David Steward’s World Wide Technology (WWT) and Robert F. Smith’s Vista Equity Partners face different versions of the same pressure: what can a large business do next that customers will pay for? WWT completed its acquisition of Softchoice in 2025, adding software, cloud, cybersecurity and AI services. Vista is working with its software portfolio as AI agents raise questions about enterprise products and pricing.

The figures often used to describe these businesses measure different things. Calendly’s $3 billion figure was its reported valuation after a 2021 financing round. WWT says annual revenue exceeds $20 billion. Vista reported $103 billion in assets under management (AUM) as of June 30, 2026. One is a reported valuation at a point in time, one is yearly sales, and one is investor capital under management.

Three Different Routes Beyond the First Breakthrough

Calendly is extending a scheduling product into the meeting itself. WWT has bought capabilities it can bring to customers alongside infrastructure work. Vista is preparing software companies for the changes AI could bring to their products and operations. Each strategy depends on proving value after the headline announcement.

Tope Awotona Wants Calendly to Own More of the Meeting

Beyond the Booking Link

Calendly’s invitation link did two jobs. It helped users find a time to meet, and it introduced the service to everyone who received one. When an existing user sent a link, a new person encountered Calendly as part of arranging a conversation.

TechCrunch reported in January 2021 that a $350 million funding round valued the company at more than $3 billion. It also reported that Calendly had 10 million monthly users at the time. That user figure describes the period around the funding round, not the current size of the service.

Now the company wants to be useful before and after a meeting. Callie can help sort out scheduling details in an email thread and answer questions about an event. Notetaker joins calls on Zoom, Google Meet and Microsoft Teams, then provides recordings, transcripts, summaries and action items. Calendly says recaps are private by default and users choose what to record and share. Callie entered beta in August 2026, alongside new plan tiers for the company’s AI tools. Initial access was in English.

Awotona explained the thinking behind the launch:

Meetings don’t exist in isolation. They’re part of a larger work cycle. — Tope Awotona

Calendly says beta customers had used Notetaker for more than 100,000 recordings. It also says users report saving three to four hours a week by reducing manual note-taking. Those figures come from the company’s launch announcement. They show early use, but not how many customers will pay for the tools or keep using them after a trial.

The product also asks customers to trust Calendly with more of their work. Meetings can include customer information, hiring discussions or confidential plans. The company’s AI data and privacy explanation describes what its features access and how users can choose whether to use them. Clear controls matter, but so does the customer’s judgment about whether the time saved is worth sharing meeting content with the service. Calendly will also have to compete with tools built into the calendar and video services people already use.

David Steward Is Expanding WWT’s Service Range

From Infrastructure to Integration

WWT helps organizations select, test and deploy technology. Its Advanced Technology Center gives customers a place to assess hardware and software before committing to large deployments. That can help when a business is weighing cloud capacity, data security or unfamiliar AI systems.

The company was founded by Steward and Jim Kavanaugh in 1990. It grew from technology distribution into a global solutions provider. Softchoice adds another layer to that business. WWT completed the acquisition on March 13, 2025, for approximately US$1.3 billion. Softchoice brought software and cloud expertise, consulting services, and customers in Canada’s commercial and small and medium-sized business markets. WWT said the combined businesses would offer a broader range of software, cloud, cybersecurity and AI services.

Steward described the intended reach in WWT’s completion announcement:

This combination will magnify our presence in global markets, across both enterprise and small and medium size businesses for all industries. — David Steward

WWT says annual revenue is more than $20 billion. With Softchoice, WWT can offer more software and consulting work alongside infrastructure deployment. That could put it in the running for a larger share of customers’ technology budgets, including ongoing cloud, security and AI support. WWT has to bring Softchoice’s expertise into its operations and keep customers satisfied during the transition.

Robert F. Smith Is Positioning Vista for Agentic AI

Enterprise Software Under Pressure to Adapt

Vista is a private investment firm focused on enterprise software. Its FAQ reports $103 billion in AUM and a portfolio of more than 85 software companies as of June 30, 2026. The figure describes capital Vista manages for investors. It is not Vista’s valuation, the combined market value of its portfolio or Smith’s personal wealth.

