OpenAI needs cash fast. The company cut its new funding goal this week. It once wanted $30 billion. Now it wants $20 billion instead.

OpenAI's valuation goal fell too. The company first pitched investors on a $1.4 trillion value. Investors now hear $1.25 trillion. That mark dropped $150 billion in one month.

Some investors pulled back their planned checks. OpenAI has secured about $11 billion in commitments so far. That sum covers less than half its first $30 billion goal. OpenAI set Friday as the deadline for new commitments.

The cut lands during a rough week for tech stocks. The Nasdaq 100 fell about 10 percent over three weeks. One single Friday brought a 4.5 percent drop. Investors worry labs spend too much on AI data centers.

Anthropic adds to that worry this week. Anthropic is OpenAI's biggest rival in the AI race. Blackstone told investors that Anthropic may earn a junk credit rating after its planned stock listing. A junk rating raises the cost of borrowing money.

The finance writer Kobeissi Letter broke the news on X. The post pulled 1,476 likes within hours. It drew 114 replies too. The Financial Times first reported the funding cut. The Wall Street Journal and Techmeme both confirmed it fast.

A Hacker News post about the cut picked up 5 points. OpenAI's own Wikipedia page drew 7,782 views on Thursday. Readers already search the company's name as the story spreads.

The reaction says something plain about this stretch in AI. Labs raised billions on pure promise for years. Investors now want proof the math works. A trillion-dollar valuation needs real revenue behind it today.

OpenAI still plans to raise more money in the weeks ahead. The first close will likely land short of $20 billion. A later offering could fill the rest of the gap. Every AI lab now watches how investors treat this one raise.

S&P Global Ratings turned gloomy on Meta this week. The agency cut its outlook from stable to negative. Meta still holds its AA- rating for now. The warning centers on how much Meta spends on AI.

Meta posted negative free cash flow in the third quarter. The loss came to $2.12 billion. That marks the first time this happened at Meta. Free cash flow is money left after a company pays its bills and builds new gear.

Meta plans heavy spending ahead too. The company guides full-year capex between $135 billion and $145 billion. Capex means money spent on buildings, chips, and other long-term gear. S&P doubts Meta will slow that pace soon.

S&P also doubts Meta's AI bets will pay off fast. The agency says it watches to see when AI revenue will outpace AI costs. Meta disagreed with that view in public. A spokesperson called the view too harsh on Meta's AI plans.

The crypto commentator Coin Bureau posted the news on X. The post pulled 71 likes within hours. It drew 18 replies too. Coin Bureau called Meta one of the world's most profitable companies now facing new financial strain.

The timing matters past Meta alone. OpenAI cut its own funding goal by $10 billion the same week. Investors now question AI spending across every major company. A rating agency doubting Meta adds fresh weight to that mood.

Four giant companies drive most AI capital spending today. Meta, Microsoft, Google, and Amazon all build huge data centers. S&P flagged Meta first among that group. Rivals now face similar questions from their own lenders.

This moment marks a shift in how Wall Street treats AI bets. Years of cheap praise gave way to hard math this fall. A company can spend big on AI only as long as lenders keep faith in the payoff.

A photo about brain cells in AI swept through Reddit overnight. A user named EmotionalGuess9229 posted it to r/singularity. The post made a bold claim. Real human brain cells now work inside AI systems. Readers reacted fast. The thread drew 61 comments within hours.

The post taps into real science from this month. Cortical Labs builds computers from real human brain cells. The cells grow in a dish. The company calls its unit the CL1. Each CL1 links its cells to a chip with tiny wires.

Cortical Labs showed a new system called HIVE in late October. HIVE links 15 CL1 units into one team. Researchers tested the linked units on a symbol task. Seven linked units scored 99 percent on that test. One unit alone scored about 80 percent.

The team ran a second test too. They tried a virtual navigation task. One CL1 unit alone hit 40 percent of the top score. A normal digital AI model hit 32 percent. Many CL1 units working together hit about 70 percent.

Cortical Labs says living cells use far less power than normal chips. The company showed HIVE at a science meeting in Washington. The meeting covers brain cells grown outside a body. Those cells can still learn simple tasks.

Reddit users split on how they felt. Some called the science thrilling and strange at once. Others asked where the cells come from. Many brain cells in this field start as stem cells from donated blood.

The story spread fast online. It touches a line many readers care about. A computer built from living tissue still feels new to most people. Cortical Labs wants that reaction. Excitement drives funding and new customers. The next test will push HIVE past a simple lab game.

A LocalLLaMA user named Relative-Ant-9249 posted fresh speed tests this week. They run Qwen3.8-Flash-Next, a large open AI model, on two RTX 5090 graphics cards. Their post compared four different software engines side by side. Each engine runs the same model in its own way.

QEngine came out fastest in the test. QEngine is a small, open tool built by a coder named bpx03. The poster said QEngine hit 504 tokens a second in one mode. A token is a small chunk of text an AI model reads or writes.

QEngine's rivals trailed well behind in the same test. Strata, another engine, hit 214 tokens a second on one setting. A tool called bucko vLLM hit 135 tokens a second in its tuned setup. The gap between tools topped 1.5 times in the poster's own numbers.

The poster aimed QEngine at hardware past its first design. QEngine's own instructions list older RTX 3090 cards only. The poster changed one build setting to target newer RTX 5090 cards. The swap worked, and the engine started serving requests within a minute.

The post pulled 21 comments inside hours on r/LocalLLaMA. Readers there run AI models on their own machines at home. They swap notes on speed, memory use, and hardware tricks daily. A clear, numbers-first post like this draws quick, detailed replies.

QEngine itself stays small today. The project carries 19 stars on GitHub as of this week. That small count hides real value for a niche crowd. A tool this fast changes how many models a single home rig can run well.

The story shows where real AI progress often starts. Big labs chase headline benchmarks with huge budgets. Small coders chase raw speed on gear they already own. Both paths feed back into the same open AI tools everyone shares.

OpenAI, Google, Microsoft, and Anthropic made a joint promise this week. Each company agreed to honor one shared web signal. The signal tells AI crawlers to skip a site. Publishers can set it once. That beats dealing with four separate rules.

The plan centers on one line in a file called robots.txt. Robots.txt tells web crawlers which pages to skip. The new line acts as one switch for AI training alone. A site can still allow search crawling. It can still block AI training at the same time.

The idea started with an engineer named Kevin Kelley. Kelley works on standards at Anthropic. He pitched the shared signal through a public standards group. The four companies then agreed to honor the same code.

Rollout speed differs by company. Anthropic already honors the new signal today. OpenAI plans its own rollout within months. Google and Microsoft both target 2027. Meta voiced strong support for the plan too.

The announcement drew light attention online so far. A Techmeme post about the deal pulled 7 likes and 6 replies on X. A tracking account called AGTP Insights shared the same news late Friday night. Wider reaction may build once each company finishes its rollout.

The plan still runs on trust past any binding rule. Each company must choose to honor the signal every time. Older, narrow crawler blocks still work as a backup. A publisher can stack both tools for stronger coverage today.

The deal matters for one reason. It marks rare teamwork among AI rivals. OpenAI, Google, Microsoft, and Anthropic compete hard for users and computing power. Publishers pushed labs for a simpler opt-out for over a year. One shared signal now gives publishers a single lever instead of four.