can see is forged does not invade private life. The dispute is whether blame reaches beyond the person who made a picture, to the firm that released a tool able to make it.
New York State
Created a dedicated AI enforcement unit. Reports say oversight at the state level is being strengthened
UK parliamentary committee
Called for a new AI law, citing human rights risks. The lawsuit over Grok fake images is also ongoing
European AI summit
Tusk announced a high-level AI summit to close the computing-resource gap with the US and China
Governments are moving too. One state has set up a team to police AI. A group of lawmakers wants new legislation to protect people. Europe plans a meeting to keep pace with its two big rivals. They are trying to build protection and competitiveness at once.
Drug companies, too, are starting to create pictures and text with AI. A picture in patient material might end up resembling someone who actually exists. If that face hurts someone, the tool maker may not protect you. With tool makers refusing to accept blame, users must check for themselves and keep records.
In the end, every story comes back to one question. Who checks for the user, and who carries the blame?
Whether the AI stock correction continues, together with mathematicians' verification of the Riemann hypothesis proof claim.
Open the full transcript
Intro
Why would lenders want computer hardware kept on the ground?Today: how to judge AI money and AI claims. A company known for rockets raised a large sum to buy processors for AI. Reports say the people who lent the money asked that those processors stay on the ground. People who put up cash seem to care less about a grand vision than about getting paid back. That same day, shares tied to AI were sold, and forecasts split. For those of us using AI in pharma work, this matters. What evidence makes us trust it, and how much should we pay?
Share prices do not answer that.
CH 01 Conditions Attached to AI Hardware
Computing for AI needs equipment that costs enormous sums. What do the people lending that money ask for in return?A company known for rockets has borrowed heavily to buy processors for AI. The headlines say its lenders want those processors kept on the ground. A machine within reach can be seized if repayments stop. That seems the natural reading of the lenders' thinking. But this reason is my own inference, not something reported.Processors alone do not compute. They need electricity. New generating machines take a long time to arrive after an order. So older engine types are reportedly being put to work feeding data centres.This is not remote from a pharma company. The AI tools we use every day run on machines in some building, fed by that electricity. If electricity or processors run short, the amount we can use, or the price, may change. If you build AI into a work process, plan first how people will keep the work going on the day it stops. Even with processors and electricity in short supply, will money keep flowing into AI?
That same day, the markets were sharply divided.
CH 02 Bulls and Bears on the Same Day
On the day shares linked to AI were sold, a forecast also said companies would keep spending more on it. Why are views so far apart?The optimists expect corporate outlays on AI to keep growing. A firm that owns everything from base components to finished products also kept a high rating. The pessimists say these share prices have already collapsed. But that view comes from a known doubter, and the figures should be checked elsewhere. One looks at how much money will be spent. The other asks whether real value sits behind the price paid.The sheer size of the forecast shows how strong expectations are. And the more money pours in, the harder the fall if results do not come.For a pharma company setting an AI budget, this split is useful. Spending more and spending well are different things. Before adopting AI, you need a way to check for yourselves whether the effect justifies the cost. If budgets rise and fall with market mood, you cannot later explain what the decision was based on. On checking effects, another announcement that day caught my eye. It was about results that AI is said to have produced.
CH 03 AI Results and How to Verify Them
When we hear that AI has achieved something big, how far should we believe it?The two announcements share one feature. Both came from the side that produced the result. One says a long-standing puzzle in mathematics has been cracked. The other says the layout of a computing part was made better. If true, both are big steps. But until outside experts have checked them, it is fair to treat them as assertions. The hardware story, too, reaches us as the developer's own account.A proof is accepted only after other researchers read it line by line and look for errors. Being widely reported and being accepted as correct can be far apart in time.This order is very like the world of medicines. A company may say a drug works, but it is not approved until an independent review passes it. Think of using AI analysis in internal documents or external explanations. If you cannot show who checked the content, the result cannot serve as evidence. Keeping claims and confirmation apart is still required when the work involves AI. Checking claims takes effort and money. That burden has started to show up in the price of AI.
CH 04 Will Free AI Last?
Will using AI at no cost stay normal from here on?In China, providers that gave service away are cutting how much people can use, or leaving the market. Running these models costs money every single day. Meanwhile, domestic capital is gathering around a leading developer, and the likely survivors are becoming clear. Some people now run models on their own machines. The choice is shifting to paying, or running it yourself. The plan of attracting users at no cost and charging later may be hitting a wall.A cap on use means the tool could stop in the middle of important work.If staff rely on no-cost tools for their work, those tools could vanish one day. With such tools, it can also be hard to confirm how the data you enter is handled. Running models in-house keeps data inside, but then upkeep and checking fall on you. The company should first decide which tools may be used for work. So how much are the big drugmakers paying for AI?
We now look at those who pay, not those who charge.
CH 05 What Pharma Majors Pay
We have talked a lot about money. But where did drug companies themselves put theirs?Only one deal came with a clear figure. It extended an existing cancer partnership. It stayed with a partner it already had. Another deal sets up shared business with a partner in lifestyle-related disease, but no sum was given. The third is a share sale by executives as individuals, and has nothing to do with company spending. In practice, that one extension is the only corporate outlay with a known size.This kind of shared business differs from a contract where you simply pay. Both gains and losses are split. Nobody picks it without planning a long relationship.Within these reports, the big drugmakers are not throwing money at AI at random. They seem to prefer long ties with partners they already know. With share prices swinging hard, that is a cautious choice. When choosing an AI partner, starting small and expanding after seeing results is a useful order. Working deeply with one partner, rather than many, also makes results easier to check. Try small, check results, then expand. This idea is taking shape inside drug factories, too.
