The day OpenAI joined Google's cloud── Compute shortage, a weakness found by AI, a drug discovery AI deal
Download the video (MP4, 9 MB)
The range of AI we can choose from has grown wide. Yet to me, it also looks as if we lean more and more on a few giants. Around pharma work, AI now turns up in research, in legal work and in hiring. So which view is right is not someone else's problem. The firms that deliver AI have their own pressures. Those pressures show in where they sell, and in which industries they enter. Today we follow them through several scenes. We ask what the receiving side should prepare. At the end, I give my answer on whom we take AI from, and on what terms.
01OpenAI arrives on Google Cloud
Source Google Cloud
Our first scene is about where you buy AI. For a user, that also decides where you get locked in.
| Item | Details |
|---|---|
| Where offered | Google Cloud Model Garden |
| Models side by side | Gemini, Claude, GPT |
| Also available | 200+ in-house, third-party and open models on one unified platform |
| How reports frame it | OpenAI leaves sole reliance on Microsoft for all 3 major clouds |
Google Cloud is the only hyperscaler offering all 3 (company statement)
Until now, using one developer's models meant running them on a partner's servers. Rival developers' models now sit side by side in one place. Users can swap and compare them without changing their base system. I think the contest is moving from model quality to who owns the shelf.
Only a handful of firms own those shelves, renting out servers worldwide. The price is keeping quality the same everywhere. If every shelf has the same models, the base is chosen on cost and ease of control.
The number on offer is far beyond what one person could test. Without selection criteria, more choice means confusion. I would set test tasks from our own work first.
For a drug maker, the key question is not which model is smartest. It is where patient data and development records sit, and under whose control. If each department picks its own model before policy catches up, no one can see where data flows.
So why did the developer need to spread beyond a single host? The reason seems to lie in pressures on the selling side.
02Not enough compute
Source Sources | Alex Heath ／ Business Insider ／ New York Post
The clue is that the power to run AI is itself scarce. I think spreading out was not only about winning customers. A seller short of supply tries to sell in as many places as it can.
| Person | Venue | What was said |
|---|---|---|
| Sundar Pichai (Google) | Podcast appearance | Discussed "Gemini 4 Argon" and the compute crunch |
| Elon Musk (SpaceX) | SpaceX earnings call | "Relentlessly pressed by every hyperscaler" to rent out compute |
| Meta | New York Post report | Speculation it will issue new stock to fund AI spending |
Meta's new stock issue is speculation, not confirmed
The head of a search giant and the head of a rocket company said the same thing. There are not enough machines to run AI, and firms are fighting over them. Another giant may even sell new shares to raise cash. That is not confirmed. What is certain is that very different leaders described the same shortage. A shortage for sellers becomes a worry for buyers.
This power grows only when buildings, electricity and chips all come together. None of them appear overnight. When demand jumps, waiting lines form.
For users, this is about price and priority. If the shortage lasts, big customers may lock in capacity first. Smaller ones may wait. If your research or document work relies on AI, it should keep running when one supplier stops. Before signing, check the fallback. Build price rises and delays into your plans.
And AI running on that power has started to find weak spots in our defenses. Next comes the scene about defense.
03AI finds a weakness in the web
If a powerful AI spots a human oversight before humans do, who fixes it? In the next scene, that question became real. And it reaches firms that do not even use AI.
The developer's AI found a large flaw in the internet's plumbing that attackers could use. It did not keep this to itself. It set up a way to line up repair work with partners worldwide. With AI, flaws may be found faster than they are fixed. Not everyone with such AI will act in good faith.
In good hands, a flaw is a clue for a fix. In bad hands, it becomes an attack. The same skill cuts both ways. So I weigh the steps after discovery as heavily as the discovery itself.
Drug makers hold tempting targets, like patient data and records of medicines in development. When a repair notice arrives, who receives it, and by when is it fixed? Without clear roles, attackers may use a flaw before defenders act. Even for rented systems, the contact point must sit inside the company.
This developer is now moving beyond defense, into the very work of making new medicines.
04OpenAI buys a drug discovery AI startup
Source Chosunbiz ／ Unite.AI ／ GeneOnline ／ Business Wire
A maker of AI is entering the process of making medicines. For pharma, this changes whether that firm is a tool vendor or a rival hunting the same candidates.
