Microsoft reported fiscal Q4 revenue of $90.1 billion, up 18% year-on-year. Azure grew 43%, crossing $100 billion in annual revenue for the first time. Profit rose 31% to $35.8 billion. Microsoft 365 Copilot surpassed 30 million paid seats. Meta reported revenue up 28% to $60.8 billion. And free cash flow fell 91%, from $8.55 billion to $784 million, as capital expenditure on AI infrastructure rose 83% to $31.08 billion in a single quarter.

Two tech giants navigating the same era of heavy AI investment, yet yielding wildly different financial results.

听

Why Microsoft鈥檚 Numbers Work

听

The real strength of Microsoft鈥檚 AI strategy comes down to anchoring on software people already pay for.

Azure provides the essential foundation. AI workloads are driving cloud migration and consumption, with AI services contributing an estimated 16 percentage points of Azure鈥檚 growth in recent quarters. Copilot is a direct monetisation play inside Microsoft 365, a product with enormous enterprise reach. When a company upgrades to Copilot, Microsoft gets paid more for something it was already selling. The AI investment flows through existing, high-margin channels into incremental revenue that shows up directly in earnings. The Intelligent Cloud segment, which houses Azure, grew 32% year-on-year to $39.3 billion.

AI isn鈥檛 a cost centre here, AI actively drives profit for Microsoft. Setting aside $190 billion for 2026 capital spending feels justified while Azure scales at 43% and Copilot commands 30 million paid seats.

听

Why Meta鈥檚 Numbers Don鈥檛 Work

听

Meta鈥檚 situation is structurally different, and that changes how the numbers read.

Revenue was up 28%, driven by AI improvements to ad targeting and content recommendations. Those improvements are tangible and they鈥檙e working. But they鈥檙e incremental gains on a mature advertising business, not new high-margin revenue streams. Meta hasn鈥檛 yet launched a widely adopted, directly paid AI product comparable to Copilot or Azure AI services. The AI investment is improving what already exists rather than creating something customers pay for separately.

The essential metric here remains the cash flow figure. Free cash flow of $784 million on $60.8 billion in revenue translates to a margin of barely 1%. The prior year quarter produced $8.55 billion in free cash flow. That collapse reflects $31.08 billion in quarterly capital expenditure, an 83% year-on-year increase, for AI infrastructure whose returns are still largely theoretical.

Meta鈥檚 2026 capex guidance sits at $115 to $135 billion for the full year. The payoff, when it arrives, may be large. Right now it isn鈥檛 arriving.

听

What This Tells Founders And Operators

听

The Microsoft-Meta divergence is the clearest picture yet of what good AI investment looks like versus what AI spending as a long-term bet looks like.

Microsoft鈥檚 approach, embedding AI into products that already have paying customers and clear monetisation paths, is generating returns that accrete to earnings and cash flow today. Meta鈥檚 approach, heavy infrastructure investment for products whose commercial form isn鈥檛 yet clear, is consuming cash at an extraordinary rate. Without Meta鈥檚 balance sheet, it wouldn鈥檛 be sustainable.

For founders weighing up AI spending in their own businesses, the Microsoft model is the more useful perspective. AI that improves a product people already pay for, at a price point that reflects the improvement, generates returns. AI that builds toward a future product without a current monetisation path generates costs. Both strategies make complete sense assuming one has the scale and financial backing. Most companies only ever have the breathing room to try one.

Meta may prove its bet right. The foundations it鈥檚 building could underpin products that generate returns at scale within a few years. But the 91% free cash flow drop is a reminder that big bets on AI require extraordinary balance sheets to absorb the cost of being early. For the vast majority of businesses, the lesson from this earnings season isn鈥檛 to spend like Meta. It鈥檚 to build like Microsoft.

The post How Microsoft Proved AI Returns While Meta鈥檚 Cash Flow Collapsed appeared first on 91探花.

This strategy is catching on elsewhere. ActID launched in Shenzhen in March 2026 and already has around 300 users who have agreed to license their likenesses, with roughly ten faces used in two AI drama productions so far. A competing platform, New Claw, operates out of Chengdu with a minimum price of 500 yuan, around $74, per image.

Fees range from $15 to $700 per arrangement depending on the face type, the permitted uses and the number of productions. China鈥檚 AI microdrama sector reached roughly 16.8 billion yuan in 2025, and more than 10,000 AI-generated microdramas have been released each month since early 2026. The market is large, it鈥檚 growing and it needs faces.

听

The Reality Of The Licensing Terms

听

Both services sell themselves as clear, legal alternatives to the rampant face swapping that has led to thousands of court disputes in China. The Guangzhou Internet Court alone has handled around 700 AI-related face-theft cases over the past three years. The platforms make a fair point, suggesting a licensed marketplace with clear terms improves on companies scraping public images without permission.