Vista’s AI strategy is to work with portfolio companies on their products and operations. “Agentic AI” refers to systems designed to carry out tasks within a workflow, rather than simply answer a prompt. Smith’s thesis is that established business software has context, customer relationships and permissions that can make AI more useful inside companies. Vista says its portfolio companies are exploring AI for product innovation and operating costs, though the work and results will vary from one business to another.

Smith has described the shift this way:

Software is evolving from a service into a worker. — Robert F. Smith

AI could make enterprise software more useful. It could also expose products with narrow functions or weak pricing power to new competition. In 2024, Reuters reported that Vista was in talks with Pluralsight’s lenders over the company’s debt after it struggled to meet its obligations. That episode does not predict the outcome of Vista’s AI strategy. It does show why a technology thesis cannot stand in for durable demand and sound financing.

Three Strategies for Ongoing Ownership

Calendly wants a place in more of the meeting cycle. WWT has completed a deal that broadens the services it can offer. Vista is directing capital and operational support toward AI’s effect on enterprise software. The work ahead differs: Calendly needs customers to pay for and keep using the new products; WWT has to integrate Softchoice and deliver the expanded services; Vista’s portfolio companies have to show that AI improves products or reduces costs.

For founders, these examples point to different ways ownership can create room to grow. Calendly’s link helped users bring the product to new customers. WWT built around institutional buyers and the infrastructure they depend on. Vista’s model gives it influence over software businesses and their operations. Each route calls for its own mix of capital, customer access and expertise.

The companies’ next results will be measured in ordinary business terms: renewals, delivery and cash flow. Calendly’s beta usage is an early sign, not proof of a durable market. WWT has closed its Softchoice deal, but integration takes time. Vista has set out an AI strategy; its portfolio companies still have to show what customers value and will pay for. The earlier milestones made expansion possible. They cannot guarantee it.

FAQs

What Is Tope Awotona Building at Calendly Now?

Calendly has added AI products around its scheduling service. Callie helps coordinate meetings and answer event questions. Notetaker records calls on major meeting platforms and prepares transcripts, summaries and action items. Callie entered beta in August 2026.

What Does WWT’s Softchoice Acquisition Add?

WWT completed the approximately US$1.3 billion acquisition in March 2025. Softchoice adds software and cloud expertise, consulting services and a stronger presence in Canada and commercial and small-business markets across North America.

Is WWT a $20 Billion Company?

WWT says it generates more than $20 billion in annual revenue. That figure measures sales, not market value. The company is privately held, so the revenue figure should not be described as its valuation.

What Does Vista’s $103 Billion Figure Measure?

Vista reported $103 billion in assets under management as of June 30, 2026. AUM is capital managed for investors. It is not Vista’s valuation, the combined value of portfolio companies or Robert F. Smith’s personal net worth.

What Is Vista’s Agentic AI Strategy?

Vista says it helps portfolio companies use generative and agentic AI to develop products and reduce operating costs. Its thesis is that software’s data and workflow context can make AI more useful to businesses. Adoption and financial results will vary by company.

The post Black Men in Tech: 3 Founders Who Built Billion-Dollar Companies and What They’re Doing Next appeared first on Black Elites.

For founders, a working space can help turn an idea into a demo. The AI boom still leaves a harder test: finding buyers and building businesses that can keep going after the demo is done.

The AI Boom Needs More Than Model Builders

The International Energy Agency estimates that global investment in data centres reached about $500 billion in 2024, nearly twice the 2022 level. That is spending on facilities and equipment, not revenue from AI products. Data centres used an estimated 415 terawatt-hours of electricity worldwide that year, about 1.5% of global consumption. The IEA projects that use could rise to around 945 terawatt-hours by 2030, with AI among the main drivers. The 2030 figure is a forecast.

Those facilities need power, cooling, servers, networks and security. Companies using them need help connecting the systems to their own software and data. Some of the resulting business goes to data-centre owners. Some goes to the firms that make the technology work.

For Black founders, cybersecurity, installation, data preparation, maintenance and system integration offer possible entry points. Each still requires skilled people and buyers willing to trust a supplier with critical work. The companies supplying the AI buildout show how much of the investment is flowing beyond the firms building AI models.

