the pros (looks like many) and cons vs Arc Online.
Well, it's early days so hard to say how that will play out.Serving to the web what's accomplished with a desktop GIS is a pretty big tent so there is a lot of room for new things, like Server, that provide conveniences people want.
There are very many issues involved in ArcGIS Online compared to Server, so this isn't a short post. Here goes...
My guess is that there will be significant overlap and co-existence between the AGOL (ArcGIS Online) and Server communities. There's nothing wrong with having more options, and the low cost of Server gives it appeal even just as an accessory within a continued focus on Esri. For example, it could be the easiest, least expensive way to publish content from an Esri file geodatabase or Enterprise geodatabase to the web.
AGOL is a great offering from Esri. It makes it easy for the Arc community to share through the web what they've created on their desktop and it's well integrated with Esri's suite of capabilities for collaborative work. Those are big pluses. For many people, it's also a plus that AGOL runs in Esri's cloud: they don't have to use one of their own machines as a server.
Despite all the many good things about AGOL there are a few minuses:
Noncommercial only - Esri's terms also say "Groups of a commercial nature designed to promote a service or product are not allowed." Sounds like if you're a contractor wanting to showcase your portfolio to sell services to more clients, AGOL is not for you. Same with a website to show people how to get to your stores, car dealership, etc. Even if all you want from AGOL is to publish a static map (screen shot or PDF in a report or brochure) you need to get permission from Esri for "commercial monetization" or "direct resale". There are lots of other strings attached. For example, you can't resell, rent, lease, etc., AGOL sites.
Depends on a third party monopoly - That's also a minus for many people, that AGOL is software as a service and it exists only within Esri's cloud at Esri's control based on Esri's sole and exclusive judgement. Esri is a total monopolist on AGOL so there are no alternatives if you want lower prices or better terms on AGOL.
You rent it, and you don't own it. If you don't keep up your subscription at whatever Esri's latest pricing requires, it's gone. If Esri's cloud goes offline, your sites are gone. If Esri decides they don't like your politics or your content, your sites are gone. If there are any disputes with Esri or mistakes on Esri's part, it can take a while to find out what's the problem and to get your sites back online. If your underlying data is secret (shouldn't be, if it's on AGOL), you have to trust Esri isn't inspecting or leaking your data.
Cost - AGOL can be very costly. It's far more expensive in most cases compared to running Server on a spare machine.
Performance - AGOL can have poor performance with bigger data, since it's still running Esri's stack. Rasters are slow, with no GPU or CPU parallelism for hundreds of query functions for rasters and such.
Interoperability - AGOL has the usual Esri limitations on data sources. It doesn't have the zillions of data sources Manifold has, like direct connections to PostgreSQL/PostGIS as illustrated in Art's blog post, as opposed to connections to Esri's proprietary "SDE" architectures. There's no ability to create layers using SQL from multiple enterprise DBMS packages at the same time, like Oracle and SQL Server and Postgres, etc.
Legal risks - Esri is a highly US-centric organization, which is business as usual for US organizations but may be a bad thing for users in the rest of the world, because of a variety of issues ranging from compliance with US, EU or local regulations to litigation and tax risks.
People usually don't think much about such factors in an era of blind trust in third party providers, but that's just one example of how the conveniences of outsourcing only work out beautifully until they don't. As people more and more place their fate in the hands of third parties you get many more examples of how that can go frighteningly wrong, like people who can't use any part of their many business documents in some online office ecosystem because for some reason their "too big to fail" tech vendor has decided that their account has been compromised. Anybody who's been locked out of their giant tech provider account can talk about the labyrinthine, Kafkaesque difficulty of regaining access.