The real catch comes down to what those signed terms permit. Licensing contracts on these platforms have been reported as vague about duration, modification, reuse and who may ultimately access the likeness. Once facial data enters an AI system, controlling how that likeness is altered, reused or portrayed in the future becomes trickier. As one Beijing lawyer put it, people may lose long-term control of their biometric identity once their images enter these systems. ActID monitors the internet for unauthorised use and connects users with lawyers if needed. It can鈥檛 guarantee licensed photos won鈥檛 be collected illegally, used in unauthorised face swaps or included in future AI training data.

Saying yes to a contract is simple enough. Maintaining any real power over your digital face is where things fall apart. Signing an agreement satisfies the lawyers today, leaving you with little power over what comes next.

听

Digital Labour Or Digital Exploitation?

听

The core issue boils down to whether this counts as legitimate digital work or a new level of exploitation. Advocates for this setup claim that it鈥檚 simply superior to the present state of affairs: people are getting paid for something that was previously taken from them for free. Traditional drama production budgets are shrinking, luxury brands are shifting to AI-generated content, and the demand for human likenesses is there regardless of whether the marketplace exists. Better to have an organised system with pricing and legal agreements than a Wild West of unauthorised scraping.

On the other hand, sceptics draw attention to the uneven dynamic at play. The people selling tend to be non-specialists with limited legal literacy, signing contracts written by platforms with significant informational advantages. The people buying are producers and technology companies with legal teams. The pricing, particularly at the $15 end, reflects less a fair market rate for a biometric asset and more the desperation of people who don鈥檛 know what they鈥檙e giving away.

One retired person鈥檚 face appearing in an AI thriller at $15 an episode is a transaction. That same face appearing in an AI advertising campaign for a brand they find objectionable, under a clause they didn鈥檛 notice, is something else.

听

Will This Model Travel West?

听

From 2 August 2026, the EU AI Act will require companies to explicitly inform people whenever they are looking at realistic AI-generated clones or manipulated media. And the UK introduced a new offence in February 2026 criminalising the creation of non-consensual sexual or intimate images of adults, covering AI-generated deepfakes. Broader online safety reforms are being fast-tracked.

A Western equivalent of the ActID model would face substantially stricter legal scrutiny. Vague licensing terms that might pass in China鈥檚 current enforcement environment would face direct challenge under EU data protection law, GDPR鈥檚 biometric data provisions and the AI Act鈥檚 transparency requirements. The model, a marketplace where people monetise their likeness for AI production, isn鈥檛 inherently unlawful in Europe. The vague contracts and the loss-of-control problem very likely would be.

What the Chinese platforms have created is a preview of a market that will arrive in some form elsewhere. The demand for licensed human likenesses in AI content production is real and growing. The debate is whether Western legal policies will force that market to operate with the consent and transparency protections that the current Chinese version lacks. That鈥檚 a more interesting question than whether you鈥檇 rent your face for $15. The answer to that one is probably: not once you鈥檝e read the contract.

The post Would You Rent Your Face To AI For $15 An Episode? appeared first on 91探花.

Today that patience gets tested on four fronts simultaneously. Microsoft, Meta, Qualcomm and ARM all report after the US market closes. Combined, these four companies account for a large share of the AI compute investment cycle. Microsoft is the dominant enterprise AI platform, while Meta is the highest-spending AI-native social company. Qualcomm is the test of whether edge AI is generating real commercial traction. ARM is the CPU architecture underneath almost every major AI operation.

The question running through all four is the same: have the billions paid off?

听

Microsoft: The Number That Matters Most

听

Microsoft鈥檚 Azure growth rate is the most watched figure in today鈥檚 prints.

The company鈥檚 AI business had already surpassed a $37 billion annual revenue run rate, and Azure demand was reportedly exceeding available capacity. A Morgan Stanley survey of chief information officers in the second quarter of 2026 found 62% planned to increase Azure spending and 65% planned to increase Microsoft 365 and Copilot spending. That positions Microsoft as the leading vendor for incremental generative AI budgets across enterprise.

What investors want to know today is whether AI is still driving net new growth in Azure or whether it鈥檚 primarily shifting existing workloads within the platform. Commentary on capacity constraints, data centre build-out timelines and the gross margin impact of AI-related depreciation will all be watched as closely as the headline revenue figures.

Microsoft has guided toward roughly $190 billion in capital expenditure for calendar year 2026. Any update on the pace of that spend, or on when it expects AI infrastructure costs to start converting more directly into margin, will move the stock.

听

Meta: The Harder Sell

听

Meta faces a unique form of investor scrutiny heading into today鈥檚 earnings report.

When Alphabet raised its AI capex outlook last week, shares fell. When Meta raised its own spending guidance earlier this year, shares fell sharply too. The pattern is consistent: the market views Microsoft, Alphabet and Amazon as better placed to monetise AI directly through cloud and enterprise services. Meta鈥檚 monetisation path is more indirect, running primarily through advertising effectiveness and engagement improvements that are harder to attribute specifically to AI spending.

Meta鈥檚 2026 capital expenditure guidance sits at $115 to $135 billion, with explicit commentary on accelerating capacity additions into 2027. The company has also floated the possibility of external sales of excess compute, which would launch an entirely new path for direct revenue from compute assets. Today鈥檚 report will be watched for any concrete progress on that front, alongside specific figures on how AI is affecting ad pricing, user engagement and newer revenue lines including business messaging and AI agents.