David Steward Built a Technology Supplier From Five People

World Wide Technology began in 1990 with five employees and 4,000 square feet. David Steward recalled those beginnings in a 2020 interview with TIME:

In 1990, I started this company called World Wide Technology with five people and 4,000 square feet. — David Steward

By then, WWT had grown into a $12 billion company. It now says it has more than 12,000 employees and is the largest Black-owned company in the United States. Steward founded the business and remains its chairman. WWT works across AI, cloud, networking and data-centre infrastructure. 

The company’s Advanced Technology Center gives customers a place to test systems before deployment. WWT says the centre has more than 500 racks and represents $1 billion in infrastructure investment. That is the company’s stated figure for the centre; it does not make WWT a hyperscale public cloud operator.

WWT built a business around getting complex technology to work together before it goes live. Testing can help customers avoid delays and costly failures. The centre supports that service; it is not an open campus competing for cloud tenants.

WWT also says it qualified for and graduated from the U.S. Small Business Administration’s Business Development Program, which provides training and technical assistance to eligible firms. That is one documented step in its history. It does not explain the company’s growth on its own. WWT’s scale reflects years of building the capacity and relationships to serve large customers.

Greenwood Is Building Local Access to AI

The federal funding behind Tulsa’s tech push went to a regional consortium, not to Black Tech Street alone. The U.S. Economic Development Administration announced approximately $51 million for the Tulsa Hub for Equitable & Trustworthy Autonomy, led by Tulsa Innovation Labs. The award covers six projects, including an AI Center of Excellence and programs linking manufacturers and startups to technology supply chains.

The Oklahoma Eagle reported that $10.6 million was allocated to the Greenwood AI Center. The Greenwood Cyber + AI Lab opened in May 2026. The paper also reported that creators will own the intellectual property they develop there. That gives founders a clearer stake in the work, though it cannot guarantee a customer or revenue.

Tyrance Billingsley II, founder of Black Tech Street, described the ambition this way:

Greenwood has an invaluable asset to ensure that its members are leading this revolution and shaping how our nation will lead. — Tyrance Billingsley II

Microsoft brings technical staff and equipment to the lab. Black Tech Street is building local participation. The wider Tulsa coalition connects the work to universities, manufacturers and public institutions. Whether that network helps startups secure paid work is still the key measure.

A demo does not pay a team while it waits for the next contract. Founders have to find buyers, negotiate terms and cover costs as they build. Felecia Hatcher’s work building an ecosystem for underrepresented founders points to the value of connecting entrepreneurs to investors and business support, alongside community institutions.

Data-Centre Growth Also Affects Local Communities

For most startups, owning a large data centre is not a realistic first move. Projects require land, power, equipment and years of financing and construction. A technical team cannot overcome those costs with a good product alone.

Black-owned firms can still pursue work around the facilities: cybersecurity, data preparation, installation, maintenance and integration. Those areas require less capital than building a campus, but buyers still expect a skilled team, reliable delivery and a track record.

People living near new data-centre projects have their own stake. The IEA says electricity demand is concentrated in the regions where facilities cluster. A project’s headline investment does not say who pays for grid upgrades, which local firms win contracts or how many jobs remain after construction. Those details matter to residents and policymakers weighing a proposal.

Partnership terms matter to founders, too. Does access include enough computing power to build and test? Who owns the code and data? Do cloud credits expire? What does a corporate partner expect in return? A short period of access may help with an experiment. A business needs customers and money to continue.

The Greenwood lab and WWT’s technology centre occupy different places in the same market. One gives local founders a place to work. The other supports a large Black-owned supplier serving major customers. Tulsa has opened the lab; the next evidence will come from what founders build there and whether buyers show up.

FAQs

What is AI infrastructure?

AI infrastructure includes data centres, chips, electricity, cloud platforms, networks, cybersecurity and services that connect and maintain these systems. Some parts require huge capital investment. Others rely on specialised skills and contracts with businesses that use AI.

Is the Greenwood Cyber + AI Lab a data centre?

No. It is a technology hub in Tulsa’s historic Greenwood district, opened by Black Tech Street and Microsoft in May 2026. Startups can access equipment, servers and technical support there, but it is not a hyperscale data-centre campus.