A developer that sold the same AI to every industry bought a young firm that hunts for new medicines. Reports say the aim is to fold research that runs itself into its biology toolkit. Until now, the developer only rented out tools. Now it has stepped into the core work of one industry.
| Deal | Reported amount | What the amount is |
|---|---|---|
| OpenAI's acquisition of Orchestra Bio | Not reported | Size unknown |
| Big pharma–startup miniprotein drug discovery partnership | Over $400 million | Deal valuation, not a confirmed total payment |
| Pharma back-office work | $400 billion | Market size, not one company's investment |
Amount types as judged by this site's "AI and the Pharmaceutical Industry"
The same day, several big figures on AI for medicines hit the headlines. But they mean very different things. One is the value a deal assumes. Another is the size of a whole sector. For the purchase itself, no price appears. I would first check whether a figure is an estimate or money actually paid.
For research teams, the question is whether to hand their lab data to this toolkit. Could your hard-won insight end up training the other side's tools? Check the data terms at the contract stage. The vendor may also become a rival.
So outside the lab, in the daily work of a company, how far has AI already reached?
05In the pharma workplace
Source Fierce Healthcare ／ Thomson Reuters Legal Solutions ／ 日刊薬業
The AI story does not end inside the lab. It is entering quieter jobs that matter to drug makers too.
Epic Connectors
Epic launched "Connectors", linking its own AI tools with third-party apps (Fierce Healthcare, headline only).
Researching 100+ jurisdictions
Thomson Reuters' Multinational Agent researches one legal question across 100+ jurisdictions, with verifiable sources.
AI conversations in hiring
A pharma industry group dropped entry sheets in new-graduate hiring and introduced AI conversations (Nikkan Yakugyo).
A firm that keeps medical records began connecting its assistant to other companies' software. In legal research, an assistant now checks rules that differ by country, and shows where each answer came from. In hiring, one company now gets to know students by talking with AI instead of reading paperwork. Until now, AI answered when asked. Now it owns a piece of the workflow.
Drug work is bound by rules that vary by country. Checking those differences has eaten a lot of staff time. With traceable answers, checking gets faster. Regulatory staff can spend the saved time on judgment.
For legal, regulatory and HR staff, AI is no longer distant. But there is a condition. People must be able to check where an answer came from. In hiring, trust needs a clear account of what shapes the AI's view. Who catches AI errors, and who is accountable? Decide that before you start.
Yet bringing AI into the workplace does not mean it takes root. The same day also showed the opposite.
06AI expands, AI retreats
Source Microsoft Source ／ Wccftech ／ BigGo Finance
That day, AI spread fast in one place and was pulled back in another. The gap reveals the situation of those on the receiving end.
| Move | Who and what | Source |
|---|---|---|
| Expands | Microsoft brings Copilot to 330,000+ VW employees | Microsoft Source |
| Shrinks | 8 Copilot features removed from Windows 11 File Explorer | Wccftech |
| Blocked | Muse, Dots, Doubao and other personal agents blocked by apps | BigGo Finance |
The Windows 11 removal reportedly followed user backlash
At one carmaker, a large share of staff will get an AI helper. Yet in the same developer's computer software, users complained, and AI functions were cut. I think features seen as forced on people get rejected, however good. And AI that acts for people keeps getting shut out by apps. In the end, users and platform owners decide whether AI spreads.
A company-wide rollout looks big as a number. But seats handed out and daily use are different things. I want to know which tasks got faster, and by how much.
Drug makers face the same thing. Handed down from the top, AI leaves staff unsure, and they drift away. Decide with users which tasks to hand over, and how far. Spread what works in one department to others. That beats one big rollout.
Only when AI takes root at work can we say we chose it. So what do today's scenes show when we put them together?
Whether OpenAI discloses the Orchestra Bio acquisition price and how the tools are offered after Rosalind integration (including pricing for drug makers).