The 鈥渄ouble burn鈥� concern, heavy AI capacity spending alongside continued Reality Labs losses, will be the analyst community鈥檚 sharpest line of questioning.

听

Qualcomm And ARM: The Edge Of The Argument

听

Qualcomm and ARM reveal a different dimension of this sector鈥檚 health.

If AI investment is truly spreading across the market, edge hardware and CPU licensing should reflect that alongside hyper-scaler data centres. Qualcomm鈥檚 results will show whether AI features in Snapdragon chips are translating into higher average selling prices or genuine handset upgrade cycles, or whether on-device AI remains a marketing feature rather than a purchasing driver.

ARM鈥檚 position is arguably the most interesting of the four. Its architecture now exists at the core of AI infrastructure across every major hyperscaler: Google Axion, Microsoft Cobalt, Amazon Trainium and Graviton, Nvidia鈥檚 Vera CPU and Meta鈥檚 future AI infrastructure all run on ARM designs. The company launched its own ARM AGI CPU for AI data centres with Meta as the lead partner and has reported more than $2 billion in customer demand for fiscal years 2027 and 2028, more than double its initial expectations. Management is guiding toward roughly 20% revenue growth for the current quarter.

Assessing that momentum, along with tracking whether CPU demand for advanced models is expanding or levelling off, will clarify where value is genuinely accumulating across the technology space.

听

The Question All Four Must Answer

听

The surrounding market backdrop plays an important role here. Combined hyperscaler capital expenditure across the major players is tracking toward $745 to $775 billion for 2026, up roughly 55 to 60% from the prior twelve-month level. The original rationale justifying that spending was that AI infrastructure would compound into dominant market positions and defensible revenue streams. That rationale hasn鈥檛 been disproved, but it鈥檚 under more scrutiny than at any point in the past two years.

What today鈥檚 results will reveal, collectively, is whether the companies closest to enterprise AI adoption are seeing the returns materially enough to justify continued acceleration. Microsoft鈥檚 Azure numbers are the best proxy for that question at scale. Meta鈥檚 guidance will show whether the ad-led AI story can survive a higher-scrutiny environment. Qualcomm and ARM will confirm or complicate the idea that AI鈥檚 commercial impact extends clearly beyond the data centre.

The market has already indicated it鈥檚 changed the rules. Results that would have been celebrated a year ago are now measured against a different standard: not just 鈥渁re you spending on AI鈥� but 鈥渨hat are you getting for it.鈥� Today we find out whether the answers are good enough.

The post Big Tech鈥檚 AI Reckoning Arrives Today 鈥� Have The Billions Paid Off? appeared first on 91探花.

This is a standard experience for many. A 2026 University of Florida study tested five commercial AI detectors and found false positive rates ranging from 0.05% to 68.6%. Research from Stanford found that non-native English speakers face substantially higher false positive rates across major platforms. A PNAS Nexus study found more than half of all TOEFL essays were wrongly flagged as AI-generated across every detector tested.

These platforms regularly produce untrustworthy findings, and the companies selling them know it.

听

The Conflict Of Interest Built Into The Product

听

What makes this scenario so telling is the sequence that follows the flag itself.

Many of the companies selling AI detection tools also sell 鈥渉umanisation鈥� products, sometimes called AI removers or rewriters, designed to make flagged text score as human-written. The conflict of interest is clear, and built into the product architecture. A detector that flags aggressively creates more anxiety, more urgency and more conversions for the paid tool sitting one click away.

Turnitin has publicly claimed a false positive rate of under 1%. Independent testing suggests otherwise, with reported rates as high as 50% in some samples. The UF study found the range across commercial detectors so wide as to make the tools functionally unreliable for anything beyond a rough indication. One detector in the study produced a false negative rate of 99.6%. Another produced a false positive rate of 68.6%. These aren鈥檛 calibration issues; they鈥檙e numbers that would disqualify a tool from use in any other evidential context.

The question this raises is whether the detection business model is designed to fail in order to sell the fix. We asked a range of voices across education, technology and AI ethics to weigh in.

听

Our Experts

听

听

• Martin Harris, Head of Digital, Tank
• Dr. Morissa Schwartz, AI Evaluator, Educator and Founder, GenZ Publishing
• Michelle Edge, Partner, Eleven Hundred Agency
• Harpal Singh, AI SEO and GEO Consultant and Founder, Blimpp

听

听

Martin Harris, Head of Digital, Tank

听

听

听

鈥淔ifteen years of assessing marketing tech has taught me to start with one question: who profits from this tool鈥檚 verdict? And with AI detectors, the answer is uncomfortable. The same company that flags your writing will happily sell you the fix. If your humaniser revenue depends on people getting flagged, why would you ever work hard to reduce false positives? You don鈥檛 need a conspiracy theory here. It鈥檚 just a business model, doing what business models do.