How much federal funding went to Tulsa’s tech hub?

The U.S. Economic Development Administration announced approximately $51 million for the broader Tulsa Hub for Equitable & Trustworthy Autonomy consortium. The Oklahoma Eagle reported that $10.6 million was allocated to the Greenwood AI Center.

How can Black founders enter the AI infrastructure market?

Founders can build services in cybersecurity, data preparation, installation, maintenance, AI testing and system integration. These areas usually require less capital than a data-centre campus, but buyers still expect specialised skills, reliable delivery and strong security.

Why does intellectual-property ownership matter?

Founders who retain the intellectual property they create can build products and companies around it, subject to their agreements. Access to a lab or equipment alone does not guarantee ownership, so founders should review partnership and funding terms before work begins.

What should communities ask about data-centre investments?

Ask who owns the facility, which local firms will win contracts, what jobs will last after construction and who pays for grid upgrades. The answers help residents and policymakers judge who benefits from a project and who carries its costs.

The post The Hidden AI Boom: How Black Tech Founders Are Securing the Infrastructure Economy appeared first on Black Elites.

Anthropic’s decision to fuse Chat, Cowork, and Design into one box kills that friction. You type what you need, and the backend handles the routing—whether that means answering in two sentences, making a slideshow, or running background tasks. It shifts the focus from software modes to getting work done, leaving rivals like OpenAI stuck defending clunky dropdown menus.

Why Anthropic Removed the Choice

The old arrangement required people to classify a job before Claude had properly encountered it. An experienced user might know when a request had become “Cowork work.” Many others stayed in Chat because that was where they had started. The tool with the deeper capability could remain untouched, even after a company had paid for access to it.

Anthropic says customers complained about deciding where a task belonged. Context did not always travel cleanly when work moved between environments either.

Those are adoption problems, not merely navigation complaints. Enterprise software regularly accumulates features faster than employees learn the labels attached to them. Anthropic’s answer is to put the classification work behind the input box. The model sees the request first and selects the machinery.

The initial rollout covers Pro and Max subscribers across web, mobile and desktop. Team and Free plans are expected later. Enterprise customers are on a slower timetable: administrators will receive at least 30 days’ notice before the unified experience reaches their organisations.

That delay is telling. A consumer can try a new interface and abandon a bad result. Inside a company, the same experiment may touch client material, shared drives or a scheduled process.

One Legal Request, Several Kinds of Work

Andrew Keller, identified by Anthropic as a senior economist, described a task that does not fit neatly inside a chat window:

I could have Claude pull up [my legal research database], and it would pull all the cases, read them, figure out which other cases I might need, download them, and store them in a folder for my personal review. -Andrew Keller

The request begins as retrieval. It soon becomes reading, gap identification and file preparation. Under the old product logic, the user might have had to notice those transitions. In the new interface, they are treated as parts of the same assignment.

Keller’s last four words matter: “for my personal review.” The work ends with a person inspecting what Claude assembled. Hiding the hand-off between search and execution may save time; hiding the point at which evidence needs checking would create a different problem.

Some employees will now use agentic features without deliberately opening an agent product—or necessarily knowing that Anthropic classifies the work that way. That may be exactly what drives usage. It also leaves the interface carrying more responsibility for explaining what Claude has done.

The Report and the Slides Share a History

An ordinary office assignment often loses detail as it moves between applications. The source sits in a drive. Colleagues discuss it in Slack. Someone checks the figures in a spreadsheet, writes the report in Word, and sends it to another person to turn into slides. A qualification that was obvious in the source may be missing from the version management sees.

Claude Docs and Slides are meant to keep more of that working history in one conversation. The same sources and instructions used for the report remain available when Claude produces the presentation. The user can edit the output, leave comments for Claude and share the work by link.

Anthropic is not forcing the finished file to remain there. Documents can be exported to Microsoft Word or Google Docs; presentations can be exported as PowerPoint or PDF files.

For enterprise buyers, the exports are useful. For Anthropic, they expose an unresolved question. Docs and Slides are beta products competing for attention against software with years of formatting features, approval routines and employee habit behind it. Claude may become the place where a document is researched and drafted, while Word, Google Docs or PowerPoint remains where it is actually finished.