Open the full transcript
Intro
The range of AI we can choose from has grown wide. Yet to me, it also looks as if we lean more and more on a few giants. Around pharma work, AI now turns up in research, in legal work and in hiring. So which view is right is not someone else's problem. The firms that deliver AI have their own pressures. Those pressures show in where they sell, and in which industries they enter. Today we follow them through several scenes. We ask what the receiving side should prepare. At the end, I give my answer on whom we take AI from, and on what terms.
CH 01 OpenAI arrives on Google Cloud
Our first scene is about where you buy AI. For a user, that also decides where you get locked in.Until now, using one developer's models meant running them on a partner's servers. Rival developers' models now sit side by side in one place. Users can swap and compare them without changing their base system. I think the contest is moving from model quality to who owns the shelf.Only a handful of firms own those shelves, renting out servers worldwide. The price is keeping quality the same everywhere. If every shelf has the same models, the base is chosen on cost and ease of control.The number on offer is far beyond what one person could test. Without selection criteria, more choice means confusion. I would set test tasks from our own work first.For a drug maker, the key question is not which model is smartest. It is where patient data and development records sit, and under whose control. If each department picks its own model before policy catches up, no one can see where data flows. So why did the developer need to spread beyond a single host?
The reason seems to lie in pressures on the selling side.
CH 02 Not enough compute
The clue is that the power to run AI is itself scarce. I think spreading out was not only about winning customers. A seller short of supply tries to sell in as many places as it can.The head of a search giant and the head of a rocket company said the same thing. There are not enough machines to run AI, and firms are fighting over them. Another giant may even sell new shares to raise cash. That is not confirmed. What is certain is that very different leaders described the same shortage. A shortage for sellers becomes a worry for buyers.This power grows only when buildings, electricity and chips all come together. None of them appear overnight. When demand jumps, waiting lines form.For users, this is about price and priority. If the shortage lasts, big customers may lock in capacity first. Smaller ones may wait. If your research or document work relies on AI, it should keep running when one supplier stops. Before signing, check the fallback. Build price rises and delays into your plans. And AI running on that power has started to find weak spots in our defenses. Next comes the scene about defense.
CH 03 AI finds a weakness in the web
If a powerful AI spots a human oversight before humans do, who fixes it?In the next scene, that question became real. And it reaches firms that do not even use AI.The developer's AI found a large flaw in the internet's plumbing that attackers could use. It did not keep this to itself. It set up a way to line up repair work with partners worldwide. With AI, flaws may be found faster than they are fixed. Not everyone with such AI will act in good faith.In good hands, a flaw is a clue for a fix. In bad hands, it becomes an attack. The same skill cuts both ways. So I weigh the steps after discovery as heavily as the discovery itself.Drug makers hold tempting targets, like patient data and records of medicines in development. When a repair notice arrives, who receives it, and by when is it fixed?
Without clear roles, attackers may use a flaw before defenders act. Even for rented systems, the contact point must sit inside the company. This developer is now moving beyond defense, into the very work of making new medicines.
CH 04 OpenAI buys a drug discovery AI startup
A maker of AI is entering the process of making medicines. For pharma, this changes whether that firm is a tool vendor or a rival hunting the same candidates.A developer that sold the same AI to every industry bought a young firm that hunts for new medicines. Reports say the aim is to fold research that runs itself into its biology toolkit. Until now, the developer only rented out tools. Now it has stepped into the core work of one industry.The same day, several big figures on AI for medicines hit the headlines. But they mean very different things. One is the value a deal assumes. Another is the size of a whole sector. For the purchase itself, no price appears. I would first check whether a figure is an estimate or money actually paid.For research teams, the question is whether to hand their lab data to this toolkit. Could your hard-won insight end up training the other side's tools?
Check the data terms at the contract stage. The vendor may also become a rival. So outside the lab, in the daily work of a company, how far has AI already reached?
CH 05 In the pharma workplace
The AI story does not end inside the lab. It is entering quieter jobs that matter to drug makers too.A firm that keeps medical records began connecting its assistant to other companies' software. In legal research, an assistant now checks rules that differ by country, and shows where each answer came from. In hiring, one company now gets to know students by talking with AI instead of reading paperwork. Until now, AI answered when asked. Now it owns a piece of the workflow.Drug work is bound by rules that vary by country. Checking those differences has eaten a lot of staff time. With traceable answers, checking gets faster. Regulatory staff can spend the saved time on judgment.For legal, regulatory and HR staff, AI is no longer distant. But there is a condition. People must be able to check where an answer came from. In hiring, trust needs a clear account of what shapes the AI's view. Who catches AI errors, and who is accountable?