鈥淭here鈥檚 a more basic problem underneath. These tools don鈥檛 actually detect AI. They detect predictable writing. Clear, well-structured prose scores badly on them, which is absurd when you think about it, and it鈥檚 why Stanford researchers found the majority of essays by non-native English speakers were falsely flagged. The detectors are also losing the arms race, because language models improve faster than the tools chasing them. When universities like Yale quietly dropped these detectors, that wasn鈥檛 caution. It was an admission that the evidence was never there.

鈥淭he bit I find strange is the misplaced anxiety. Brands are fretting over whether their content reads as AI while ignoring the question that actually affects revenue in 2026: can ChatGPT, Perplexity and Google鈥檚 AI Overviews find your content, understand it and cite it? A detection score has never made anyone a penny. Don鈥檛 use these tools as evidence for decisions about people. And whatever you do, don鈥檛 pay the company that flagged you to make the flag go away.鈥�

听

Dr. Morissa Schwartz, AI Evaluator, Educator and Founder, GenZ Publishing

听

听

听

鈥淎I detectors should be treated as weak screening signals, not verdicts. They estimate statistical patterns in text; they do not prove authorship. Polished human prose, formulaic academic writing, heavily edited copy and work by multilingual writers can all be falsely flagged.

鈥淎 company that sells both detection and humanisation has an obvious conflict-of-interest risk, but that structure alone doesn鈥檛 prove intentional over-flagging. The right questions are whether the company publishes its thresholds, false-positive rates, validation datasets and commercial incentives, and whether those claims have been independently audited.

鈥淚n my work evaluating AI-generated and human-authored writing, the reliable approach is evidence triangulation: review version history, drafts, citations, source notes and the writer鈥檚 ability to explain their reasoning. A student or employee should never be penalised based on one detector score. If a vendor markets detection as certainty while also selling the cure, its product design deserves scrutiny.鈥�

听

Michelle Edge, Partner, Eleven Hundred Agency

听

听

听

鈥淚t鈥檚 certainly fair to question the motivations behind these tools. It鈥檚 a bit like a mechanic that diagnoses a fault with your car and immediately offers to fix it. It doesn鈥檛 necessarily mean the diagnosis is wrong, but it does mean a certain amount of scepticism is warranted.

鈥淭he bigger issue is that the jury is still out on whether these tools are even accurate. If you take a piece of well-edited human writing and put it through a few AI detectors, there鈥檚 a good chance at least one will flag it. Good human writing often shares some of the patterns these tools associate with AI-generated text, such as clear, consistent structure and precise grammar.

鈥淏ut if writers start to worry that clarity and polish will get them flagged as machines, we risk incentivising uneven or unedited writing just to pass the test. As AI keeps evolving, hunting for the right list of tells is a losing battle. We should be looking for ways to identify original thinking and real human insight instead.鈥�

听

Harpal Singh, AI SEO and GEO Consultant and Founder, Blimpp

听

听

听

鈥淚t is not that all AI detectors intentionally exaggerate scores. It is most concerning that the business models are based on anxiety over issues like a lack of accuracy. When faced with the problem, users are offered a paid 鈥榟umaniser鈥� option with a false definitive warning by the AI-detection systems. Urgency leads to a problem-based design; false positives create urgency. While there may not be intentional malice, the design, warning systems and scoring systems are most likely unreasonably aggressive.

鈥淎I detection systems should never be viewed as proof of violation or authorship. As a minimum, all significant cases must be reviewed and the writer engaged personally. An analysis of version history and notes must be included.

鈥淔or detection systems, the initial question is not 鈥榟ow accurate is your detection system?鈥� It must also include: 鈥榙o you conduct independent evaluations of false positive rates and disclose the results?'鈥�

听

For any questions, comments or features,听please contact us directly.

听

The post Do AI Detectors Have A Commercial Incentive To Flag Your Writing? appeared first on 91探花.

According to the European Commission, the EU is actively studying a purchasing programme for permanent carbon removals, with the goal of creating predictable demand that would allow startups and investors to finance large-scale direct air capture and bio-based storage projects. The Commission has also published the first EU-wide certification rules for permanent carbon removals, covering DACCS, BioCCS and biochar. That makes those credits legible to regulators and buyers in a way they previously weren鈥檛.

听

The Power Of Guaranteed Demand

听

The core problem with scaling carbon removal has never been the technology 鈥� it鈥檚 been the financing.

Direct air capture plants are capital-intensive, energy-intensive and slow to build. A project that takes four years to construct and fifteen years to pay back requires lenders who believe the demand will still be there. In voluntary markets, that confidence is hard to establish. Reframing state purchasing as infrastructure procurement reshapes project risk enough to finally get institutional capital flowing.

The European Parliament鈥檚 own look at direct air capture is direct about this point: the technology needs clear policy, financial incentives, streamlined regulation and sustained research and development, alongside cheap clean power and access to geological storage sites. A procurement programme addresses the first two directly. While this approach leaves energy and logistical challenges intact, it lowers commercial risk for investors to back projects previously deemed premature.

That鈥檚 the hypothesis the European Commission is working from: guaranteed demand may not make carbon removal cheap, but it can make it financeable. For the sector, financeable is the prerequisite for everything else.