If that is how people use the product, Anthropic has improved the preparation layer. It has not yet captured the whole workspace.

What Happens After the Prompt

The same box can now produce a short answer, consult a connector, create a document or keep working after the employee leaves. There is no mode selector requiring the user to announce which one is needed.

So the routing decision becomes less visible just as it becomes more consequential.

Consider a vague request written between meetings: “Check what changed and update the team.” Claude could interpret that as a summary. It might search several sources, alter a file or prepare a recurring task. The correct scope depends on company practice and on details the employee may assume are already understood.

Anthropic says Claude asks for approval before acting by default. Users can grant more freedom and have the system interrupt only when closer review is needed. Neither setting settles the central question for enterprise deployment: how well does Claude recognise when the request is underspecified?

The useful tests will be untidy ones. Administrators should watch what sources Claude consults after an ambiguous instruction, when it asks a follow-up question, and how plainly it records its activity. They should also time how long it takes an employee to stop or redirect work that began on the wrong premise. A launch demonstration, written to make the route obvious, cannot answer much of that.

KPMG’s Live Test

KPMG is already putting Anthropic’s proposition into a working system. In May 2026, the professional-services firm said Claude would be made available to more than 276,000 employees. Cowork and Managed Agents are also being embedded in Digital Gateway, KPMG’s platform for tax and legal work.

This is not a blank workspace. Digital Gateway contains proprietary tools and client information. A wrong assumption can survive into advice; material from one engagement cannot casually turn up in another.

Rema Serafi, vice chair of tax at KPMG US, described the earlier process this way:

Building an AI agent to help clients adjust to changing tax regulations used to take weeks and required teams to switch between multiple tools and chat windows. — Rema Serafi

Serafi said the integrated system cut that work to minutes. Anthropic’s published material does not provide an independent test of the claim, so the comparison remains an account from the two partners responsible for the deployment.

Even with that qualification, KPMG offers a useful test case because most of its staff will not be building agents from scratch. They will meet Claude inside assignments they already recognise. Administrators choose the available systems; employees ask for work to be done.

The less glamorous details will decide whether that scales. Which client context did Claude retain? Why did it select a particular source? What happens when project memory carries forward a mistaken assumption? Tax and legal teams cannot treat those questions as interface trivia.

The Pressure on Rival AI Workspaces

Reuters reported that Anthropic’s announcement followed OpenAI’s July introduction of ChatGPT Work, which brought conversational work closer to Codex capabilities. The two companies are confronting the same product problem: their AI systems can do more, but the growing collection of tools and workspaces can make that capability harder to approach.

OpenAI has models and specialised work surfaces serving different users. The choices are useful when someone already understands them. They are less helpful when an employee arrives with an assignment that will change shape after the first prompt. Anthropic’s redesign puts pressure on rivals to justify every mode decision they leave to the user.

Microsoft and Google start elsewhere. Their employees already know where a document, spreadsheet or presentation belongs, and AI has been inserted into those familiar applications. Anthropic does not have an established office suite that constrains the design. It can organise the experience around a continuing conversation.

That freedom does not supply the mature formatting, administrative controls or workplace habits already attached to Microsoft 365 and Google Workspace. The exports acknowledge as much.

Claude Code remains separate, while Design can still be opened on its own. For now, Anthropic has concentrated the merger on assignments that move among conversation, research and everyday office files. Whether users accept the model’s routing choices will matter more than whether the missing selector makes the screen look cleaner.

Questions Employees and Buyers May Have

Has Anthropic discontinued Cowork? Cowork’s capabilities are moving into ordinary Claude conversations rather than disappearing. Anthropic says existing chats, projects, skills and connectors will remain available.

Who receives the unified experience first? Pro and Max subscribers are first in the rollout. Team and Free access is expected later. Enterprise administrators will control internal availability after receiving at least 30 days’ notice.

Can Claude Docs replace Microsoft Word? That is not what the available product evidence establishes. Claude Docs may be useful when research, drafting instructions, and revision need to remain in the same conversation. Users can still export the file to Word or Google Docs, where established formatting and review processes may continue.