Decide that before you start. Yet bringing AI into the workplace does not mean it takes root. The same day also showed the opposite.
CH 06 AI expands, AI retreats
That day, AI spread fast in one place and was pulled back in another. The gap reveals the situation of those on the receiving end.At one carmaker, a large share of staff will get an AI helper. Yet in the same developer's computer software, users complained, and AI functions were cut. I think features seen as forced on people get rejected, however good. And AI that acts for people keeps getting shut out by apps. In the end, users and platform owners decide whether AI spreads.A company-wide rollout looks big as a number. But seats handed out and daily use are different things. I want to know which tasks got faster, and by how much.Drug makers face the same thing. Handed down from the top, AI leaves staff unsure, and they drift away. Decide with users which tasks to hand over, and how far. Spread what works in one department to others. That beats one big rollout. Only when AI takes root at work can we say we chose it. So what do today's scenes show when we put them together?
Wrap-up
Our options have truly widened. But the power to run AI is limited. AI now finds serious flaws. And its makers are stepping into medicine. So deciding what to entrust, what to delegate, and who keeps watch is the receiving side's job. Being able to write those terms in your own words is what choice really means. Next, I will watch what terms that developer offers to drug makers.
- CH 01Google Cloud「Google Cloud expands model choice and flexibility, bringing OpenAI models to customers」 cloud.google.com
- CH 02Sources | Alex Heath「Sundar Pichai on Gemini 4 and the AI compute crunch」 sources.news
- CH 02Business Insider「AI Rivals Are Thirsty for SpaceX's Compute, Elon Musk Says」 businessinsider.com
- CH 02New York Post「Speculation — and anxiety — growing that Meta will issue new stock to pay for lavish AI spending」 nypost.com
- CH 03OpenAI「Patch the Planet: global response coordination as agents uncover critical vulnerabilities」 openai.com
- CH 03Forbes「OpenAI Says AI Found One Of The Biggest Web Weaknesses Ever」 forbes.com
- CH 04Chosunbiz「OpenAI acquires Orchestra Bio, accelerates AI-driven drug discovery - CHOSUNBIZ」 biz.chosun.com
- CH 04Unite.AI「OpenAI Acquires Orchestra Bio to Expand Rosalind Life Sciences Tools」 unite.ai
- CH 04GeneOnline「The $400B Pharma Backend Is Getting an AI Shakeup—and Hourly Billing Is Dead」 geneonline.com
- CH 04Business Wire「AI Proteins Announces Research Collaboration and License Agreement With Astellas Pharma to Discover and Develop Novel Miniprotein-Based Therapeutics Valued at Over $400 Million」 businesswire.com
- CH 05Fierce Healthcare「Epic launches Connectors to link its AI tools with third-party apps」 fiercehealthcare.com
- CH 05Thomson Reuters Legal Solutions「AI for cross-border legal research: the Multinational Agent」 legal.thomsonreuters.com
- CH 05日刊薬業「AI対話導入「自分の言葉で話して」 新卒採用でES廃止のシミックグループ | 日刊薬業 - 医薬品産業の総合情報サイト」 nk.jiho.jp
- CH 06Microsoft Source「Microsoft empowers more than 330,000 Volkswagen Group employees with Copilot」 news.microsoft.com
- CH 06Wccftech「Microsoft Scales Back Copilot In Windows 11, Removing Eight AI Actions From File Explorer Amid User Backlash」 wccftech.com
- CH 06BigGo Finance「Personal AI Agents Hit Platform Blockades: Meta's Muse, ByteDance's Doubao Bump Into App "No-Entry" Walls」 finance.biggo.com
Articles used
- AI Daily News 2026-10-22
- AI and the Pharmaceutical Industry 2026-10-22
- AI Daily News, Pharmacy Edition 2026-10-22 Morning
Today's related reports
- AI Daily News October 22, 2026
- AI & Economy News October 22, 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.