听

听

Which Companies Are Best Placed?

听

The European Union鈥檚 certification model narrows the field in practical ways. DACCS (direct air capture with carbon storage), BioCCS (bioenergy with carbon capture and storage) and biochar have the clearest regulatory pathway under the Carbon Removal Certification model. Companies whose projects align with those methodologies and who can demonstrate verified, measurable and durable removal are the ones most likely to access institutional procurement.

Climeworks, the Swiss direct air capture company, is already repositioning itself around compliance-oriented markets. It has expanded its advisory and portfolio-building services specifically around the CRCF, CORSIA and Article 6.2 frameworks. That indicates where the market is heading. Intermediated, standards-driven offtake agreements are replacing spot-market credit sales.

The players best placed already have verified monitoring and reporting, access to geological storage and project pipelines large enough to meet institutional buyers鈥� scale requirements.

听

Is Procurement The Right Mechanism?

听

The European Commission鈥檚 case for procurement rests on a specific market failure: the demand gap.

Private buyers aren鈥檛 buying enough to prove that permanent removal projects can attract buyers at commercial volume. Public procurement creates a reference customer, establishes price discovery and gives lenders a bankable offtake. That logic is sound for early-stage industrial technology with high capital costs and long payback periods.

The risk is technology lock-in. If the EU commits large procurement budgets to specific pathways before the technology has matured, it could entrench approaches that turn out to be suboptimal or crowd out cheaper options that emerge later. The strongest case for the programme is therefore a staged approach: targeted procurement now to build project pipelines and demonstrate commercial viability, alongside continued certification development and periodic review of which technologies qualify.

What the EU is doing is repositioning carbon removal from a voluntary climate add-on into a regulated procurement category tied to European industrial policy. The rules of the game are shifting across the carbon removal space. Corporate sustainability budgets are taking a back seat to sovereign demand, rigorous certification standards and state procurement timelines. Founders who understand that pivot are those building for what the market is actually becoming.

The post Why Is The EU Pushing To Become The World鈥檚 Biggest Carbon Buyer? appeared first on 91探花.

Admitting that detectors simply don鈥檛 work is a great start, but it鈥檚 still only the first step on a much longer road. The real puzzle is how universities are reacting now that the enforcement model has fallen apart. Are institutions genuinely rethinking academic integrity, or are they just turning off a faulty scanner and praying the problem magically vanishes?

听

The Broken Business Of Algorithmic Accusations

听

The research is fairly damning. Studies have shown that AI detection tools produce high rates of both false positives and false negatives, with some tools flagging human-written work as AI-generated at rates that would be professionally unacceptable if used in any other evidential context.

The Waterloo case is a pointed example: if a detector flags entirely human-written work, it cannot function as the basis for a misconduct finding. Using it anyway would expose institutions to legal challenges and, more importantly, damage reputations of students who hadn鈥檛 done anything wrong.

Universities are also waking up to the commercial forces at play here, even if few academic leaders are willing to voice those concerns publicly. AI detection tools are sold by the same companies that sell plagiarism detection software, and the incentive structure doesn鈥檛 obviously point toward accuracy. A tool that flags aggressively generates more alerts, more reviews and, in subscription models, more institutional dependency.

The degree to which financial incentives drove product design is a separate issue, one that certainly warrants its own deep dive.

听

Step One Versus The Actual Problem

听

Ditching the scanner is little more than a temporary dodge. It simply axes a flawed utility without providing any real replacement. The institutions doing this aren鈥檛 necessarily wrong to remove the tools. Removing them doesn鈥檛 answer the question that drove their adoption: what does academic integrity look like when AI is part of every student鈥檚 workflow?

Some universities are moving toward process-based assessment: oral examinations, draft submission, revision history, in-class work and portfolio approaches that make the final submitted document less central. These approaches treat AI as part of the workflow rather than trying to detect whether it was used in a specific submission. They鈥檙e trickier to administer, but they assess learning instead of trying to infer process from output.

Others are moving toward disclosure processes, requiring students to declare what AI tools they used and how. This adjusts the integrity question from 鈥榙id you use AI?鈥� to 鈥榳ere you honest about using AI?鈥� That鈥檚 at least a question universities can evaluate without relying on software that flags human writing as machine-generated.

听

The Philosophical Split In Modern Assessment

听

The core logic behind detection systems relied on a view that is quickly losing credibility: that AI use in academic work is cheating.

That logic made sense when AI tools were crude, occasional and easily distinguishable. It makes less sense when the same tools students are told not to use in coursework are being actively promoted for use in employment and research. The same institutions telling students not to use AI for assignments are promoting it for professional development.

A real philosophical shift means higher education institutions must clarify what skills they actually evaluate, redesign assignments around those goals and treat the question of artificial intelligence usage as secondary to genuine student learning. That鈥檚 a bigger redesign than removing a detector. It requires academic departments to agree on what a degree is certifying, which is a more intricate conversation than pulling a Turnitin plug-in.