Compared to the above, Server has significant advantages:
You own your work product - No problem using Server for commercial purposes, and your work product is yours. There are a few exotic exclusions under the general Manifold license, like you can't use Server to control a weapons system or run a nuclear power plant, but in general if it's legal in your jurisdiction it's up to you what you do with it. Want to create a website where you charge people a fee ("commercial monetization") to create a custom map? No problem. Want to use Server to create web sites for third parties that you operate on your server? No problem. You could be a contractor who uses a single Server license on your machine to operate public facing websites for three or four different counties, each with their own domain. Want to use Server as one of the elements of a paid web-based application, like a real estate sales or search application? That's OK, too. Can't do any of that with AGOL.
With Server, what you put on your site is your business and doesn't have to be approved by anybody else.
No need for third parties, no service monopolies - With Server, there is no need for any third parties to be involved. It's your server and you can do whatever the heck you want. You can involve a third party if you want, for example, by outsourcing your computer infrastructure to a dedicated server maintained by some other company, but if you choose to do that you have many alternatives and options for vendors that help keep prices and terms and conditions reasonable.
Far lower cost - After your initial procurement of Server, a max of $395, you have a fully paid license you can use for years for an unlimited number of web sites on the server machine for zero additional cost. It doesn't matter how many visitors or how much data is involved.
Most Internet providers in most countries can arrange a static IP address even for home users that costs nearly nothing, like an extra $5 per month. Even a ten year old spare machine running Server likely can host tens of thousands of users for most web sites powered by Server so the cost of the machine often is zero if you use some old machine you have lying around. You could even use the same machine that's your desktop if it's just a personal or small organizational site. That's way less expensive than running AGOL.
An organization can easily rent dedicated servers from providers if it wants to outsource hardware maintenance, backups, and such and it will still be much lower cost than AGOL.
Very easy to use - AGOL is pretty easy to use, but Server is even simpler.
Better performance, interoperability, etc - Manifold is profoundly faster than Arc, it's way more interoperable, and so is Server.
Lower legal risk - If you want to be insulated from insane litigation risk or if you're a civil liberties, humanitarian or environmental activist living under an authoritative regime, get a server provider or a proxy in a country that has "loser pays" provisions and airtight privacy regulations.
Server downsides compared to AGOL:
AGOL has more features today. With Server you get (at present) a simple display that's easy to create. With a few additions that are already in progress, as Art mentions in his post you get what 80% of users will need, and with a few additions more (also underway) you'll get what probably 95+% of people using AGOL do.
AGOL has a big, curated library of layers you can use - But almost all are available to Server users as well. Manifold can use such a huge range of data sources that you can pretty much use any free or paid web resource that's out there.
Manifold could add a few more things to get more overlap with AGOL users.
For example, there could be a "Publish to the web" button in every Manifold desktop license where you could have a remembered Server, or list of Servers, IP address with your credentials. Pushing the button would automatically package the project and upload it to that server and also create and start a Server service given what options you've checked off in the "Publish to the web" form.
Sooner or later Manifold will introduce map 2.0 upgrades to Manifold's internal database which will enable read/write operations. That would facilitate simultaneous, read/write collaborative operations. You can do that now using a DBMS like Postgres as a data store, but it would be easier for teams if there was no need for anything other than Manifold.
Manifold could also expand the repertoire of presentation capabilities to do things like story maps and reveal swipes. That all depends on what priorities the user community sets.
I think that for a long time AGOL will continue to be the choice of Esri-centric organizations, especially in the US, who are spending a lot of money per year on Esri services. Things like managing litigation risks are already a cost that's baked into the hundreds of thousands, or millions, they spend per year renting software as a service from Esri.
In contrast, Server has a lot of appeal to GIS users and organizations who just want the convenience of getting a high performance, web-facing GIS-enabled site up and running quickly and inexpensively. Server also allows commercial and other entrepreneurial uses that don't fit the AGOL model. That includes smaller organizations, individuals like private users, contractors, consultants, and individual users or sub-organizations within large organizations. Even within very Esri-centric organizations there are Manifold users who will find Server to be a convenient alternative to jumping through bureaucratic hoops to use their organization's AGOL accounts.