What if Claude selects the wrong workflow? The employee can redirect the assignment, but companies should test the recovery process themselves. The important questions are whether Claude shows what it is doing early enough, whether a user can stop it easily, and whether an unsuitable route reaches a connected system before anyone notices.

Is Claude Code included? No. Claude Code remains a separate developer product. The September announcement concerns Chat, Cowork, Design, Docs, and Slides.

The post How Claude’s Single-Interface Pivot Redefines the Enterprise AI Workspace appeared first on Black Elites.

Notehouse was a six-year-old CRM for nonprofit leaders, social workers and counsellors. It had about $100,000 in annual recurring revenue, 250-plus customer organisations and a 4.7-star Capterra rating. It also had years of technical debt. Waterglass founder Bernhard Hauser later said engineering diligence pointed toward one conclusion: much of the codebase should be replaced.

The buyer went ahead anyway, then rebuilt the application.

That sounds irrational only if the code was the asset. Hauser’s explanation was different: Waterglass had acquired an audience, recurring revenue and a place inside a specialised workflow. The purchase price was never disclosed. But that almost misses the point. Waterglass wasn’t really buying the code. It was buying customers, recurring revenue and a place inside a specialised workflow.

That distinction matters now because AI is changing the economics of software faster than it is changing the economics of distribution.

Code Is Getting Cheaper. Customers Are Not.

The barrier to producing software has fallen sharply. A 2025 field study involving 4,867 developers at Microsoft, Accenture and a Fortune 100 company found that developers given an AI coding assistant completed about 26% more tasks. Y Combinator managing partner Jared Friedman said that in the accelerator’s Winter 2025 batch, a quarter of startups had codebases that were 95% AI-generated.

Reliable software is not trivial—security, architecture, QA and maintenance still matter. But “we can build the product” is becoming less impressive as a standalone advantage.

Finding customers has not enjoyed the same deflation.

Benchmarkit’s 2025 SaaS benchmarks found that the median company spent $2 in sales and marketing to acquire $1 of new annual recurring revenue in 2024, up 14% from the previous year. The weakest quartile spent $2.82. It is not a universal formula, but it captures the problem: software supply is exploding while attention, trust and switching willingness remain scarce.

So the founder-economics question changes. Why spend a year building an AI product and another two trying to manufacture demand if a small, ignored SaaS company has already done the second job?

The Ugly Software May Contain the Real Moat

Vertical SaaS is often unimpressive from the outside. The interface can look dated. The feature set may be narrow. Growth can be slow. Yet a scheduling tool for clinics, CRM for photographers or case-management platform for social workers may sit inside a workflow customers repeat every day.

That produces assets a fresh AI app lacks: customer relationships, billing history, usage patterns, support conversations, domain vocabulary, integrations and knowledge about what users actually do.

What we bought was access to an audience. — Bernhard Hauser

The small-acquisition market shows buyers will pay for those economics. Acquire.com’s 2025 transaction data, covering more than 136 deals with reported sale prices and multiples, showed profitable SaaS businesses selling at a median 3.9 times annual profit. Businesses with less than $100,000 in annual net income averaged 3.7 times profit. In rough terms, a $100,000 purchase at that multiple would correspond to about $27,000 of annual profit before deal-specific adjustments. Acquire.com’s sample naturally skews toward smaller, founder-led transactions, but it offers a useful view of a market that is usually opaque.

There are literal small examples too. Jonathon Ringeisen sold Essential Studio Manager, a niche CRM for photographers, for $90,000 through Acquire.com. A functioning product and customer base already existed.

Software Equity Group counted 2,698 SaaS transactions in 2025, up 28% year on year; 72% referenced AI in some form. Vertical software represented 55% of SaaS M&A activity in the first quarter of 2026. Buyers are increasingly interested in software that already owns workflows and data.

Buy the Demand, Then Add the AI

The contrarian play is not “buy old SaaS because old SaaS is better.” It is to separate customer acquisition from product improvement.

A founder starting from zero must discover a problem, ship a product, earn trust, find a channel, convert users, retain them and only then learn which AI features matter. An acquirer can begin halfway down that list.