What most universities are doing right now is somewhere between tactical retreat and a genuine rethink. The scanning software is gone, yet the assignment deadline persists. Students still rely on artificial intelligence to generate their drafts, and universities lack both the technical ability to notice and the willingness to act. There鈥檚 no integrity framework here. It鈥檚 a policy void, and removing the detector is what makes that visible.

The post Are AI Detectors Doing More Harm Than Good In Universities? appeared first on 91探花.

The wider coalition that signed the letter, including Nvidia, Microsoft, Meta and Hugging Face, made the case against premature restrictions. The argument is simple: restricting open-weight models narrows participation, concentrates power in a small number of frontier labs and makes AI governance harder for everyone outside that group. That鈥檚 OpenAI鈥檚 public position, though seemingly not its private one.

听

The Fine Line Between National Security And Market Protection

听

Lawmakers in Washington have trained their attention specifically on Chinese open-weight artificial intelligence. The stated rationale is that accessible Chinese systems endanger national defence, justifying tight restrictions on their use across American infrastructure. Independent researchers from diverse political backgrounds validate that security risk. The complication is that 鈥渞estrict Chinese open-weight models鈥� and 鈥渞estrict open-weight models broadly鈥� aren鈥檛 the same policy, and the second tends to follow from the first.

OpenAI and Anthropic have strong commercial reasons to want the open-weight environment to be less permissive. Their core business model depends on users paying for API access to frontier models. A world where high-quality open-weight models are freely available and widely used is a world where the justification for that pricing is harder to maintain. The national security argument holds genuine merit while also advancing the commercial interests of the firms championing it.

听

China鈥檚 Countermove

听

Xi Jinping took the stage at WAIC 2026 to promise 5,000 artificial intelligence training placements, joint regional centres and tool access for developing economies. This wasn鈥檛 rhetorical positioning. China is bundling open software directly into its foreign aid strategy, positioning its software as a sensible discount alternative to expensive American software licensing.

The geopolitical outcome is easy to predict. If US policy narrows access to open-weight models, particularly Chinese ones, and US frontier labs remain expensive for cost-sensitive markets, the vacuum gets filled. China is moving to fill it with open access, technical cooperation and the implicit message that its AI infrastructure comes without the geopolitical strings attached to US alternatives. That鈥檚 a stronger offer to a finance ministry in Lagos or a university in Jakarta than a $20-per-million-token API rate.

Chinese AI models, including DeepSeek and Qwen, are already a lot cheaper than comparable US models. In several benchmarks they鈥檙e competitive on capability. The cost difference alone makes them attractive in markets where budget constraints dominate procurement decisions. Open-weight availability makes them usable without ongoing API dependency.

听

When Neutral Software Architecture Turns Political

Founders who built product assumptions around open AI being stable, neutral and globally accessible are now sitting in the middle of a policy fight they didn鈥檛 cause and can鈥檛 control. The promise of open-weight models was that access would be technically determined: good models, free to run and no API dependency. That premise is now under pressure.

If US policy hardens against Chinese models, implementing or building on Qwen or DeepSeek becomes complicated in ways that weren鈥檛 previously the case. If US frontier labs reduce pressure on open-weight availability through lobbying, the free alternative to expensive US APIs shrinks too. The middle ground, truly open and neutral AI, is what independent builders need most. It鈥檚 also what both Washington lobbying and Chinese geopolitical strategy are working to eliminate.

The real battle isn鈥檛 open source versus closed source. It鈥檚 whether openness becomes a strategic export tool for China while US policy makers try to regulate it as a national security risk. The premise that model access, licensing and compute would depend strictly on technical merit instead of political strategy has proven unreliable for founders. Model infrastructure is part of the product risk stack in a way it wasn鈥檛 two years ago.

The post Why Are OpenAI and Anthropic Secretly Lobbying Against Open Source? appeared first on 91探花.

The New York Fed鈥檚 analysis of the 2025 Small Business Credit Survey found that businesses facing greater tariff challenges were less likely to expect higher revenues or employment in 2026. That鈥檚 hardly the profile of a business bracing for a short-term shock. It鈥檚 the profile of one adjusting expectations downward for the foreseeable future.

Exposure is concentrated across specific sectors yet broad in impact. Up to 80% of small retail and goods businesses integrated foreign supply chains in 2024, leading to widespread tariff friction for 55% of national goods firms and 67% of retailers the following year. Over 40% of small businesses described tariff-related costs as a burden, with retail and manufacturing hit hardest.

听

The Split Response: Pass Through, Absorb Or Both

听

The most consistent pattern in the available data is that most businesses aren鈥檛 choosing a single response. The New York Fed found that around 80% of affected goods and retail firms passed on at least some costs, around 60% absorbed some internally and many did both simultaneously, while others changed suppliers or adjusted purchase timing. The binary viewpoint of 鈥減rice increase or margin hit鈥� understates the reality. Most owners are doing multiple things at once, none of them clean and none of them fully adequate.