CH 06 On the Pharma Plant Floor
When new machinery arrives, what happens if you test it live and it fails?Every hour it is down delays medicines reaching patients.At one site, AI reads the instructions for checking machinery. The machinery is then run inside a factory built in a computer, before trials on the real line. Finding problems early cuts the time real lines stand idle. For a site making medicines every day, testing without stopping the line is a big gain.AI does not skip the real check. It sits just before it. The effort falls, but the check itself remains.At a closely watched site, every change to machinery or methods raises the question of records. The newspaper behind the story also said moving quickly to a site without people is hard. The realistic path keeps human checks, and uses AI to speed up the preparation before them. Problems found by AI are still judged by people at the end, and that judgement is recorded. Quality and manufacturing staff need a design that keeps AI on the helper side of checking. Keeping human checks also raises a question. When something goes wrong, who is to blame?
CH 07 What Is Being Overlooked
When a forged picture made by AI hurts someone, is the company that built the tool to blame?The people suing say the tool was built from the start to produce such pictures. The company being sued replies that a picture anyone can see is forged does not invade private life. The dispute is whether blame reaches beyond the person who made a picture, to the firm that released a tool able to make it.Governments are moving too. One state has set up a team to police AI. A group of lawmakers wants new legislation to protect people. Europe plans a meeting to keep pace with its two big rivals. They are trying to build protection and competitiveness at once.Drug companies, too, are starting to create pictures and text with AI. A picture in patient material might end up resembling someone who actually exists. If that face hurts someone, the tool maker may not protect you. With tool makers refusing to accept blame, users must check for themselves and keep records. In the end, every story comes back to one question. Who checks for the user, and who carries the blame?
Wrap-up
Money comes with conditions. Prices swing on hope. Results arrive as mere claims, and no-cost tools are fading. What we can rely on is not share prices, but results we have checked ourselves. The big drugmakers and the factory both put checking first. Tomorrow, we watch where shaken share prices settle, and whether the proof claim holds up.
- CH 01The Motley Fool「SpaceX Just Raised $32 Billion for AI Chips. Lenders Want Elon Musk to Keep Them on Earth.」 fool.com
- CH 01The Spectator「How old-fashioned gas engines are fueling the AI revolution」 spectator.com
- CH 02BigGo Finance「UBS Sees AI Spending Hitting $1.5 Trillion by 2028, Picks Meta and Amazon as Winners」 finance.biggo.com
- CH 02BigGo Finance「BNP Paribas Sees Microsoft AI Stack Paying Off, Keeps $632 Target」 finance.biggo.com
- CH 02Marcus on AI | Substack「Breaking: Anthropic stock is crashing, and AI stocks have lost over 50% in 5 months.」 garymarcus.substack.com
- CH 02Fortune「AI stocks are selling off after Anthropic’s record IPO, and Wall Street is split on whether this is the top」 fortune.com
- CH 03WSJ「Anthropic Mathematician Says AI Has Solved the Most Famous Problem in Math」 wsj.com
- CH 03New Scientist「Anthropic mathematician claims AI-assisted proof of the Riemann hypothesis」 newscientist.com
- CH 03Forbes「OpenAI Says Its AI Models Improved AMD's Flagship GPU Design By Up To 14%」 forbes.com
- CH 0436Kr「All DeepSeek-Style AI Players Are Exiting the Market: Are Usage Quotas Not the End of Free AI?」 eu.36kr.com
- CH 04Dealroom「Hengdian fund gathers $247M from Zhejiang firms for DeepSeek stake」 app.dealroom.co
- CH 04finance.biggo.com「Former DeepSeek Core Member Zhang Bo Open-Sources Personal Agent That Runs a 125B Model Locally on a 12GB GPU With No Token Fees」 finance.biggo.com
- CH 05Oncodaily「AstraZeneca and Tempus AI Renew $200M Oncology Deal」 oncodaily.com
- CH 05Foreign Policy Journal「Novo Nordisk (NYSE: NVO) And AI Firm nference Launch Joint Venture Framework To Tackle Obesity, Diabetes And Heart Disease」 foreignpolicyjournal.com
- CH 05TradingView「Tempus AI CEO, Data Sold Shares Worth Over $645K」 tradingview.com
- CH 06매일경제「製薬工場に新しい設備を導入する前に、人工知能(AI)が試験手続きを読み、デジタルツインで先に作動させてみる方式が導入されている。 ノボノディスクは、実際の生産設備を利用した検証に先立ち、仮想環境で試験..」 mk.co.kr
- CH 06매일경제「人工知能(AI)を活用した製薬·バイオ工場の自動化がはやくなっているが、人が消えた「完全無人工場」に直ちに切り替えることは難しいという診断が出た。 規制が厳しい製薬産業では、AI導入自体より設備と工程..」 mk.co.kr
- CH 07Bergen Record「Muslim Girl founder sues xAI over sexualized Grok edits」 northjersey.com
- CH 07Investing.com「Musk’s SpaceXSI denies liability in UK lawmaker’s lawsuit over Grok images By Reuters」 investing.com
- CH 07PYMNTS.com「New York Creates AI Enforcement Unit as States Step Up Oversight」 pymnts.com
- CH 07Anadolu Ajansı「UK parliamentary committee calls for new AI law over human rights risks」 aa.com.tr
Articles used
Today's related reports
- AI Daily News November 21, 2026
- AI & Economy News November 21, 2026
- AI & Finance News November 21, 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.