AI then becomes an upgrade layer.

Support tickets can be classified automatically. Repetitive data entry can be reduced. Customer records can support better search, summaries or recommendations. Engineering and support costs can fall. Most importantly, those upgrades can be tested against paying users instead of an imagined market.

Waterglass followed almost exactly that sequence with Notehouse. It kept the customer base, rebuilt the application in roughly six months and used AI-assisted software engineering to accelerate the work. The crucial point is not that AI rescued bad code. It is that existing demand gave the rebuild somewhere to land.

That is the difference between buying existing demand and creating new demand. One uses capital to purchase a proven revenue stream and then improve it. The other uses capital to fund uncertainty.

I am much better at taking something that exists and bringing it to the next level. — Pascal Levy-Garboua

The Trap Is Buying a Problem and Calling It an AI Opportunity

This strategy can go wrong quickly.

Notehouse itself is a warning. During a payment migration after the acquisition, some customers had to reconfirm billing details and cancellations increased. Technical debt was so severe that the buyer chose a full rebuild. Because the product handled sensitive information, security and HIPAA compliance also mattered. A cheap acquisition can become expensive once migration, remediation and support work are counted.

Churn can turn $100,000 of ARR into $70,000 after twelve months. Customer concentration can turn “recurring revenue” into one renewal decision. Weak code may conceal unsupported libraries, security holes or a single developer who understands the system. Data may be legally restricted or useless for AI. Buyers can also overpay for growth customers never asked for.

The acquisition thesis has to work before the AI story does.

A Practical Acquisition Test

Before buying a small SaaS company, an operator should be able to answer seven questions:

1. Is the revenue genuinely durable? Examine cohort retention, cancellations, failed payments, contract terms and gross revenue retention—not just MRR on the day of sale.
2. How concentrated is the customer base? Calculate what happens if the largest one, three and five customers leave.
3. How deeply does the product own a workflow? Daily or weekly operational use is more valuable than occasional convenience.
4. What proprietary data or feedback loops actually exist? “We have data” is meaningless unless the buyer has lawful access and the data can improve the product.
5. What will the code cost after closing? Commission independent technical, security and dependency diligence. Price the rebuild before signing.
6. Can AI produce a measurable customer or margin improvement? Faster support, fewer manual steps, higher ARPU or lower operating cost is a thesis. “Add a chatbot” is not.
7. Does the deal work without heroic growth? Compare purchase price with current owner earnings, required reinvestment and realistic churn. AI upside should be optionality, not the only way to justify the valuation.

That last point is where acquisition fantasies die. If the deal only works after doubling prices, halving churn and launching a perfect AI agent, the buyer is not acquiring certainty. The buyer is paying upfront for a turnaround.

FAQ

Is buying a small SaaS business safer than starting an AI company?

Not necessarily. It replaces product-market-fit risk with acquisition, technical and retention risk. The advantage is that revenue and customer behaviour can be diligenced before the purchase.

What makes a small vertical SaaS business attractive?

Strong retention, fragmented customers, recurring usage, healthy margins, low platform dependency and a workflow customers are reluctant to disrupt. A dated interface can be fixed. A customer base that never cared about the product cannot.

How much should a buyer pay?

There is no universal number. Acquire.com’s 2025 data put the median profitable SaaS transaction at 3.9 times annual profit and businesses below $100,000 in net income at 3.7 times on average. Those figures are a starting benchmark, not a valuation rule.

The most interesting AI company may therefore begin with software nobody wants to brag about owning.

The interface can be ugly. The code can need work. The category can sound boring at dinner.

But if customers keep paying, the workflow is embedded and the data is useful, the buyer starts with the one thing thousands of new AI apps are still desperately trying to manufacture: distribution.

In the AI era, the moat may not be who can build the smartest software. It may be who already has someone paying for it.

The post Why Smart Founders are Buying Ugly $100k SaaS Tools Instead of Building AI Apps appeared first on Black Elites.

Most of the industry scrambled for patches. Kunle Olukotun had already bet on something bigger: put multiple processors on one chip. That idea reshaped how modern processors get built. Decades later, it quietly became one of the engines behind the AI boom.