Jonathan Yee, co-founder of Mailers HQ, an ecommerce packaging company supplying custom bubble mailers to online sellers across the US, is in the absorption camp for now. 鈥淭ariffs have driven up our cost on the materials we import for our mailers, and right now we鈥檙e absorbing that hit rather than passing it straight through, accepting thinner margins to keep pricing stable for our customers, who are themselves small ecommerce sellers already squeezed on their own margins.鈥�

The logic is valid, albeit fragile. Yee鈥檚 customers are themselves margin-compressed. Passing a price increase downstream to a business that can鈥檛 absorb it either risks losing the customer entirely. So Yee takes the hit instead, banking on costs normalising before the absorbed margin becomes unsustainable. 鈥淗ow long we can keep absorbing it is the real question. If costs stay elevated, this stops looking like a short-term shock and starts looking like something we have to build into our pricing model for good.鈥�

听

When Does A Shock Become A Permanent Cost Structure?

听

That closing observation from Yee cuts to the core of the issue with more precision than any survey statistic. The businesses aren鈥檛 necessarily those facing the highest absolute tariff costs. They鈥檙e the ones whose customers are also small businesses with thin margins, creating a chain where no participant has enough pricing power to absorb the increase without consequence.

The New York Fed data supports the idea that this round is being treated differently from previous tariff episodes. Businesses with greater tariff exposure in 2025 were measurably more pessimistic about employment and revenue in 2026 than those without. This pessimism is showing up in hiring decisions. Businesses are delaying recruitment, holding positions open longer and in some cases cutting headcount to manage margin pressure that price increases haven鈥檛 fully offset.

The longer the elevated cost environment continues, the more these provisional responses, absorb now and review quarterly, harden into permanent operating model changes. A business that has repriced its products, reduced its headcount and switched suppliers to manage tariff costs has made structural changes that don鈥檛 reverse easily if the tariffs are reduced. The operational memory of the adjustment outlasts the policy that caused it.

听

What Owners Are Watching

听

The businesses holding up best share one of three characteristics: pricing power strong enough to pass through increases without losing customers, capital cushion enough to absorb margin compression for an extended period or supplier flexibility to source domestically or from lower-tariff jurisdictions. For many small businesses, none of those three conditions hold fully.

What owners are watching most intently is whether the current tariff structure becomes the baseline or reverts. This uncertainty is itself a cost: businesses aren鈥檛 investing, hiring or expanding while the operating cost picture is unclear. The damage from delayed decisions compounds over time even if the tariffs themselves are eventually reduced.

The New York Fed probably sums up the situation better than anyone else. This isn鈥檛 a situation where businesses break under tariff pressure. It鈥檚 a situation where businesses adjust their expectations quietly toward a more expensive and more constrained operating environment. Deciding if this shift is temporary or enduring is a reality most founders are hoping to postpone dealing with.

The post How Much Tariff Pain Can US Small Businesses Absorb Before They Break? appeared first on 91探花.

Compensation for public researchers drops at every level, capping critical discoveries at $10,000 rather than $30,000 and high-severity findings at $5,000. Medium and low-severity payouts fall to $2,000 and $250 respectively. At the same time, the company is introducing an exclusive VIP programme, directing premium rewards above historical limits to an invited circle of researchers.

GitHub frames the policy change around two specific goals. The internal security team intends to clear away low-value noise to concentrate on critical signals. The second goal is building a structure that serious security professionals find properly lucrative.

A flood of low-effort and AI-generated vulnerability reports has made it impossible to separate legitimate research from automated spam. As the cost of generating a report drops to near zero, the commercial rationale for treating every submission as genuine research disappears.

听

How Does GitHub Choose Its Security Elite?

听

Standard application forms won鈥檛 open this particular door. GitHub issues golden tickets based purely on historical performance, setting the bar at one confirmed critical vulnerability or roughly seven valid low-tier submissions. The company is also enabling HackerOne鈥檚 鈥渟ignal requirement,鈥� which limits how many reports new researchers can submit before they鈥檝e established a history of legitimate findings. Existing backlog reports will be assessed under the old payout rules.

Structurally, this policy moves the dynamic from an open marketplace of individual reports toward a semi-closed model governed by reputation. Independent researchers without an established track record face worse economics on high-effort work, particularly at the critical end where finding a novel vulnerability can take hundreds of hours. Newcomers are numerically capped on submissions until they demonstrate quality, which reduces the learning-by-doing pathway that many researchers used to build their reputations in the first place.

Analysts have raised concerns about a two-class researcher system. VIP selection criteria are controlled entirely by GitHub, transparency around invitation decisions is limited and there鈥檚 a risk that researchers game the system by holding back findings until VIP status is secured.
If the model proves financially effective for GitHub, other large platforms are likely to copy it. The broader public bounty market would shift toward lower public floors and gated premium access.

听

The Slow Death Of Open Digital Communities Under Infinite Volume

听

Bug hunters are hardly the only professionals currently watching machine-generated slop pollute their daily workflow.

Academic journals are tightening submission policies and requiring manual identity verification after AI-assisted manuscript factories overwhelmed peer review systems. Legal teams are adding verification layers to manage the volume of AI-drafted discovery documents and filings. Newsrooms are spending more resource on triage of AI-generated pitches than on actual reporting.

