How AMD Can Compete
So how can AMD hope to compete? First, it can wisely pick its targets. AMD did a great job of this with its Fusion C and E series processors. They're designed to compete against Intel's Atom processors in the market where Atom is delivering unsatisfactory experiences: low-cost netbooks. AMD saw the ways Atom disappoints - sluggish in-order execution processing cores and lackluster graphics and video support - and made a tiny, energy-efficient processor that is cheap and low-power enough to go into laptops that addresses those key issues. As Intel focuses on making Atom smaller and more energy efficient to end up in tablets and smartphones, AMD could continue to do well by making designs laser-targeted at the small and light $400 laptop. For the time being, AMD would be wise not to take on Intel were it is strongest.
Second, AMD should continue to focus on graphics. Video and graphics performance increasingly defines the way people use their PCs, and the focus on using graphics hardware to accelerate the web will only increase its importance. Intel has recognized this as put forth serious effort into dramatically improving its integrated graphics in the "Sandy Bridge" 2nd generation Core processors, but they're still quite a bit behind what AMD can do. Fusion A series processors aimed at mid-priced laptops (code-named Llano), will push integrated graphics far beyond the already impressive low-power Fusion C and E processor. That's should be just the start. AMD should do everything in its power to make sure that, if you don't buy a system with a discrete graphics card, the graphics and video are going to be night-and-day better on an AMD system than an Intel system. And, if you are going with discrete graphics, make sure a Radeon brand card offers the best bang for the buck and lowest power utilization at all price tiers. That's a tall order, but it's totally in AMD's wheelhouse.
AMD should just concede the tablet and phone markets for the time being. It will take years of fighting for marketshare and enormous engineering resources to chase down the competition in that arena. To make a real competitive solution, AMD has to beat the efforts of ARM, Intel, Nvidia, PowerVR, Samsung, TI, Qualcomm... It will consume R&D resources that could be spent making awesome processors for more traditional computers. Sure, smartphones and tablets are growth industries and the next Windows will run on ARM processors in addition to x86, but traditional computers aren't going away anytime soon. They're still going to sell in the hundreds of millions a year. AMD isn't big enough to fight both fights right now - it should focus on the PC, Mac, and server markets for at least the next few years.
Will the Fusion A series that will hit the market in a couple months be the chip that puts AMD back on top, at least in terms of delivering the processor you want to have in your new laptop or all-in-one desktop? I'm not convinced. The architectural tweaks to the Phenom II-based CPU cores, 32nm manufacturing process, new power-saving features, and powerful integrated graphics make for a potent combo. Still, Intel is just too far ahead. Llano is likely to be the king of integrated graphics performance by a wide margin, but tasks that stress the CPU cores will still be considerably faster on 2nd generation Core CPUs. What AMD needs is a new CPU core design. That's coming later this summer (probably) with the product code-named Bulldozer. It's such an unusual take on multi-core design that I really have a hard time predicting if it can beat the Intel chips of comparable price and power usage. Even if it does, it'll be another year before Bulldozer cores end up in a Fusion-style combination CPU and GPU, and Intel will be well into its 22nm chips by then.
For AMD to stay relevant, let alone deliver the "gotta have it" processor for future laptops and desktops, it really needs to remain focused, shorten the time between its major architectural design revisions, and work with its manufacturing partners at GlobalFoundries and TSMC to somehow shorten the manufacturing process lead Intel enjoys.