Betting Against the Bigger-Core Race

Back in 1996, everyone was building wider, more complex processors—more instruction issue, more speculative execution, more silicon dedicated to squeezing parallelism out of a single stream of instructions.

Olukotun, along with four Stanford colleagues, thought this was a dead end. Their pitch was simple: take one complicated processor, replace it with four simpler ones, keep the silicon area roughly the same. They ran the numbers assuming equal 500 MHz clocks and similar 430 mm² dies. The catch? For anything that couldn’t be parallelized, the old superscalar design still won—by about 30%. But for parallel and multiprogramming workloads, the four-core setup crushed it, beating the single-core design by 50% to 100%.

Olukotun wasn’t content to leave the idea in academia. He turned it into Stanford’s Hydra chip-multiprocessor project before founding Afara Websystems to bring the concept to market. Sun Microsystems bought Afara in 2002. That technology became the backbone of Niagara. By December 2005, Sun was shipping the UltraSPARC T1—eight cores, 32 simultaneous threads. According to the IEEE Computer Society, Niagara-derived chips eventually pulled in billions in revenue.

Here’s the thing: the bet wasn’t “multicore will be faster.” It was that future software would have enough parallelism baked in to make many simple cores beat one very complicated core. That’s a much riskier bet. It paid off anyway.

Design Around the Constraint, Not the Fashion

Find what actually limits performance: Don’t chase the metric everyone’s talking about. Find the resource that gets scarce as things scale.

“The inference problem is not really a compute problem.” That’s Olukotun, speaking on AI:AM in July 2026. His argument now: AI inference is bottlenecked by moving model parameters and KV-cache data around, not by raw FLOPS. SambaNova’s dataflow architecture is built around exactly that—maximizing memory-bandwidth use and overlapping communication with computation.

Don’t specialize past the market: Specialization should cut waste, not lock you into an assumption that might not hold in five years. “I’ve learned never to bet against the innovation capabilities of software people and algorithm people,” he’s said. SambaNova went with reconfigurable dataflow instead of hard-wiring a single algorithm into silicon. The chip can specialize, sure—but it can also adapt as models evolve.

Push complexity away from the user: Even great technology stalls if people have to babysit its internals. “We need to raise the level of abstraction for parallel programming,” Olukotun told The Register back in 2008. His Stanford Pervasive Parallelism Lab chased domain-specific languages and compilers that could find parallelism on their own—no manual thread-juggling required. Sun, AMD, Nvidia, IBM, HP, and Intel backed the lab with $6 million over three years.

When Afara Hit a Wall

Co-founder Les Kohn later remembered it plainly: Afara had closed one funding round, and then September 11 happened. Fundraising froze overnight. With fresh capital suddenly unavailable, Afara had little room to stay independent. Sun acquired the company, preserving the technology even as the startup disappeared.

There’s no solid evidence Olukotun ever pinned a specific lesson on that moment. But something did shift—the capital surrounding his work got a lot sturdier. PPL had six major tech companies behind it. SambaNova later drew SoftBank Vision Fund 2, BlackRock, Intel Capital, GV, Temasek, GIC, and others. In 2026 alone, SambaNova pulled in over $350 million in Series E funding, plus a planned strategic partnership with Intel.

The Network Behind the Architecture

DARPA funded the original 1996 research. Stanford gave him the platform to build on. Chipmakers funded PPL. Institutional investors funded SambaNova. One protégé, Hassan Chafi, did his PhD under Olukotun—working on transactional memory and the Delite DSL—before moving on to lead research at Oracle Labs.

Put it together, and you get something rare: academic research, chip talent, enterprise buyers, semiconductor relationships, and patient capital, all connected. That’s what it actually takes to commercialize an idea that might not look obvious for years.

Find the Constraint That Gets Worse With Scale

Before you optimize your product, do this instead: find the one resource whose cost—or inefficiency—grows fastest when usage jumps 10x. Then run one experiment aimed squarely at cutting that in half.

That’s the whole Olukotun pattern, really. Don’t polish an architecture whose bottleneck is already structural. Replace it.

The post Kunle Olukotun Solved Computing’s Scaling Problem, Years Before Anyone Cared About AI appeared first on Black Elites.