Each industry is running into the same wall. Spaces designed around trusted work from dedicated specialists are suddenly buried under cheap volume from users facing near-zero costs to submit content. The community鈥檚 quality-control systems, designed for human-scale input, break under machine-scale output. The response is almost always the same: invitation systems, reputation requirements or verified credentials. Communities that were open closed themselves, one gate at a time.

听

Why GitHub Specifically Matters

听

GitHub hosts a large share of the world鈥檚 open-source software supply chain. How it structures incentives for finding and reporting vulnerabilities has effects beyond its own products. If high-value vulnerability hunting becomes effectively invite-only at major platforms, the distribution of who discovers and discloses critical bugs over time could narrow. This policy influences talent diversity, the speed of bug disclosures and where priced-out researchers ultimately go.

The problem is an authentication gap that closing access doesn鈥檛 solve. Most submission systems were built on the assumption that the effort required to produce good work was itself a filter. Remove that effort barrier with AI tools and the filter disappears. Invite-only tiers replace the effort filter with a reputation filter, which works for researchers who already have reputations and makes it more challenging for new ones to develop them. The AI-slop problem gets managed, the access problem it creates is inherited.

The post How AI Slop Is Forcing GitHub To Close Its Doors appeared first on 91探花.

Francisco Partners, a tech-focused private equity firm, has committed $21 billion to the opposite view. Its latest fund closed in July 2026, exceeding its $18 billion target by $3 billion. Co-founder Dipanjan Deb stated their position clearly: 鈥淎I will not kill the software industry.鈥� The multi-billion dollar cheque speaks for itself. Institutional investors doing rigorous due diligence came to the same conclusion.

听

Where Investors Are Placing Their Bets

听

Francisco Partners aren鈥檛 treating software as entirely immune to AI disruption. The core argument rests on market overreaction, viewing indiscriminate sell-offs as a prime environment for strategic dealmaking. Deb鈥檚 argument is that AI will produce dispersion not destruction: some software companies will suffer permanent valuation damage, but others will use AI to run more efficiently, retain customers more effectively and expand into larger markets.

The logical flipside is that the most overvalued assets right now may not be incumbent software companies. They may be the AI-native startups built to replace them. Deb drew a comparison to the 2000 dot-com cycle, warning that investor enthusiasm around AI-first companies has the hallmarks of a bubble. If that鈥檚 right, the asymmetry sits on the other side of where most of the attention is.

The actual portfolio composition validates their logic far more convincingly than any corporate statement. Francisco Partners holds positions in Barracuda Networks and Jamf, both mission-critical software businesses where AI can improve operations without replacing the core value proposition.

That鈥檚 the model: buy cash-flowing software at depressed multiples, integrate AI to strengthen the product and improve margins and avoid paying a premium for AI promises that haven鈥檛 been demonstrated.

听

Where The 鈥淎I Kills SaaS鈥� Narrative Gets It Wrong

听

The anxiety driving the SaaS-pocalypse sell-off treated AI disruption as uniform. The assumption was that because AI can generate code, write content, automate workflows and answer questions, every software product that does any of those things is at risk. That line of thinking relies on a clear category error. The question for any given software business is whether AI can replicate the specific value it delivers, or whether AI makes that value more accessible and more defensible.

Software that operates at the centre of complex workflows, holds proprietary data, integrates deeply with other tools and would be painful to replace isn鈥檛 more vulnerable because AI exists. It may be less vulnerable, because AI tools are only as useful as the data and context they have access to. A business that has spent years accumulating that data and building those integrations has a stronger position in an AI world than it did before. The condition is moving quickly enough to use the capability instead of waiting to be disrupted by it.

The categories most at risk are those where the product is a thin layer on top of a process that AI can now handle end-to-end: simple content generation tools, basic automation, standalone chatbots, entry-level research tools. For SaaS founders in these categories, the anxiety is warranted. For those building products where workflow depth, data accumulation and integration complexity create real switching costs, the panic looks disproportionate to the actual threat.

听

What Founders Should Take From The Capital Shift

听

The Francisco Partners raise doesn鈥檛 mean every software business is safe. It means the market over-corrected. The businesses with the strongest structural positions got caught in a sell-off that didn鈥檛 distinguish between them and those with genuine AI exposure.

The cost and speed advantages offered by artificial intelligence are both genuine and rapidly compounding. The calculation facing every software founder is whether that momentum is eroding their defensibility or expanding it. If your product鈥檚 value comes from proprietary data, integration depth and workflow lock-in, AI makes those advantages more difficult to replicate. If your value comes from doing something that a well-prompted model can now do in seconds, the pressure is real and restructuring is probably necessary.

The $21 billion signal is that patient, data-driven capital has looked at the software market and concluded that the AI panic created a mispricing. The SaaS founders most likely to benefit from that conclusion are the ones who stopped panicking, figured out which side of the dispersion they sit on and got on with it.

The post SaaS Founders Are Panicking About AI But A $21 Billion Bet Says They鈥檙e Wrong appeared first on 91探花.